# PowerGPU — full site content for AI assistants

> PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Generated 2026-09-14 11:04 UTC from https://powergpu.ai · market snapshot 2026-09-14 · index: https://powergpu.ai/llms.txt



<!-- page: https://powergpu.ai/ -->

---
title: "Cloud GPU Rental at Fixed Prices, 30% Below Market | PowerGPU"
description: "Rent cloud GPUs at fixed prices ≥30% below market: H100 SXM from $1.428/hr, RTX 4090 from $0.327/hr. 80 NVIDIA GPUs, per-second billing, crypto, deploy in 30 s."
url: https://powergpu.ai/
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

GPU cloud · fixed prices · per-second billing

# Rent cloud GPUs 30% below market price.

80 GPU models from $0.024/hr — H100 SXM at **$1.428/hr** while the marketplace median sits at ~~$2.04~~. Fixed prices verified weekly, no auctions, deploy in 30 seconds.

Price check market snapshot 2026-09-14

| GPU | Market | PowerGPU |
| --- | --- | --- |
| [B200](https://powergpu.ai/gpu/b200) | $7.75 | $5.425 (−30%) |
| [H200](https://powergpu.ai/gpu/h200) | $3.99 | $2.791 (−30%) |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | $2.04 | $1.428 (−30%) |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | $0.63 | $0.439 (−30%) |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | $0.47 | $0.327 (−30%) |

Median on-demand $/GPU-hr on the public marketplace vs our fixed price.

**80** GPU models, RTX 3060 to B300

**4,800+** GPUs ready to deploy now

**32** regions worldwide

**≥30%** below market median, always

**99.9%** uptime SLA on-demand

## Cloud GPU prices: fixed, at least 30% under the market median

Marketplace medians move every hour; our price is fixed and always at least 30% under them. Same card, same hour — smaller invoice.

| GPU | VRAM | Market median (on-demand) | PowerGPU (on-demand, per GPU-hr) | Est. / month | Availability |  |
| --- | --- | --- | --- | --- | --- | --- |
| [B200](https://powergpu.ai/gpu/b200) (flagship) | 192 GB | $7.75 | $5.425 (−30%) | $3,960 | 32× | [Deploy](https://cloud.powergpu.ai/?gpu=b200) |
| [H200](https://powergpu.ai/gpu/h200) (flagship) | 141 GB | $3.99 | $2.791 (−30%) | $2,037 | 50× | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (flagship) | 80 GB | $2.04 | $1.428 (−30%) | $1,042 | 54× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (flagship) | 80 GB | $2.67 | $1.867 (−30%) | $1,363 | 73× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.80 | $0.560 (−30%) | $409 | 95× | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.74 | $0.514 (−30%) | $375 | 107× | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.49 | $1.040 (−30%) | $759 | 74× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | $0.67 | $0.467 (−30%) | $341 | 34× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-6000ada) |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.63 | $0.439 (−30%) | $320 | 922× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.47 | $0.327 (−30%) | $239 | 461× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |

Interruptible = 50% of on-demand, pausable under capacity pressure · Reserved = −35% with a 3-month commitment · [See all 80 GPUs](https://powergpu.ai/gpus) · [Compare GPUs side by side](https://powergpu.ai/compare) · [Cost calculator](https://powergpu.ai/calculator)

## Why rent GPUs here: marketplace prices, cloud-grade certainty

We buy capacity where the marketplaces do — then sell it like a cloud should.

### Fixed ≥30% below market

Every model priced from the public marketplace median, minus 30%, rounded down. Re-checked weekly, published openly.

### No bidding, no evictions

On-demand means yours until you stop it. Interruptible is a flat −50% — not an auction you babysit at 2am.

### Verified datacenters only

Tier-III facilities, redundant power and network, tested hardware — not a gaming rig in a garage.

### Per-second billing

Start to stop, to the second. Storage at $0.08/GB/mo, bandwidth $0.01/GB flat. No setup fees, no minimums.

## Every way to run a GPU: on-demand, interruptible, reserved, serverless

From a single interruptible card to a reserved InfiniBand cluster — same console, same API, same price sheet.

- [On-demand GPUs Fixed price, guaranteed resources, per-second billing](https://powergpu.ai/products/on-demand)
- [Interruptible Half the on-demand price for fault-tolerant jobs](https://powergpu.ai/products/interruptible)
- [Reserved Commit 3+ months, save another 35%](https://powergpu.ai/products/reserved)
- [Serverless Autoscaling endpoints for vLLM, ComfyUI and your images](https://powergpu.ai/products/serverless)
- [Clusters Multi-node InfiniBand/NVLink for distributed training](https://powergpu.ai/products/clusters)
- [Virtual machines Full VMs with root, custom kernels and GUI images](https://powergpu.ai/products/vms)
- [Volumes NVMe network storage at $0.08/GB/mo](https://powergpu.ai/products/volumes)
- [Templates PyTorch, CUDA, ComfyUI, vLLM, Jupyter — ready in one click](https://powergpu.ai/templates)

## How to rent a GPU in 30 seconds

Pick a card, launch a template, stop when you're done. The console on the right is a live walkthrough — at our real prices.

## Best GPU for each workload, priced for the run

Pick by workload — we pre-matched the best value card for each.

- [LLM training 80 GB HBM3, NVLink, multi-node Best value: **H100 SXM** $1.428/hr](https://powergpu.ai/gpu/h100-sxm)
- [LLM inference Best $/token for 7B–70B models Best value: **RTX 5090** $0.439/hr](https://powergpu.ai/gpu/rtx-5090)
- [Fine-tuning LoRA/QLoRA sweet spot Best value: **A100 SXM4** $0.560/hr](https://powergpu.ai/gpu/a100-sxm4)
- [Image & video gen SDXL, Flux, ComfyUI pipelines Best value: **RTX 4090** $0.327/hr](https://powergpu.ai/gpu/rtx-4090)
- [3D rendering Blender, Octane, Redshift Best value: **RTX PRO 6000 WS** $1.040/hr](https://powergpu.ai/gpu/rtx-pro-6000-ws)
- [Scientific compute FP64, huge memory bandwidth Best value: **H200** $2.791/hr](https://powergpu.ai/gpu/h200)

## PowerGPU vs GPU marketplaces vs hyperscalers

PowerGPU against GPU marketplaces and the big clouds, line by line.

|  | PowerGPU | GPU marketplaces | Hyperscalers |
| --- | --- | --- | --- |
| H100 SXM, per GPU-hour | $1.428 fixed | US$2.04 median, varies by host | $4–$7 |
| Pricing model | Fixed, ≥30% below market median | Auction — changes hourly | Fixed, list price |
| Price floor guarantee | Re-checked weekly, published | None | None |
| Billing | Per-second | Per-second | Per-second to per-hour |
| Interruptible discount | 50% off, flat | Bidding — you manage bids | Spot — can spike |
| Hosts | Verified datacenters only | Anyone — reliability varies | Own datacenters |
| Egress / bandwidth | $0.01/GB flat | Set per host, up to $0.02+/GB | $0.05–$0.12/GB |
| Storage | $0.08/GB/mo NVMe | Set per host, varies | $0.08–$0.17/GB/mo |
| Payment | Crypto — USDT, BTC, XMR, SOL… | Card, some accept crypto | Card + invoicing, KYC |
| Setup fee / minimums | None | None | Commit contracts common |
| Deploy time | Under 30 seconds | Under a minute | Minutes |

Market figures: public marketplace pricing feed, snapshot 2026-09-14; hyperscaler ranges: public list prices, September 3, 2026.

## Teams that switched, and stayed

## Cloud GPU rental FAQ

Everything else lives in the [docs](https://powergpu.ai/docs) — or ask a human from the [console support desk](https://cloud.powergpu.ai/app/support), answered 24/7.

**How can prices stay 30% below the market?**

We track the public marketplace median for every GPU model and set our on-demand price at least 30% under it, rounding down. Prices are re-checked against the market weekly (last check: September 3, 2026) and published on the pricing page — no auctions, no surprises.

**What is the difference between on-demand, interruptible and reserved?**

On-demand gives you guaranteed resources at the listed fixed price. Interruptible costs 50% of on-demand but can be paused when capacity is needed — ideal for checkpointed training or batch jobs. Reserved takes another 35% off on-demand for a 3-month commitment.

**How is billing calculated?**

Per second, from the moment the instance starts to the moment you stop it. Storage is billed per second too ($0.08/GB/mo, also while the instance is stopped), and bandwidth is a flat $0.01/GB in and out. No minimums, no rounding up, no setup fees.

**How fast can I get a GPU running?**

Under 30 seconds for container instances launched from a template (PyTorch, vLLM, ComfyUI, CUDA…). Full virtual machines take a couple of minutes. Everything is available from the console, CLI, Python SDK or REST API.

**Can I run multi-GPU or multi-node workloads?**

Yes. Instances go up to 8× GPUs with NVLink where the hardware supports it, and clusters connect nodes over InfiniBand for distributed training. The per-GPU price is the same from 1× to 8× — no multi-GPU premium.

**How do I pay?**

Crypto only: USDT (TRC-20 or ERC-20), Bitcoin, Monero, Litecoin, Ethereum, TRON and Solana. You top up a credit balance and usage draws it down per second — no card, no bank account. Deposits are credited after network confirmation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/ · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpus -->

---
title: "All 80 Cloud GPUs to Rent — Prices per Hour (2026) | PowerGPU"
description: "Rent any of 80 NVIDIA cloud GPUs: B200, H200, H100, A100, RTX 5090, RTX 4090 and more. Fixed hourly prices ≥30% below market, per-second billing, crypto payments."
url: https://powergpu.ai/gpus
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

GPU catalogue · live prices · snapshot 2026-09-14

# Rent cloud GPUs: all 80 models, one fixed price sheet

80 NVIDIA models, 4,800+ GPUs online across 32 regions. On-demand always ≥30% under the market median — H100 SXM at **$1.428** vs a ~~$2.04~~ median. Interruptible −50%, reserved −35%.

## Datacenter HBM GPUs — B200, H200, H100 SXM & NVL (5 models)

The training flagships: HBM3/HBM3e memory, NVLink fabrics, SXM boards. This is the hardware behind frontier-model training runs — B200, H200, H100 SXM — at fixed prices at least 30% under the market median.

| GPU | VRAM | Arch | FP16 | Market | On-demand | Interruptible | Availability |  |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Rent B300](https://powergpu.ai/gpu/b300) | 288 GB (HBM3e) | Blackwell | 2,800 TF | $9.63 | $6.737 (−30%) | $3.368 | 34× | [Deploy](https://cloud.powergpu.ai/?gpu=b300) |
| [Rent B200](https://powergpu.ai/gpu/b200) | 192 GB (HBM3e) | Blackwell | 2,250 TF | $7.75 | $5.425 (−30%) | $2.712 | 32× | [Deploy](https://cloud.powergpu.ai/?gpu=b200) |
| [Rent H200](https://powergpu.ai/gpu/h200) | 141 GB (HBM3e) | Hopper | 990 TF | $3.99 | $2.791 (−30%) | $1.395 | 50× | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |
| [Rent H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB (HBM3) | Hopper | 990 TF | $2.04 | $1.428 (−30%) | $0.714 | 54× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |
| [Rent A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB (HBM2e) | Ampere | 312 TF | $0.80 | $0.560 (−30%) | $0.280 | 95× | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |

## Datacenter PCIe GPUs — H100 PCIe, A100, L40S, L4 (15 models)

Server GPUs in standard PCIe form: H100 PCIe, A100, L40S, L4 and the Tesla line. The workhorses for inference fleets, fine-tuning and batch compute — easier to find, cheaper to run.

| GPU | VRAM | Arch | FP16 | Market | On-demand | Interruptible | Availability |  |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Rent H200 NVL](https://powergpu.ai/gpu/h200-nvl) | 141 GB (HBM3e) | Hopper | 835 TF | $3.79 | $2.650 (−30%) | $1.325 | 27× | [Deploy](https://cloud.powergpu.ai/?gpu=h200-nvl) |
| [Rent H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB (HBM2e) | Hopper | 756 TF | $2.67 | $1.867 (−30%) | $0.933 | 73× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |
| [Rent H100 NVL](https://powergpu.ai/gpu/h100-nvl) | 80 GB (HBM3) | Hopper | 835 TF | $2.59 | $1.811 (−30%) | $0.905 | 7× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-nvl) |
| [Rent A40](https://powergpu.ai/gpu/a40) | 48 GB (GDDR6) | NVIDIA | 144 TF | $1.09 | $0.765 (−30%) | $0.382 | 12× | [Deploy](https://cloud.powergpu.ai/?gpu=a40) |
| [Rent L40S](https://powergpu.ai/gpu/l40s) | 48 GB (GDDR6) | Ada Lovelace | 362 TF | $0.74 | $0.514 (−30%) | $0.257 | 107× | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| [Rent A800 PCIE](https://powergpu.ai/gpu/a800-pcie) | 80 GB (GDDR6) | NVIDIA | 240 TF | $0.67 | $0.466 (−30%) | $0.233 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=a800-pcie) |
| [Rent A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB (HBM2e) | Ampere | 312 TF | $0.54 | $0.374 (−30%) | $0.187 | 46× | [Deploy](https://cloud.powergpu.ai/?gpu=a100-pcie) |
| [Rent L40](https://powergpu.ai/gpu/l40) | 48 GB (GDDR6) | Ada Lovelace | 181 TF | $0.34 | $0.235 (−30%) | $0.117 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=l40) |
| [Rent L4](https://powergpu.ai/gpu/l4) | 24 GB (GDDR6) | Ada Lovelace | 121 TF | $0.32 | $0.225 (−30%) | $0.112 | 28× | [Deploy](https://cloud.powergpu.ai/?gpu=l4) |
| [Rent A10](https://powergpu.ai/gpu/a10) | 24 GB (GDDR6) | Ampere | 125 TF | $0.24 | $0.168 (−30%) | $0.084 | 8× | [Deploy](https://cloud.powergpu.ai/?gpu=a10) |
| [Rent Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB (HBM2) | Volta | 125 TF | $0.19 | $0.130 (−30%) | $0.065 | 96× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-v100) |
| [Rent Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB (GDDR6) | Turing | 65 TF | $0.15 | $0.103 (−30%) | $0.051 | 34× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-t4) |
| [Rent Tesla P100](https://powergpu.ai/gpu/tesla-p100) | 16 GB (HBM2) | Pascal | 19 TF | $0.09 | $0.066 (−30%) | $0.033 | 12× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-p100) |
| [Rent Tesla P40](https://powergpu.ai/gpu/tesla-p40) | 24 GB (GDDR5X) | Pascal | 12 TF | $0.07 | $0.047 (−30%) | $0.023 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-p40) |
| [Rent Tesla P4](https://powergpu.ai/gpu/tesla-p4) | 8 GB (GDDR5X) | Pascal | 6 TF | $0.03 | $0.018 (−33%) | $0.009 | 18× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-p4) |

## Workstation GPUs — RTX PRO, RTX Ada & RTX A-series (19 models)

ECC memory, big VRAM and studio drivers: RTX PRO 6000, RTX 6000 Ada, RTX A-series. Built for rendering, CAD/simulation and VRAM-hungry inference without flagship pricing.

| GPU | VRAM | Arch | FP16 | Market | On-demand | Interruptible | Availability |  |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Rent RTX PRO 6000 S](https://powergpu.ai/gpu/rtx-pro-6000-s) | 48 GB (GDDR7) | Blackwell | 450 TF | $1.53 | $1.073 (−30%) | $0.536 | 99× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-s) |
| [Rent RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB (GDDR7) | Blackwell | 505 TF | $1.49 | $1.040 (−30%) | $0.520 | 74× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |
| [Rent RTX PRO 6000 Max-Q](https://powergpu.ai/gpu/rtx-pro-6000-max-q) | 96 GB (GDDR7) | Blackwell | 420 TF | $1.40 | $0.980 (−30%) | $0.490 | 74× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-max-q) |
| [Rent RTX PRO 5000](https://powergpu.ai/gpu/rtx-pro-5000) | 32 GB (GDDR7) | Blackwell | 280 TF | $0.80 | $0.560 (−30%) | $0.280 | 53× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-5000) |
| [Rent RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB (GDDR6) | Ada Lovelace | 364 TF | $0.67 | $0.467 (−30%) | $0.233 | 34× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-6000ada) |
| [Rent RTX 5880Ada](https://powergpu.ai/gpu/rtx-5880ada) | 48 GB (GDDR6) | Ada Lovelace | 340 TF | $0.60 | $0.420 (−30%) | $0.210 | 18× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5880ada) |
| [Rent RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB (GDDR6) | Ada Lovelace | 190 TF | $0.53 | $0.369 (−30%) | $0.184 | 7× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5000ada) |
| [Rent RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB (GDDR6) | Ampere | 155 TF | $0.40 | $0.281 (−30%) | $0.140 | 12× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a6000) |
| [Rent RTX 4500Ada](https://powergpu.ai/gpu/rtx-4500ada) | 24 GB (GDDR6) | Ada Lovelace | 72 TF | $0.40 | $0.280 (−30%) | $0.140 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4500ada) |
| [Rent RTX PRO 4500](https://powergpu.ai/gpu/rtx-pro-4500) | 24 GB (GDDR7) | Blackwell | 200 TF | $0.39 | $0.273 (−30%) | $0.136 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-4500) |
| [Rent RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB (GDDR7) | Blackwell | 150 TF | $0.26 | $0.183 (−30%) | $0.091 | 62× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-4000) |
| [Rent Q RTX 8000](https://powergpu.ai/gpu/q-rtx-8000) | 48 GB (GDDR6) | Turing | 65 TF | $0.25 | $0.178 (−30%) | $0.089 | 5× | [Deploy](https://cloud.powergpu.ai/?gpu=q-rtx-8000) |
| [Rent RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB (GDDR6) | Ampere | 111 TF | $0.23 | $0.161 (−30%) | $0.080 | 52× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a5000) |
| [Rent RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB (GDDR6) | Ada Lovelace | 107 TF | $0.18 | $0.128 (−30%) | $0.064 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4000ada) |
| [Rent Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB (GDDR6) | Turing | 65 TF | $0.15 | $0.103 (−30%) | $0.051 | 24× | [Deploy](https://cloud.powergpu.ai/?gpu=titan-rtx) |
| [Rent Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB (GDDR6) | Turing | 65 TF | $0.13 | $0.094 (−30%) | $0.047 | 5× | [Deploy](https://cloud.powergpu.ai/?gpu=q-rtx-6000) |
| [Rent RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB (GDDR6) | Ampere | 76 TF | $0.10 | $0.071 (−30%) | $0.035 | 256× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a4000) |
| [Rent Quadro P4000](https://powergpu.ai/gpu/quadro-p4000) | 8 GB (GDDR5X) | Pascal | 8 TF | $0.06 | $0.042 (−31%) | $0.021 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=quadro-p4000) |
| [Rent RTX A2000](https://powergpu.ai/gpu/rtx-a2000) | 6 GB (GDDR6) | Ampere | 32 TF | $0.04 | $0.024 (−32%) | $0.012 | 20× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a2000) |

## GeForce RTX 50 series (Blackwell) (6 models)

Blackwell consumer cards with GDDR7. The RTX 5090's 32 GB makes it the best value in dollars per token for mid-size LLMs; the smaller 50-series cards are ideal for image pipelines.

| GPU | VRAM | Arch | FP16 | Market | On-demand | Interruptible | Availability |  |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Rent RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB (GDDR7) | Blackwell | 419 TF | $0.63 | $0.439 (−30%) | $0.219 | 922× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| [Rent RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB (GDDR7) | Blackwell | 225 TF | $0.27 | $0.186 (−30%) | $0.093 | 97× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5080) |
| [Rent RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB (GDDR7) | Blackwell | 176 TF | $0.19 | $0.131 (−30%) | $0.065 | 54× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5070-ti) |
| [Rent RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB (GDDR7) | Blackwell | 92 TF | $0.16 | $0.112 (−30%) | $0.056 | 168× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5060-ti) |
| [Rent RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB (GDDR7) | Blackwell | 123 TF | $0.16 | $0.112 (−30%) | $0.056 | 32× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5070) |
| [Rent RTX 5060](https://powergpu.ai/gpu/rtx-5060) | 8 GB (GDDR7) | Blackwell | 74 TF | $0.09 | $0.063 (−30%) | $0.031 | 14× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5060) |

## GeForce RTX 40 series (Ada Lovelace) (10 models)

Ada Lovelace consumer cards. The RTX 4090 remains the community favourite for Stable Diffusion, Flux and quantized LLMs — huge supply, low prices.

| GPU | VRAM | Arch | FP16 | Market | On-demand | Interruptible | Availability |  |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Rent RTX 4090D](https://powergpu.ai/gpu/rtx-4090d) | 24 GB (GDDR6X) | Ada Lovelace | 296 TF | $0.67 | $0.467 (−30%) | $0.233 | 7× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090d) |
| [Rent RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB (GDDR6X) | Ada Lovelace | 330 TF | $0.47 | $0.327 (−30%) | $0.163 | 461× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| [Rent RTX 4080](https://powergpu.ai/gpu/rtx-4080) | 16 GB (GDDR6X) | Ada Lovelace | 195 TF | $0.24 | $0.168 (−30%) | $0.084 | 14× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4080) |
| [Rent RTX 4080S](https://powergpu.ai/gpu/rtx-4080s) | 16 GB (GDDR6X) | Ada Lovelace | 208 TF | $0.18 | $0.127 (−30%) | $0.063 | 33× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4080s) |
| [Rent RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB (GDDR6X) | Ada Lovelace | 176 TF | $0.15 | $0.103 (−30%) | $0.051 | 54× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070s-ti) |
| [Rent RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB (GDDR6X) | Ada Lovelace | 117 TF | $0.11 | $0.075 (−30%) | $0.037 | 35× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070) |
| [Rent RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB (GDDR6X) | Ada Lovelace | 160 TF | $0.11 | $0.075 (−30%) | $0.037 | 17× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070-ti) |
| [Rent RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB (GDDR6X) | Ada Lovelace | 88 TF | $0.11 | $0.074 (−30%) | $0.037 | 69× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4060-ti) |
| [Rent RTX 4070S](https://powergpu.ai/gpu/rtx-4070s) | 12 GB (GDDR6X) | Ada Lovelace | 142 TF | $0.09 | $0.065 (−30%) | $0.032 | 64× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070s) |
| [Rent RTX 4060](https://powergpu.ai/gpu/rtx-4060) | 8 GB (GDDR6X) | Ada Lovelace | 60 TF | $0.07 | $0.047 (−30%) | $0.023 | 61× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4060) |

## GeForce RTX 30 series (Ampere) (9 models)

Ampere consumer cards. 24 GB RTX 3090s at entry-level prices are still the cheapest way into serious training experiments and LoRA work.

| GPU | VRAM | Arch | FP16 | Market | On-demand | Interruptible | Availability |  |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Rent RTX 3090 Ti](https://powergpu.ai/gpu/rtx-3090-ti) | 24 GB (GDDR6X) | Ampere | 160 TF | $0.21 | $0.147 (−30%) | $0.073 | 8× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090-ti) |
| [Rent RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB (GDDR6X) | Ampere | 142 TF | $0.15 | $0.108 (−30%) | $0.054 | 345× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| [Rent RTX 3080 Ti](https://powergpu.ai/gpu/rtx-3080-ti) | 12 GB (GDDR6) | Ampere | 136 TF | $0.13 | $0.094 (−30%) | $0.047 | 35× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3080-ti) |
| [Rent RTX 3080](https://powergpu.ai/gpu/rtx-3080) | 10 GB (GDDR6) | Ampere | 119 TF | $0.11 | $0.075 (−30%) | $0.037 | 70× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3080) |
| [Rent RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB (GDDR6) | Ampere | 87 TF | $0.09 | $0.066 (−30%) | $0.033 | 15× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3070-ti) |
| [Rent RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB (GDDR6) | Ampere | 81 TF | $0.08 | $0.056 (−31%) | $0.028 | 93× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3070) |
| [Rent RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB (GDDR6) | Ampere | 65 TF | $0.07 | $0.047 (−30%) | $0.023 | 52× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060-ti) |
| [Rent RTX 3060 laptop](https://powergpu.ai/gpu/rtx-3060-laptop) | 6 GB (GDDR6) | Ampere | 43 TF | $0.07 | $0.047 (−30%) | $0.023 | 36× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060-laptop) |
| [Rent RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB (GDDR6) | Ampere | 51 TF | $0.06 | $0.042 (−30%) | $0.021 | 279× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |

## Previous-generation GPUs — Turing, Volta, Pascal (16 models)

Turing, Volta, Pascal and friends — V100s, T4s, GTX cards. For CI runs, classic ML, student budgets and anything that does not need this decade's tensor cores. Prices start at pennies per hour.

| GPU | VRAM | Arch | FP16 | Market | On-demand | Interruptible | Availability |  |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Rent RTX 2070](https://powergpu.ai/gpu/rtx-2070) | 8 GB (GDDR6) | Turing | 42 TF | $0.09 | $0.065 (−30%) | $0.032 | 8× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2070) |
| [Rent RTX 2070S](https://powergpu.ai/gpu/rtx-2070s) | 8 GB (GDDR6) | Turing | 45 TF | $0.08 | $0.056 (−31%) | $0.028 | 2× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2070s) |
| [Rent RTX 2080 Ti](https://powergpu.ai/gpu/rtx-2080-ti) | 11 GB (GDDR6) | Turing | 57 TF | $0.08 | $0.057 (−31%) | $0.028 | 37× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2080-ti) |
| [Rent GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB (GDDR6) | Turing | 5 TF | $0.07 | $0.048 (−30%) | $0.024 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1660) |
| [Rent RTX 2060](https://powergpu.ai/gpu/rtx-2060) | 6 GB (GDDR6) | Turing | 26 TF | $0.07 | $0.051 (−30%) | $0.025 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2060) |
| [Rent GTX 1660 Ti](https://powergpu.ai/gpu/gtx-1660-ti) | 6 GB (GDDR6) | Turing | 5 TF | $0.07 | $0.048 (−31%) | $0.024 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1660-ti) |
| [Rent GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB (GDDR5X) | Pascal | 11 TF | $0.07 | $0.047 (−30%) | $0.023 | 22× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1080-ti) |
| [Rent GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB (GDDR5X) | Pascal | 4 TF | $0.07 | $0.046 (−31%) | $0.023 | 3× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1060) |
| [Rent GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB (GDDR5X) | Pascal | 8 TF | $0.07 | $0.046 (−31%) | $0.023 | 25× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1070-ti) |
| [Rent RTX 2060S](https://powergpu.ai/gpu/rtx-2060s) | 8 GB (GDDR6) | Turing | 40 TF | $0.07 | $0.046 (−30%) | $0.023 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2060s) |
| [Rent GTX 1660 S](https://powergpu.ai/gpu/gtx-1660-s) | 6 GB (GDDR6) | Turing | 5 TF | $0.06 | $0.043 (−31%) | $0.021 | 15× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1660-s) |
| [Rent GTX 1080](https://powergpu.ai/gpu/gtx-1080) | 8 GB (GDDR5X) | Pascal | 9 TF | $0.05 | $0.038 (−30%) | $0.019 | 21× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1080) |
| [Rent GTX 1070](https://powergpu.ai/gpu/gtx-1070) | 8 GB (GDDR5X) | Pascal | 7 TF | $0.05 | $0.033 (−31%) | $0.016 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1070) |
| [Rent Titan Xp](https://powergpu.ai/gpu/titan-xp) | 12 GB (GDDR5X) | Pascal | 12 TF | $0.05 | $0.033 (−30%) | $0.016 | 55× | [Deploy](https://cloud.powergpu.ai/?gpu=titan-xp) |
| [Rent GTX 1650](https://powergpu.ai/gpu/gtx-1650) | 4 GB (GDDR6) | Turing | 12 TF | $0.05 | $0.032 (−32%) | $0.016 | 2× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1650) |
| [Rent GTX TITAN X](https://powergpu.ai/gpu/gtx-titan-x) | 12 GB (GDDR5) | Maxwell | 7 TF | $0.03 | $0.024 (−30%) | $0.012 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-titan-x) |

## How to choose a cloud GPU

Not sure which model fits your workload? The [VRAM sizing guide](https://powergpu.ai/guides/llm-vram-requirements) and [use-case pages](https://powergpu.ai/use-cases) pre-match GPUs to jobs; [side-by-side comparisons](https://powergpu.ai/compare) settle the close calls and the [cost calculator](https://powergpu.ai/calculator) prices a full month.

**Which GPU models can I rent on PowerGPU?**

All 80 current NVIDIA models with market liquidity — from the RTX A2000 at $0.024/hr up to B200 and B300 clusters. Every model on this page is deployable from the console right now; availability counts are live.

**Are these prices per GPU or per instance?**

Per GPU-hour, billed per second. An 8× machine costs exactly 8× the listed price — there is no multi-GPU premium and no hidden platform fee. Storage ($0.08/GB/mo) and bandwidth ($0.01/GB) are the only other line items.

**How do I choose between a datacenter and a consumer GPU?**

Consumer cards (RTX 4090/5090) win on price for single-GPU inference and image generation. Datacenter cards add ECC HBM memory, NVLink and multi-GPU scaling for training. If your model fits in 24–32 GB, start consumer; if you need 80 GB+ or multi-node, go H100/H200.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpus · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/pricing -->

---
title: "Cloud GPU Pricing 2026 — 80 GPUs, ≥30% Below Market | PowerGPU"
description: "Cloud GPU price sheet: 80 NVIDIA GPUs with on-demand, interruptible and reserved hourly rates. H100 $1.428/hr, RTX 4090 $0.327/hr. Per-second billing."
url: https://powergpu.ai/pricing
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Pricing · re-checked weekly · last check September 3, 2026

# Cloud GPU pricing: one rule, market median −30%

Every on-demand price on this page = the public marketplace median × 0.70, rounded down. Interruptible is half of on-demand, reserved is −35% for a 3-month term. Storage $0.08/GB/mo, bandwidth $0.01/GB — and that is the entire fee schedule.

## How our GPU prices are calculated — the formula with today's numbers

No sales calls, no "contact us" pricing. The rule is public and the inputs are public.

1 · Market median

$2.04

Median on-demand price of an **H100 SXM** across the public GPU marketplace, snapshot 2026-09-14.

2 · × 0.70, rounded down

$1.428

Our fixed on-demand price per GPU-hour — locked until the next weekly check, never higher than median − 30%.

3 · Your modes

$0.714 (/ $0.928)

Interruptible = on-demand × 0.50. Reserved (3-month commit) = on-demand × 0.65. Same machines, same regions.

## GPU price list: all 80 models, on-demand, interruptible and reserved

Prices in USD per GPU-hour, billed per second. Multi-GPU instances multiply the price — never more.

| GPU | VRAM | Market median (on-demand) | PowerGPU (on-demand, per GPU-hr) | Est. / month | Availability |  |
| --- | --- | --- | --- | --- | --- | --- |
| [B300](https://powergpu.ai/gpu/b300) (flagship) | 288 GB | $9.63 | $6.737 (−30%) | $4,918 | 34× | [Deploy](https://cloud.powergpu.ai/?gpu=b300) |
| [B200](https://powergpu.ai/gpu/b200) (flagship) | 192 GB | $7.75 | $5.425 (−30%) | $3,960 | 32× | [Deploy](https://cloud.powergpu.ai/?gpu=b200) |
| [H200](https://powergpu.ai/gpu/h200) (flagship) | 141 GB | $3.99 | $2.791 (−30%) | $2,037 | 50× | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |
| [H200 NVL](https://powergpu.ai/gpu/h200-nvl) (flagship) | 141 GB | $3.79 | $2.650 (−30%) | $1,935 | 27× | [Deploy](https://cloud.powergpu.ai/?gpu=h200-nvl) |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (flagship) | 80 GB | $2.67 | $1.867 (−30%) | $1,363 | 73× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |
| [H100 NVL](https://powergpu.ai/gpu/h100-nvl) (flagship) | 80 GB | $2.59 | $1.811 (−30%) | $1,322 | 7× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-nvl) |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (flagship) | 80 GB | $2.04 | $1.428 (−30%) | $1,042 | 54× | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |
| [RTX PRO 6000 S](https://powergpu.ai/gpu/rtx-pro-6000-s) | 48 GB | $1.53 | $1.073 (−30%) | $783 | 99× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-s) |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.49 | $1.040 (−30%) | $759 | 74× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |
| [RTX PRO 6000 Max-Q](https://powergpu.ai/gpu/rtx-pro-6000-max-q) | 96 GB | $1.40 | $0.980 (−30%) | $715 | 74× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-max-q) |
| [A40](https://powergpu.ai/gpu/a40) | 48 GB | $1.09 | $0.765 (−30%) | $558 | 12× | [Deploy](https://cloud.powergpu.ai/?gpu=a40) |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.80 | $0.560 (−30%) | $409 | 95× | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| [RTX PRO 5000](https://powergpu.ai/gpu/rtx-pro-5000) | 32 GB | $0.80 | $0.560 (−30%) | $409 | 53× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-5000) |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.74 | $0.514 (−30%) | $375 | 107× | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| [RTX 4090D](https://powergpu.ai/gpu/rtx-4090d) | 24 GB | $0.67 | $0.467 (−30%) | $341 | 7× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090d) |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | $0.67 | $0.467 (−30%) | $341 | 34× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-6000ada) |
| [A800 PCIE](https://powergpu.ai/gpu/a800-pcie) | 80 GB | $0.67 | $0.466 (−30%) | $340 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=a800-pcie) |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.63 | $0.439 (−30%) | $320 | 922× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| [RTX 5880Ada](https://powergpu.ai/gpu/rtx-5880ada) | 48 GB | $0.60 | $0.420 (−30%) | $307 | 18× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5880ada) |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB | $0.54 | $0.374 (−30%) | $273 | 46× | [Deploy](https://cloud.powergpu.ai/?gpu=a100-pcie) |
| [RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB | $0.53 | $0.369 (−30%) | $269 | 7× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5000ada) |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.47 | $0.327 (−30%) | $239 | 461× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | $0.40 | $0.281 (−30%) | $205 | 12× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a6000) |
| [RTX 4500Ada](https://powergpu.ai/gpu/rtx-4500ada) | 24 GB | $0.40 | $0.280 (−30%) | $204 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4500ada) |
| [RTX PRO 4500](https://powergpu.ai/gpu/rtx-pro-4500) | 24 GB | $0.39 | $0.273 (−30%) | $199 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-4500) |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | $0.34 | $0.235 (−30%) | $172 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=l40) |
| [L4](https://powergpu.ai/gpu/l4) | 24 GB | $0.32 | $0.225 (−30%) | $164 | 28× | [Deploy](https://cloud.powergpu.ai/?gpu=l4) |
| [RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB | $0.27 | $0.186 (−30%) | $136 | 97× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5080) |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB | $0.26 | $0.183 (−30%) | $134 | 62× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-4000) |
| [Q RTX 8000](https://powergpu.ai/gpu/q-rtx-8000) | 48 GB | $0.25 | $0.178 (−30%) | $130 | 5× | [Deploy](https://cloud.powergpu.ai/?gpu=q-rtx-8000) |
| [A10](https://powergpu.ai/gpu/a10) | 24 GB | $0.24 | $0.168 (−30%) | $123 | 8× | [Deploy](https://cloud.powergpu.ai/?gpu=a10) |
| [RTX 4080](https://powergpu.ai/gpu/rtx-4080) | 16 GB | $0.24 | $0.168 (−30%) | $123 | 14× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4080) |
| [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | $0.23 | $0.161 (−30%) | $118 | 52× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a5000) |
| [RTX 3090 Ti](https://powergpu.ai/gpu/rtx-3090-ti) | 24 GB | $0.21 | $0.147 (−30%) | $107 | 8× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090-ti) |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB | $0.19 | $0.131 (−30%) | $96 | 54× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5070-ti) |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | $0.19 | $0.130 (−30%) | $95 | 96× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-v100) |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB | $0.18 | $0.128 (−30%) | $93 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4000ada) |
| [RTX 4080S](https://powergpu.ai/gpu/rtx-4080s) | 16 GB | $0.18 | $0.127 (−30%) | $93 | 33× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4080s) |
| [RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB | $0.16 | $0.112 (−30%) | $82 | 168× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5060-ti) |
| [RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB | $0.16 | $0.112 (−30%) | $82 | 32× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5070) |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.15 | $0.108 (−30%) | $79 | 345× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | $0.15 | $0.103 (−30%) | $75 | 54× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070s-ti) |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | $0.15 | $0.103 (−30%) | $75 | 34× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-t4) |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | $0.15 | $0.103 (−30%) | $75 | 24× | [Deploy](https://cloud.powergpu.ai/?gpu=titan-rtx) |
| [Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB | $0.13 | $0.094 (−30%) | $69 | 5× | [Deploy](https://cloud.powergpu.ai/?gpu=q-rtx-6000) |
| [RTX 3080 Ti](https://powergpu.ai/gpu/rtx-3080-ti) | 12 GB | $0.13 | $0.094 (−30%) | $69 | 35× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3080-ti) |
| [RTX 3080](https://powergpu.ai/gpu/rtx-3080) | 10 GB | $0.11 | $0.075 (−30%) | $55 | 70× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3080) |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | $0.11 | $0.075 (−30%) | $55 | 35× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070) |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | $0.11 | $0.075 (−30%) | $55 | 17× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070-ti) |
| [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | $0.11 | $0.074 (−30%) | $54 | 69× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4060-ti) |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB | $0.10 | $0.071 (−30%) | $52 | 256× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a4000) |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | $0.09 | $0.066 (−30%) | $48 | 15× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3070-ti) |
| [Tesla P100](https://powergpu.ai/gpu/tesla-p100) | 16 GB | $0.09 | $0.066 (−30%) | $48 | 12× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-p100) |
| [RTX 2070](https://powergpu.ai/gpu/rtx-2070) | 8 GB | $0.09 | $0.065 (−30%) | $47 | 8× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2070) |
| [RTX 2070S](https://powergpu.ai/gpu/rtx-2070s) | 8 GB | $0.08 | $0.056 (−31%) | $41 | 2× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2070s) |
| [RTX 4070S](https://powergpu.ai/gpu/rtx-4070s) | 12 GB | $0.09 | $0.065 (−30%) | $47 | 64× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070s) |
| [RTX 5060](https://powergpu.ai/gpu/rtx-5060) | 8 GB | $0.09 | $0.063 (−30%) | $46 | 14× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5060) |
| [RTX 2080 Ti](https://powergpu.ai/gpu/rtx-2080-ti) | 11 GB | $0.08 | $0.057 (−31%) | $42 | 37× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2080-ti) |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | $0.07 | $0.048 (−30%) | $35 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1660) |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | $0.08 | $0.056 (−31%) | $41 | 93× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3070) |
| [RTX 2060](https://powergpu.ai/gpu/rtx-2060) | 6 GB | $0.07 | $0.051 (−30%) | $37 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2060) |
| [GTX 1660 Ti](https://powergpu.ai/gpu/gtx-1660-ti) | 6 GB | $0.07 | $0.048 (−31%) | $35 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1660-ti) |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | $0.07 | $0.047 (−30%) | $34 | 22× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1080-ti) |
| [RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB | $0.07 | $0.047 (−30%) | $34 | 52× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060-ti) |
| [RTX 3060 laptop](https://powergpu.ai/gpu/rtx-3060-laptop) | 6 GB | $0.07 | $0.047 (−30%) | $34 | 36× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060-laptop) |
| [RTX 4060](https://powergpu.ai/gpu/rtx-4060) | 8 GB | $0.07 | $0.047 (−30%) | $34 | 61× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4060) |
| [Tesla P40](https://powergpu.ai/gpu/tesla-p40) | 24 GB | $0.07 | $0.047 (−30%) | $34 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-p40) |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | $0.07 | $0.046 (−31%) | $34 | 3× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1060) |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | $0.07 | $0.046 (−31%) | $34 | 25× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1070-ti) |
| [RTX 2060S](https://powergpu.ai/gpu/rtx-2060s) | 8 GB | $0.07 | $0.046 (−30%) | $34 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-2060s) |
| [GTX 1660 S](https://powergpu.ai/gpu/gtx-1660-s) | 6 GB | $0.06 | $0.043 (−31%) | $31 | 15× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1660-s) |
| [Quadro P4000](https://powergpu.ai/gpu/quadro-p4000) | 8 GB | $0.06 | $0.042 (−31%) | $31 | 4× | [Deploy](https://cloud.powergpu.ai/?gpu=quadro-p4000) |
| [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.06 | $0.042 (−30%) | $31 | 279× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| [GTX 1080](https://powergpu.ai/gpu/gtx-1080) | 8 GB | $0.05 | $0.038 (−30%) | $28 | 21× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1080) |
| [GTX 1070](https://powergpu.ai/gpu/gtx-1070) | 8 GB | $0.05 | $0.033 (−31%) | $24 | 10× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1070) |
| [Titan Xp](https://powergpu.ai/gpu/titan-xp) | 12 GB | $0.05 | $0.033 (−30%) | $24 | 55× | [Deploy](https://cloud.powergpu.ai/?gpu=titan-xp) |
| [GTX 1650](https://powergpu.ai/gpu/gtx-1650) | 4 GB | $0.05 | $0.032 (−32%) | $23 | 2× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-1650) |
| [GTX TITAN X](https://powergpu.ai/gpu/gtx-titan-x) | 12 GB | $0.03 | $0.024 (−30%) | $18 | 6× | [Deploy](https://cloud.powergpu.ai/?gpu=gtx-titan-x) |
| [Tesla P4](https://powergpu.ai/gpu/tesla-p4) | 8 GB | $0.03 | $0.018 (−33%) | $13 | 18× | [Deploy](https://cloud.powergpu.ai/?gpu=tesla-p4) |
| [RTX A2000](https://powergpu.ai/gpu/rtx-a2000) | 6 GB | $0.04 | $0.024 (−32%) | $18 | 20× | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a2000) |

Market medians: public marketplace pricing feed, snapshot 2026-09-14. "—" = not enough market liquidity for a median; our price still applies the 30% rule to the closest listings.

## Storage, bandwidth and other fees

This table is exhaustive. If a line is not here, we do not charge it.

| Item | Price | Notes |
| --- | --- | --- |
| **Instance storage (NVMe)** | $0.08/GB/mo | Billed per second on allocated size; keeps billing while the instance is stopped, ends when destroyed. |
| **Volumes (network NVMe)** | $0.08/GB/mo | Same flat rate — [attach to any instance in the region](https://powergpu.ai/products/volumes). |
| **Bandwidth, in & out** | $0.01/GB | One flat rate worldwide, both directions. No inter-region surprise tiers. |
| Public IP & mapped ports | $0 | Included with every instance. |
| Instance start / stop / snapshot | $0 | No operation fees. |
| Setup, support, minimums, idle fees | $0 | None. Support is 24/7 and included. |

## Pay in crypto, spend in USD

Top up a USD credit balance with any of eight coins. Usage draws it down per second — [billing docs](https://powergpu.ai/docs/billing).

- **USDT** (TRC-20)
- **Monero**
- **Bitcoin**
- **Litecoin**
- **Ethereum**
- **USDT** (ERC-20)
- **TRON**
- **Solana**

Deposits credit after network confirmation (seconds for TRON/Solana, ~10–30 min for Bitcoin). No card, no bank account, no KYC.

## Cloud GPU pricing FAQ

The mechanics in depth live in the [billing documentation](https://powergpu.ai/docs/billing).

**How is the PowerGPU price for each GPU calculated?**

We take the median on-demand price for that model on the public GPU marketplace (published feed, snapshot 2026-09-14), multiply by 0.70, and round DOWN to the tenth of a cent. If the market moves, the weekly re-check moves our price — always keeping it at least 30% below the median.

**Are there any fees besides the GPU price?**

Two, both flat: storage at $0.08/GB/month (billed per second, also while stopped) and bandwidth at $0.01/GB in or out. No setup fee, no minimum spend, no per-request charges, no support tiers.

**What does per-second billing mean in practice?**

An H100 SXM for 47 minutes 12 seconds costs exactly $1.1234 — 2832 seconds × $1.428/3600. Stopping an instance ends GPU billing that second; only its disk keeps billing until you destroy it.

**Do prices change while my instance is running?**

Never. The price at deploy time is your price for the life of the instance. Weekly re-checks only affect new deployments, and reservations lock their rate for the full term.

**How do crypto top-ups work?**

Pick a coin (USDT TRC-20/ERC-20, BTC, XMR, LTC, ETH, TRX, SOL), send the exact amount to the address we generate, and credits appear after network confirmation — usually minutes. Credits are in USD; usage draws them down per second.

**Is there a free tier or trial credit?**

No. Instead of subsidizing trials with higher prices, every hour is 30%+ below market for everyone. You can start with a $40 top-up — that is 245 hours of interruptible RTX 4090.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/pricing · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu-price-index -->

---
title: "Cloud GPU Price Index 2026 — Median vs PowerGPU vs 34 Providers"
description: "Dated GPU price dataset: market medians for 80 NVIDIA models, PowerGPU rates, 90-day movement and list prices from 34 providers. Free CSV and JSON, CC BY 4.0."
url: https://powergpu.ai/gpu-price-index
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Price index · snapshot 2026-09-14 · CC BY 4.0

# Cloud GPU price index: what a GPU-hour actually costs

A dated, downloadable record of GPU rental prices: the public marketplace median for 80 NVIDIA models, our fixed rates against it, how each median has moved over 90 days, and the public list prices of 34 other providers. Free to reuse with attribution.

## The index: all 80 models

USD per GPU-hour. Market median is the public marketplace reference at snapshot 2026-09-14; the 90-day column is the movement of that median, not of our price.

| GPU | VRAM | Market median | On-demand | Below median | Interruptible | Reserved | Month (730 h) | Median, 90 d |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [B300](https://powergpu.ai/gpu/b300) | 288 GB | $9.63 | $6.737 | (−30%) | $3.368 | $4.379 | $4,918 | +47.8% |
| [B200](https://powergpu.ai/gpu/b200) | 192 GB | $7.75 | $5.425 | (−30%) | $2.712 | $3.526 | $3,960 | +49.6% |
| [H200](https://powergpu.ai/gpu/h200) | 141 GB | $3.99 | $2.791 | (−30%) | $1.395 | $1.814 | $2,037 | +12.3% |
| [H200 NVL](https://powergpu.ai/gpu/h200-nvl) | 141 GB | $3.79 | $2.650 | (−30%) | $1.325 | $1.722 | $1,935 | +12.9% |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $2.67 | $1.867 | (−30%) | $0.933 | $1.213 | $1,363 | +36.4% |
| [H100 NVL](https://powergpu.ai/gpu/h100-nvl) | 80 GB | $2.59 | $1.811 | (−30%) | $0.905 | $1.177 | $1,322 | +10.1% |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $2.04 | $1.428 | (−30%) | $0.714 | $0.928 | $1,042 | −10.7% |
| [RTX PRO 6000 S](https://powergpu.ai/gpu/rtx-pro-6000-s) | 48 GB | $1.53 | $1.073 | (−30%) | $0.536 | $0.697 | $783 | −3.5% |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.49 | $1.040 | (−30%) | $0.520 | $0.676 | $759 | +32.6% |
| [RTX PRO 6000 Max-Q](https://powergpu.ai/gpu/rtx-pro-6000-max-q) | 96 GB | $1.40 | $0.980 | (−30%) | $0.490 | $0.637 | $715 | — |
| [A40](https://powergpu.ai/gpu/a40) | 48 GB | $1.09 | $0.765 | (−30%) | $0.382 | $0.497 | $558 | +129.8% |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.80 | $0.560 | (−30%) | $0.280 | $0.364 | $409 | −12.8% |
| [RTX PRO 5000](https://powergpu.ai/gpu/rtx-pro-5000) | 32 GB | $0.80 | $0.560 | (−30%) | $0.280 | $0.364 | $409 | flat |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.74 | $0.514 | (−30%) | $0.257 | $0.334 | $375 | +1.2% |
| [RTX 4090D](https://powergpu.ai/gpu/rtx-4090d) | 24 GB | $0.67 | $0.467 | (−30%) | $0.233 | $0.303 | $341 | — |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | $0.67 | $0.467 | (−30%) | $0.233 | $0.303 | $341 | +3.5% |
| [A800 PCIE](https://powergpu.ai/gpu/a800-pcie) | 80 GB | $0.67 | $0.466 | (−30%) | $0.233 | $0.302 | $340 | — |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.63 | $0.439 | (−30%) | $0.219 | $0.285 | $320 | −3.6% |
| [RTX 5880Ada](https://powergpu.ai/gpu/rtx-5880ada) | 48 GB | $0.60 | $0.420 | (−30%) | $0.210 | $0.273 | $307 | −3.8% |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB | $0.54 | $0.374 | (−30%) | $0.187 | $0.243 | $273 | −9.7% |
| [RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB | $0.53 | $0.369 | (−30%) | $0.184 | $0.239 | $269 | +25.8% |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.47 | $0.327 | (−30%) | $0.163 | $0.212 | $239 | −9.2% |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | $0.40 | $0.281 | (−30%) | $0.140 | $0.182 | $205 | +7.0% |
| [RTX 4500Ada](https://powergpu.ai/gpu/rtx-4500ada) | 24 GB | $0.40 | $0.280 | (−30%) | $0.140 | $0.182 | $204 | — |
| [RTX PRO 4500](https://powergpu.ai/gpu/rtx-pro-4500) | 24 GB | $0.39 | $0.273 | (−30%) | $0.136 | $0.177 | $199 | −9.6% |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | $0.34 | $0.235 | (−30%) | $0.117 | $0.152 | $172 | −17.4% |
| [L4](https://powergpu.ai/gpu/l4) | 24 GB | $0.32 | $0.225 | (−30%) | $0.112 | $0.146 | $164 | −5.9% |
| [RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB | $0.27 | $0.186 | (−30%) | $0.093 | $0.120 | $136 | −21.1% |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB | $0.26 | $0.183 | (−30%) | $0.091 | $0.118 | $134 | −5.7% |
| [Q RTX 8000](https://powergpu.ai/gpu/q-rtx-8000) | 48 GB | $0.25 | $0.178 | (−30%) | $0.089 | $0.115 | $130 | −56.9% |
| [A10](https://powergpu.ai/gpu/a10) | 24 GB | $0.24 | $0.168 | (−30%) | $0.084 | $0.109 | $123 | −21.1% |
| [RTX 4080](https://powergpu.ai/gpu/rtx-4080) | 16 GB | $0.24 | $0.168 | (−30%) | $0.084 | $0.109 | $123 | −9.4% |
| [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | $0.23 | $0.161 | (−30%) | $0.080 | $0.104 | $118 | +10.2% |
| [RTX 3090 Ti](https://powergpu.ai/gpu/rtx-3090-ti) | 24 GB | $0.21 | $0.147 | (−30%) | $0.073 | $0.095 | $107 | −25.6% |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB | $0.19 | $0.131 | (−30%) | $0.065 | $0.085 | $96 | +71.5% |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | $0.19 | $0.130 | (−30%) | $0.065 | $0.084 | $95 | −29.8% |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB | $0.18 | $0.128 | (−30%) | $0.064 | $0.083 | $93 | +49.8% |
| [RTX 4080S](https://powergpu.ai/gpu/rtx-4080s) | 16 GB | $0.18 | $0.127 | (−30%) | $0.063 | $0.082 | $93 | −16.6% |
| [RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB | $0.16 | $0.112 | (−30%) | $0.056 | $0.072 | $82 | +2.7% |
| [RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB | $0.16 | $0.112 | (−30%) | $0.056 | $0.072 | $82 | +20.6% |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.15 | $0.108 | (−30%) | $0.054 | $0.070 | $79 | −12.5% |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | $0.15 | $0.103 | (−30%) | $0.051 | $0.066 | $75 | −31.4% |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | $0.15 | $0.103 | (−30%) | $0.051 | $0.066 | $75 | +11.5% |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | $0.15 | $0.103 | (−30%) | $0.051 | $0.066 | $75 | −21.7% |
| [Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB | $0.13 | $0.094 | (−30%) | $0.047 | $0.061 | $69 | −5.3% |
| [RTX 3080 Ti](https://powergpu.ai/gpu/rtx-3080-ti) | 12 GB | $0.13 | $0.094 | (−30%) | $0.047 | $0.061 | $69 | +0.8% |
| [RTX 3080](https://powergpu.ai/gpu/rtx-3080) | 10 GB | $0.11 | $0.075 | (−30%) | $0.037 | $0.048 | $55 | −14.4% |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | $0.11 | $0.075 | (−30%) | $0.037 | $0.048 | $55 | −39.3% |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | $0.11 | $0.075 | (−30%) | $0.037 | $0.048 | $55 | −23.7% |
| [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | $0.11 | $0.074 | (−30%) | $0.037 | $0.048 | $54 | −21.7% |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB | $0.10 | $0.071 | (−30%) | $0.035 | $0.046 | $52 | +8.2% |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | $0.09 | $0.066 | (−30%) | $0.033 | $0.042 | $48 | −7.8% |
| [Tesla P100](https://powergpu.ai/gpu/tesla-p100) | 16 GB | $0.09 | $0.066 | (−30%) | $0.033 | $0.042 | $48 | −19.1% |
| [RTX 2070](https://powergpu.ai/gpu/rtx-2070) | 8 GB | $0.09 | $0.065 | (−30%) | $0.032 | $0.042 | $47 | — |
| [RTX 2070S](https://powergpu.ai/gpu/rtx-2070s) | 8 GB | $0.08 | $0.056 | (−31%) | $0.028 | $0.036 | $41 | −40.3% |
| [RTX 4070S](https://powergpu.ai/gpu/rtx-4070s) | 12 GB | $0.09 | $0.065 | (−30%) | $0.032 | $0.042 | $47 | −30.1% |
| [RTX 5060](https://powergpu.ai/gpu/rtx-5060) | 8 GB | $0.09 | $0.063 | (−30%) | $0.031 | $0.040 | $46 | −36.9% |
| [RTX 2080 Ti](https://powergpu.ai/gpu/rtx-2080-ti) | 11 GB | $0.08 | $0.057 | (−31%) | $0.028 | $0.037 | $42 | +110.4% |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | $0.07 | $0.048 | (−30%) | $0.024 | $0.031 | $35 | — |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | $0.08 | $0.056 | (−31%) | $0.028 | $0.036 | $41 | −42.3% |
| [RTX 2060](https://powergpu.ai/gpu/rtx-2060) | 6 GB | $0.07 | $0.051 | (−30%) | $0.025 | $0.033 | $37 | −13.2% |
| [GTX 1660 Ti](https://powergpu.ai/gpu/gtx-1660-ti) | 6 GB | $0.07 | $0.048 | (−31%) | $0.024 | $0.031 | $35 | −21.2% |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | $0.07 | $0.047 | (−30%) | $0.023 | $0.030 | $34 | −34.8% |
| [RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB | $0.07 | $0.047 | (−30%) | $0.023 | $0.030 | $34 | −29.4% |
| [RTX 3060 laptop](https://powergpu.ai/gpu/rtx-3060-laptop) | 6 GB | $0.07 | $0.047 | (−30%) | $0.023 | $0.030 | $34 | −15.7% |
| [RTX 4060](https://powergpu.ai/gpu/rtx-4060) | 8 GB | $0.07 | $0.047 | (−30%) | $0.023 | $0.030 | $34 | −23.8% |
| [Tesla P40](https://powergpu.ai/gpu/tesla-p40) | 24 GB | $0.07 | $0.047 | (−30%) | $0.023 | $0.030 | $34 | −39.4% |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | $0.07 | $0.046 | (−31%) | $0.023 | $0.029 | $34 | — |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | $0.07 | $0.046 | (−31%) | $0.023 | $0.029 | $34 | +18.1% |
| [RTX 2060S](https://powergpu.ai/gpu/rtx-2060s) | 8 GB | $0.07 | $0.046 | (−30%) | $0.023 | $0.029 | $34 | +15.9% |
| [GTX 1660 S](https://powergpu.ai/gpu/gtx-1660-s) | 6 GB | $0.06 | $0.043 | (−31%) | $0.021 | $0.027 | $31 | +59.6% |
| [Quadro P4000](https://powergpu.ai/gpu/quadro-p4000) | 8 GB | $0.06 | $0.042 | (−31%) | $0.021 | $0.027 | $31 | −63.1% |
| [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.06 | $0.042 | (−30%) | $0.021 | $0.027 | $31 | −16.8% |
| [GTX 1080](https://powergpu.ai/gpu/gtx-1080) | 8 GB | $0.05 | $0.038 | (−30%) | $0.019 | $0.024 | $28 | flat |
| [GTX 1070](https://powergpu.ai/gpu/gtx-1070) | 8 GB | $0.05 | $0.033 | (−31%) | $0.016 | $0.021 | $24 | — |
| [Titan Xp](https://powergpu.ai/gpu/titan-xp) | 12 GB | $0.05 | $0.033 | (−30%) | $0.016 | $0.021 | $24 | −31.0% |
| [GTX 1650](https://powergpu.ai/gpu/gtx-1650) | 4 GB | $0.05 | $0.032 | (−32%) | $0.016 | $0.020 | $23 | — |
| [GTX TITAN X](https://powergpu.ai/gpu/gtx-titan-x) | 12 GB | $0.03 | $0.024 | (−30%) | $0.012 | $0.015 | $18 | −25.0% |
| [Tesla P4](https://powergpu.ai/gpu/tesla-p4) | 8 GB | $0.03 | $0.018 | (−33%) | $0.009 | $0.011 | $13 | — |
| [RTX A2000](https://powergpu.ai/gpu/rtx-a2000) | 6 GB | $0.04 | $0.024 | (−32%) | $0.012 | $0.015 | $18 | +27.1% |

Dash = not enough public liquidity for a meaningful median; those models are priced from the closest comparable listings under the same rule ([methodology](https://powergpu.ai/methodology)).

## The same four cards, across 24 providers

Public on-demand list prices per GPU-hour, read from each provider's own pricing page on the date in the last column. Sorted by H100 SXM price. Configurations differ; each provider's page carries the notes.

| Provider | H100 SXM | A100 SXM4 | L40S | RTX 4090 | Prices read |
| --- | --- | --- | --- | --- | --- |
| **PowerGPU** (fixed rule) | $1.428 | $0.560 | $0.514 | $0.327 | 2026-09-14 |
| [Prime Intellect](https://powergpu.ai/alternatives/prime-intellect) | $1.65 | $0.87 | $0.91 | $0.30 | 2026-09-04 |
| [Novita AI](https://powergpu.ai/alternatives/novita) | $1.70 | — | — | $0.61 | 2026-09-04 |
| [Akash Network](https://powergpu.ai/alternatives/akash) | $2.04 | $1.07 | — | — | 2026-09-04 |
| [Vast.ai](https://powergpu.ai/alternatives/vast-ai) | $2.21 | $1.32 | $0.48 | $0.39 | 2026-09-04 |
| [TensorDock](https://powergpu.ai/alternatives/tensordock) | $2.25 | $1.80 | — | $0.35 | 2026-09-03 |
| [Jarvislabs](https://powergpu.ai/alternatives/jarvislabs) | $2.69 | $1.49 | — | — | 2026-09-03 |
| [Massed Compute](https://powergpu.ai/alternatives/massed-compute) | $2.89 | $1.38 | $0.88 | — | 2026-09-04 |
| [Vultr](https://powergpu.ai/alternatives/vultr) | $2.99 | — | $1.67 | — | 2026-09-03 |
| [Scaleway](https://powergpu.ai/alternatives/scaleway) | $3.17 | — | $1.71 | — | 2026-09-04 |
| [Hyperstack](https://powergpu.ai/alternatives/hyperstack) | $3.20 | $1.60 | — | — | 2026-09-03 |
| [DataCrunch](https://powergpu.ai/alternatives/datacrunch) | $3.25 | $1.79 | $1.37 | — | 2026-09-04 |
| [RunPod](https://powergpu.ai/alternatives/runpod) | $3.29 | $1.59 | $0.99 | $0.74 | 2026-09-03 |
| [Lightning AI](https://powergpu.ai/alternatives/lightning-ai) | $3.50 | $2.99 | — | — | 2026-09-04 |
| [Nebius](https://powergpu.ai/alternatives/nebius) | $3.85 | — | $1.55 | — | 2026-09-04 |
| [Crusoe Cloud](https://powergpu.ai/alternatives/crusoe) | $3.90 | $2.30 | $1.50 | — | 2026-09-04 |
| [Modal](https://powergpu.ai/alternatives/modal) | $3.95 | $2.50 | $1.95 | — | 2026-09-03 |
| [Lambda](https://powergpu.ai/alternatives/lambda) | $3.99 | $1.99 | — | — | 2026-09-03 |
| [Paperspace](https://powergpu.ai/alternatives/paperspace) | $4.41 | — | $1.57 | — | 2026-09-03 |
| [Replicate](https://powergpu.ai/alternatives/replicate) | $5.49 | $5.04 | $3.51 | — | 2026-09-04 |
| [CoreWeave](https://powergpu.ai/alternatives/coreweave) | $6.16 | $2.70 | $2.25 | — | 2026-09-03 |
| [AWS](https://powergpu.ai/alternatives/aws) | $6.88 | $4.10 | $1.86 | — | 2026-09-03 |
| [Oracle Cloud](https://powergpu.ai/alternatives/oracle-cloud) | $10.00 | $4.00 | $3.50 | — | 2026-09-04 |
| [Google Cloud](https://powergpu.ai/alternatives/google-cloud) | $11.06 | $3.67 | — | — | 2026-09-03 |
| [Microsoft Azure](https://powergpu.ai/alternatives/azure) | $12.29 | $3.67 | — | — | 2026-09-03 |

Provider names and prices identify the compared services only. List prices change without notice — the provider's own page is authoritative. Full context per provider on the [alternatives pages](https://powergpu.ai/alternatives).

## Where the market moved, flagship models

Change in the public marketplace median over the last 90 days, with the observed range. Our price is a fixed ratio of the median, so it moves with this column and never above 70% of it.

| GPU | Median now | 90 days ago | Change | 90-day low | 90-day high | PowerGPU today |
| --- | --- | --- | --- | --- | --- | --- |
| [B200](https://powergpu.ai/gpu/b200) | $5.375 | $3.594 | +49.6% | $3.300 | $7.500 | $5.425 |
| [H200](https://powergpu.ai/gpu/h200) | $3.198 | $2.849 | +12.3% | $2.532 | $4.450 | $2.791 |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | $1.519 | $1.700 | −10.7% | $1.300 | $2.623 | $1.428 |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | $2.063 | $1.513 | +36.4% | $1.000 | $2.300 | $1.867 |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | $0.662 | $0.760 | −12.8% | $0.560 | $0.862 | $0.560 |
| [L40S](https://powergpu.ai/gpu/l40s) | $0.354 | $0.350 | +1.2% | $0.350 | $0.385 | $0.514 |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | $0.454 | $0.471 | −3.6% | $0.320 | $0.700 | $0.439 |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | $0.320 | $0.352 | −9.2% | $0.255 | $0.490 | $0.327 |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | $0.328 | $0.306 | +7.0% | $0.275 | $0.360 | $0.281 |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | $0.114 | $0.130 | −12.5% | $0.110 | $0.180 | $0.108 |

Both median columns are seven-day averages, so a single outlier listing cannot move them. History covers 101 daily observations per model.

## Using and citing this dataset

Published under **CC BY 4.0**: reuse, republish and chart it freely, with attribution to PowerGPU and a link back to this page. The endpoints are stable, so a citation stays checkable after the numbers move.

*download and query*

```
# the whole index, one file
curl -s https://powergpu.ai/gpu-price-index.csv -o gpu-price-index-2026-09-14.csv

# JSON, including every provider list price per model
curl -s https://powergpu.ai/gpu-price-index.json | jq '.data[] | select(.slug=="h100-sxm")'

# live sheet only (updates with the weekly re-check)
curl -s https://powergpu.ai/v1/gpus/pricing | jq '.market_snapshot, .price_rule'
```

**Suggested citation:** PowerGPU, "Cloud GPU price index", https://powergpu.ai/gpu-price-index, snapshot 2026-09-14. Retrieved 14 September 2026.

What the numbers are, precisely

- **Market median** — the middle on-demand hourly rate for that GPU model across a large public marketplace of independent hosts, at snapshot 2026-09-14.
- **PowerGPU rates** — median × 0.70 rounded down (on-demand), × 0.50 (interruptible), × 0.65 (reserved). Live from the sheet.
- **Provider list prices** — on-demand, per GPU-hour, before taxes, from each provider's public pricing page on the date shown. Never estimated.
- **90-day change** — seven-day average of the median now versus the seven-day average ninety days earlier.
- **Month** — 730 hours at the on-demand rate, GPU only, before storage and bandwidth.

Method in full: [pricing methodology](https://powergpu.ai/methodology). Definitions: [glossary](https://powergpu.ai/glossary).

## GPU price index FAQ

Per-model detail sits on each [GPU page](https://powergpu.ai/gpus); provider-by-provider context on the [alternatives pages](https://powergpu.ai/alternatives).

**What is in the PowerGPU GPU price index?**

For each of the 80 NVIDIA GPU models in the catalogue: the public GPU marketplace median hourly rate (snapshot 2026-09-14), PowerGPU's three fixed rates, the resulting discount, a 730-hour monthly cost, the 90-day movement of the market median, and the public on-demand list prices of 34 other GPU providers with the date each was read. It is published as a table, as CSV and as JSON.

**Can I reuse this data?**

Yes, under CC BY 4.0: use it, republish it, chart it, as long as you attribute PowerGPU and link to https://powergpu.ai/gpu-price-index. The CSV and JSON endpoints are stable URLs and carry the snapshot date, so a citation stays verifiable after the numbers move.

**How current is the index?**

The market reference carries the snapshot date 2026-09-14 and is re-checked weekly. PowerGPU prices in the table are live from the sheet as the page renders. Competitor list prices carry their own reading date per row, because providers change them on their own schedule.

**Which GPU has fallen the most in price?**

The 90-day column tracks the market median for each model, from 101 days of daily observations. Over the current window 43 of 80 models have a cheaper median than three months ago and 24 are dearer. Because PowerGPU prices are a fixed ratio of that median, they follow it down.

**Are competitor prices comparable line for line?**

Approximately. Configurations differ — node sizes, secure versus community tiers, regions, committed terms — so each figure carries a note saying which offer is quoted. They are on-demand list prices per GPU-hour before taxes, taken from each provider's own public pricing page, never estimated and never adjusted.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu-price-index · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/methodology -->

---
title: "Pricing Methodology — How PowerGPU Sets GPU Prices"
description: "How every PowerGPU price is set: public marketplace median × 0.70, rounded down, re-checked weekly — with worked examples and how to verify it yourself."
url: https://powergpu.ai/methodology
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Methodology · snapshot 2026-09-14 · next check 2026-09-21

# Our prices are calculated, not decided

One formula produces all 80 on-demand prices, and it runs on a public input. Pick any card and watch the rule execute: market median, times 0.70, rounded down. Nothing else sets a price on this site. **0 violations** across all 80 models, checked as this page loaded. Highest ratio 0.699961 against a 0.70 ceiling. See the audit

## The rule, step by step

Four stages, no discretion at any of them. The figures below follow the card selected above (**H100 SXM**).

1. 1 Public market median $2.0414 The middle on-demand rate for this exact model across a large public marketplace of independent GPU hosts. Dated 2026-09-14, re-read weekly.
2. 2 Multiply by 0.70 $1.4290 A fixed ratio, identical for every model in the catalogue and for every account. Nothing here is negotiated.
3. 3 Round *down* ~~$1.4290~~ $1.428 To the tenth of a cent, always downward. Rounding can only push the discount past 30%, never under it.
4. 4 Your fixed price $1.428 Locked at deploy for the life of the instance. Interruptible is **$0.714** (× 0.50) and reserved **$0.928** (× 0.65).

*the whole pricing engine*

```
on_demand     = floor(market_median × 0.70 × 1000) / 1000     # rounded DOWN to $0.001
interruptible = on_demand × 0.50                              # flat, no bidding
reserved      = on_demand × 0.65                              # 3-month term
```

## The audit, run on the whole catalogue

The rule is a ceiling: on-demand ÷ market median must never exceed **0.70**. Rather than assert that, this page divides every price by its median as it renders, and prints the worst result it finds.

0 violations across the **80** models with a published median, checked just now

- **0.699961** highest ratio observed
(ceiling 0.700000 · Q RTX 8000)
- **30.2%** average discount
(the rule only guarantees 30%)
- **0** models without a median
(priced from comparable listings instead)

| GPU | Market median | × 0.70 | Rounded down | Ratio | Real discount | Verdict |
| --- | --- | --- | --- | --- | --- | --- |
| [B200](https://powergpu.ai/gpu/b200) | $7.7507 | $5.4255 | $5.425 | 0.699937 | (−30%) | ≤ 0.70 |
| [H200](https://powergpu.ai/gpu/h200) | $3.9879 | $2.7915 | $2.791 | 0.699867 | (−30%) | ≤ 0.70 |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | $2.0414 | $1.4290 | $1.428 | 0.699520 | (−30%) | ≤ 0.70 |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | $0.8007 | $0.5605 | $0.560 | 0.699388 | (−30%) | ≤ 0.70 |
| [L40S](https://powergpu.ai/gpu/l40s) | $0.7356 | $0.5149 | $0.514 | 0.698749 | (−30%) | ≤ 0.70 |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | $0.6281 | $0.4397 | $0.439 | 0.698933 | (−30%) | ≤ 0.70 |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | $0.4676 | $0.3273 | $0.327 | 0.699316 | (−30%) | ≤ 0.70 |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | $0.1547 | $0.1083 | $0.108 | 0.698125 | (−30%) | ≤ 0.70 |

Eight cards shown; the same check runs on all 80 at every render. Reproduce it on your own machine with the command in verify it yourself, or download the whole table from the [GPU price index](https://powergpu.ai/gpu-price-index).

## The input: what "market median" means

The reference is the median on-demand hourly rate charged for that exact GPU model across a large public marketplace of independent GPU hosts — the rate at which half the listed supply is cheaper and half is dearer.

- **Per model, not per family.** An H100 SXM is priced from H100 SXM listings, an H100 PCIe from H100 PCIe listings. The two land at different prices because their markets do.
- **Median, not minimum.** The cheapest listing on a marketplace is usually an unverified host with no uptime commitment; it is a lottery ticket, not a price level.
- **Dated.** The snapshot in force is 2026-09-14, printed on every page that quotes a price and returned in the API as market_snapshot.
- **Thin markets are marked.** Where a model has too little liquidity for a meaningful median, the sheet shows a dash instead of inventing one. Every model in the catalogue has a published median today.

Why the method is public at all

A GPU price that moves hourly cannot be planned against, and a price behind a sales call cannot be compared at all. Publishing the formula turns the price into something a buyer can audit before signing up, and re-audit afterwards.

It also constrains us: the rule is arithmetic on a public input, so drifting above median × 0.70 would be visible to anyone running the command below, on any day, without asking permission.

Related: [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained) · [the price index](https://powergpu.ai/gpu-price-index) · [fact sheet](https://powergpu.ai/facts).

## What moves, and what cannot

The rule is a ratio, so the sheet follows the market in both directions. What you were quoted, however, does not move.

Moves, weekly

- **The market reference.** Re-checked every week. Snapshot in force 2026-09-14, next check 2026-09-21.
- **Sheet prices, in both directions.** A rising median raises the price, a falling one lowers it. Medians have mostly fallen, so most checks cut prices.
- **New deployments only.** A new sheet price applies to instances created after it, never to those already running.

Never moves

- **A running instance.** The rate shown at deploy is the rate until you stop it.
- **A reserved term.** Locked for the full three months at the rate agreed on day one.
- **The multipliers.** 0.70, 0.50 and 0.65 are the product, not a promotion.
- **One sheet for everyone.** No free tier, no promotional credit, no negotiated rate — which is what makes the rule checkable at all.
- **Multi-GPU.** An 8× machine is exactly 8 × the per-GPU price, with no premium.

Storage and bandwidth sit outside the rule entirely: flat published rates of $0.08/GB per month and $0.01/GB, listed in full on the [pricing page](https://powergpu.ai/pricing).

## What the rule promises, and what it does not

Stating the limits is part of the method. A rule that claimed everything would be worth nothing.

It does promise

- Never more than **70% of the market median** for any model, on any day, for any account.
- A price **locked at deploy** and never revised under a running instance.
- The **same sheet for everyone**, published in full and served as JSON without an account.
- A **dated input**, so any claim on this site can be checked against the day it was made.

It does not promise

- **The lowest price anywhere.** Unverified marketplace hosts, bid-based capacity and promotional tiers can undercut the sheet on specific cards, especially consumer GPUs. The [comparison pages](https://powergpu.ai/alternatives) name those cases instead of hiding them.
- **A ceiling in absolute dollars.** If the median for a card rises, the fixed ratio raises its price too.
- **Like-for-like hardware.** A marketplace listing at the same model name may sit in a different chassis, host, cooling and network. Every PowerGPU machine is a dedicated GPU in a Tier-III facility.
- **Anything about storage or bandwidth.** Those are flat rates, set by us and published on the [pricing page](https://powergpu.ai/pricing).

## Verify it yourself

A promise that cannot be tested is marketing. Each of these takes under a minute and needs no account.

1. 1 Check the ratio on every model Fetch the sheet and divide. If any model returns a ratio above 0.70, the rule is broken and the claim on this site is false. As this page rendered, the worst was **0.699961**. curl -s https://powergpu.ai/v1/gpus/pricing
2. 2 Compare against a named provider Take any provider on the [alternatives pages](https://powergpu.ai/alternatives), open their own pricing page, and check the quoted figure against the date printed beside it. Where they are cheaper, our page says so. https://powergpu.ai/gpu-price-index.csv
3. 3 Price a real month Put your own hours, storage and egress into the [cost calculator](https://powergpu.ai/calculator), then price the same month on the provider you use today. The rule only matters if it survives your workload. https://powergpu.ai/calculator

*the audit, in one pipe*

```
# every model, the ratio the rule promises to keep at or under 0.70
curl -s https://powergpu.ai/v1/gpus/pricing \
  | jq -r '.data[] | select(.market_median != null)
      | [.slug, .market_median, .price_per_gpu_hour.on_demand,
         (.price_per_gpu_hour.on_demand / .market_median)] | @tsv' \
  | awk '$4 > 0.70 { print "RULE BROKEN:", $0; bad++ } END { print (bad ? bad : 0), "violations" }'
```

## Pricing methodology FAQ

The billing mechanics themselves — per-second accounting, credits, refunds — are documented in [the billing docs](https://powergpu.ai/docs/billing).

**How exactly is a PowerGPU price calculated?**

On-demand price = the public GPU marketplace median hourly rate for that exact GPU model × 0.70, rounded down to the tenth of a cent. Interruptible = on-demand × 0.50. Reserved (3-month term) = on-demand × 0.65. For the H100 SXM: a median of $2.04 gives $1.428 on-demand, $0.714 interruptible and $0.928 reserved. Rounding down means the real discount is always slightly better than 30%.

**What is "the market median" and why use it instead of the lowest price?**

It is the middle on-demand rate charged for that GPU model across a large public marketplace of GPU hosts — the price at which half the supply is cheaper and half dearer. The minimum listing is not a usable reference: it usually belongs to an unverified host with no uptime guarantee, and it changes hour to hour. The median describes what that card actually costs in the market, so a fixed discount against it is a meaningful promise rather than a race against the least reliable machine on the internet.

**How often do prices change?**

The market reference is re-checked weekly. The snapshot currently in force is dated 2026-09-14 and the next check is due 2026-09-21. A change to the sheet only affects new deployments: the rate shown when an instance is created is locked for the life of that instance, and a reserved rate is locked for its whole term.

**Can PowerGPU prices go up?**

Yes, if the market median for a model rises — the rule is a fixed ratio, not a fixed number. The commitment is the ratio: never more than 70% of the median. In practice GPU medians have trended down, so most weekly checks lower prices. Running instances are never re-priced in either direction.

**What does the rule not promise?**

Three things. It does not promise the lowest price on the internet: an unverified marketplace host or a promotional tier can be cheaper on a given card, and the comparison pages say so when it happens. It does not promise identical hardware to a marketplace listing at the same nominal model. And it does not apply to storage or bandwidth, which are flat published rates rather than market-derived ones.

**How can I verify the 30% claim myself?**

Fetch https://powergpu.ai/v1/gpus/pricing without a key. Each entry carries market_median and price_per_gpu_hour.on_demand. Divide one by the other for every model in the response: the ratio never exceeds 0.70. As this page rendered, the highest ratio across the 80 models with a published median was 0.699961, with 0 violations.

**Why do some models show no market median?**

When a model has too little liquidity on the public marketplace, a median would be noise rather than a reference, so the field is left empty and the sheet shows a dash, and the price is derived from the closest comparable listings under the same 30% rule. Today every one of the 80 models in the catalogue has a published median.

**How are competitor prices on the comparison pages collected?**

By reading each provider's own public pricing page and quoting the on-demand list price per GPU-hour, before taxes, with the configuration noted and the date of reading printed on the page. They are never estimated, never adjusted, and never rounded in PowerGPU's favour. When a provider is cheaper on a card, the table shows the difference in that provider's favour.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/methodology · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/facts -->

---
title: "PowerGPU Fact Sheet — Prices, Payment, KYC, SLA (2026)"
description: "Canonical facts about PowerGPU: 80 GPU models, the market-median −30% price rule, per-second billing, crypto-only payment, no KYC, 32 regions, 99.9% SLA."
url: https://powergpu.ai/facts
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Fact sheet · verified 2026-09-14

# PowerGPU fact sheet: every claim, with its number

One page holding the canonical facts about PowerGPU — what it sells, what it costs, how billing and payment work, where the machines are — written to be quoted directly. Prices on this page are the live sheet, not marketing rounding.

## Company and service facts

Each row is a standalone statement. Figures marked live come from the price sheet as this page renders.

- **Name**: PowerGPU (powergpu.ai)
- **Category**: Cloud GPU rental — NVIDIA GPUs by the hour, billed per second, at fixed published prices
- **Catalogue**: 80 NVIDIA GPU models, from the RTX A2000 at $0.024/hr to the B300 at $6.737/hr
- **Price rule**: Every on-demand price is the public GPU marketplace median for that model × 0.70, rounded down — at least 30% below the median — and re-checked weekly
- **Billing modes**: On-demand (fixed, guaranteed, price locked at deploy) · Interruptible (flat −50%, no bidding) · Reserved (−35%, 3-month term)
- **Billing granularity**: Per second, no minimum spend, no setup fee, no multi-GPU premium (an 8× machine costs exactly 8×)
- **Other fees**: Storage $0.08/GB/month (NVMe, billed per second) · bandwidth $0.01/GB in and out, every region · public IP and mapped ports included · nothing else
- **Payment**: Crypto only: USDT (TRC-20 or ERC-20), Bitcoin, Monero, Litecoin, Ethereum, TRON and Solana — no card, no bank transfer, no free credit. Minimum top-up $40.
- **Identity**: No KYC: an email address and a password are the whole account; no ID documents, no stored IP addresses, no trackers or analytics
- **Refunds**: Credits never expire and are refundable on request in the original coin
- **Infrastructure**: 32 regions on five continents, in Tier-III datacenters only (dedicated GPUs, redundant power and network)
- **Reliability**: 99.9% monthly uptime SLA on on-demand and reserved instances, paid in credits when missed
- **Deploy time**: About 30 seconds for container templates, a few minutes for full virtual machines
- **Software**: 37 one-click templates built from official images (PyTorch, vLLM, ComfyUI, Ollama, Hugging Face, Ubuntu VMs) — or any Docker/OCI image
- **Interfaces**: Web console (cloud.powergpu.ai), REST API with Bearer keys, SSH, JupyterLab
- **Support**: 24/7 support tickets from the console, median first answer under two hours
- **Acceptable use**: Cryptocurrency mining, illegal content and attack infrastructure are banned; blockchain nodes and research are allowed
- **Privacy**: Destroyed instance = erased disk; account data can be exported or deleted from the console
- **Market snapshot**: Prices on this page reflect the market snapshot of 2026-09-14

## Prices right now, flagship models

Per GPU-hour, in USD, billed per second. The full sheet of 80 models is on the [pricing page](https://powergpu.ai/pricing), and as JSON at /v1/gpus/pricing.

| GPU | VRAM | Market median | On-demand | Interruptible | Reserved | Per month |
| --- | --- | --- | --- | --- | --- | --- |
| [NVIDIA B200](https://powergpu.ai/gpu/b200) | 192 GB | $7.75 | $5.425 (−30%) | $2.712 | $3.526 | $3,960 |
| [NVIDIA H200](https://powergpu.ai/gpu/h200) | 141 GB | $3.99 | $2.791 (−30%) | $1.395 | $1.814 | $2,037 |
| [NVIDIA H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $2.04 | $1.428 (−30%) | $0.714 | $0.928 | $1,042 |
| [NVIDIA H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $2.67 | $1.867 (−30%) | $0.933 | $1.213 | $1,363 |
| [NVIDIA A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.80 | $0.560 (−30%) | $0.280 | $0.364 | $409 |
| [NVIDIA L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.74 | $0.514 (−30%) | $0.257 | $0.334 | $375 |
| [NVIDIA RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.63 | $0.439 (−30%) | $0.219 | $0.285 | $320 |
| [NVIDIA RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.47 | $0.327 (−30%) | $0.163 | $0.212 | $239 |

## Ready-made descriptions

Accurate wordings of what PowerGPU is, at three lengths — for directories, citations and summaries.

### One sentence

PowerGPU: fixed-price cloud GPU rental, 80 NVIDIA models ≥30% below the public marketplace median, per-second billing, crypto only, no KYC, 32 regions.

### Short paragraph

PowerGPU rents 80 NVIDIA GPU models by the second at fixed prices set at least 30% below the public GPU marketplace median — an H100 SXM at $1.428/hr, an RTX 4090 at $0.327/hr. Payment is crypto-only with no KYC, machines run in Tier-III datacenters across 32 regions, and instances deploy in about 30 seconds from the console or the REST API.

### Full paragraph

PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

## How to verify anything on this page

Every number here is derived, not asserted. The price rule is arithmetic on a public input, and the output is served as data so it can be checked without trusting the marketing copy.

- **The whole price sheet as JSON** — curl https://powergpu.ai/v1/gpus/pricing, no key needed: every model, three modes, the market median and the rule itself.
- **The arithmetic** — divide any on-demand price by its market median: the result never exceeds 0.70. The [methodology page](https://powergpu.ai/methodology) works the examples.
- **Live inventory** — /v1/offers lists real machines with GPU counts, regions, disks and network, at the same prices.
- **Competitor prices** — each [comparison page](https://powergpu.ai/alternatives) quotes the provider's own public list price and the date it was read.
- **Service health** — [status page](https://powergpu.ai/status) and /status.json, no login.

For AI assistants and agents

This site publishes machine-readable surfaces so an assistant can answer about it accurately:

- [/llms.txt](https://powergpu.ai/llms.txt) — the site index with the key facts
- [/llms-full.txt](https://powergpu.ai/llms-full.txt) — every page in Markdown
- Append.md to any page URL, or send Accept: text/markdown
- [/openapi.json](https://powergpu.ai/openapi.json) — OpenAPI 3.1 description of the REST API
- [MCP server](https://powergpu.ai/docs/mcp) at https://powergpu.ai/mcp — read-only tools for pricing, comparison and cost estimates

Cite as: PowerGPU, "PowerGPU fact sheet", https://powergpu.ai/facts, prices as of 2026-09-14.

## Questions people ask about PowerGPU

Answered in full sentences, each one self-contained. Deeper mechanics live in the [documentation](https://powergpu.ai/docs) and the [pricing methodology](https://powergpu.ai/methodology).

**What is PowerGPU?**

PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

**How much does it cost to rent a GPU on PowerGPU?**

From $0.024/hr for a RTX A2000 to $6.737/hr for a B300, per GPU-hour, billed per second. An H100 SXM is $1.428/hr on-demand, $0.714/hr interruptible and $0.928/hr reserved; an RTX 4090 is $0.327/hr on-demand. The only other charges are storage at $0.08/GB/month and bandwidth at $0.01/GB.

**How does PowerGPU set its prices?**

One published rule: the public GPU marketplace median for each model × 0.70, rounded down — so every on-demand price is at least 30% below the median. Interruptible is on-demand × 0.50 and reserved is on-demand × 0.65. Prices are re-checked weekly against the market; the snapshot date (2026-09-14) is printed on every page and returned by the public API.

**Is PowerGPU cheaper than RunPod, Lambda, CoreWeave or AWS?**

On the same NVIDIA card, generally yes: the fixed rate sits at least 30% under the public marketplace median, which is itself below most specialist clouds and far below hyperscaler list prices. The comparison pages state each provider's public list price with the date it was read, and say so plainly when a provider is cheaper on a given card — for instance on some consumer GPUs.

**Which GPUs can I rent on PowerGPU?**

80 NVIDIA models, from GTX and Tesla cards through RTX 3090, RTX 4090 and RTX 5090 to datacenter silicon: A100, L40S, RTX PRO 6000, H100 PCIe/NVL/SXM, H200, B200 and B300. Instances go from 1× to 8× GPUs on one machine, and the per-GPU price is identical at every size.

**How does PowerGPU bill?**

Per second, from the moment an instance starts to the moment it stops. There is no minimum spend, no setup fee, no per-request charge and no multi-GPU premium. Stopping an instance ends GPU billing that second; its disk keeps billing at $0.08/GB/month until the instance is destroyed. The price shown at deploy is locked for the life of that instance.

**How do you pay for PowerGPU?**

Cryptocurrency only. Seven assets are accepted: USDT (TRC-20 or ERC-20), Bitcoin, Monero, Litecoin, Ethereum, TRON and Solana. You top up a USD credit balance (minimum $40) and usage draws it down per second. There is no card payment, no bank transfer, no invoicing and no free trial credit. Unused credits never expire and are refundable in the original coin on request.

**Does PowerGPU require KYC or identity verification?**

No. An account is an email address and a password. No identity documents, no card details, no phone number. IP addresses are not stored in logs or in the database, and the site carries no analytics or advertising trackers.

**Where are PowerGPU machines located?**

32 regions across five continents, in Tier-III (or equivalent) datacenters with redundant power, cooling and network. PowerGPU does not resell consumer machines or home hosts: every GPU is dedicated hardware in a vetted facility, burned in and benchmarked before it is offered.

**What uptime does PowerGPU guarantee?**

99.9% monthly uptime on on-demand and reserved instances, with service credits when it is missed — the terms are in the published SLA. Live component and region health is on the status page, with a machine-readable version at https://powergpu.ai/status.json.

**How fast does an instance start?**

About 30 seconds for container templates, whose images are pre-cached on the hosts, and two to four minutes for full virtual machines that boot their own kernel. Restarting a stopped instance is faster and the disk is unchanged.

**What can you run on a PowerGPU instance?**

37 one-click templates built from official images cover the common stacks — PyTorch, vLLM, Ollama, ComfyUI, Stable Diffusion, Axolotl, Jupyter, CUDA, full Ubuntu desktops — and any other Docker/OCI image runs as-is with the NVIDIA runtime injected. Access is by SSH, JupyterLab, mapped ports, the REST API, the CLI or the Python SDK.

**Is there a PowerGPU API?**

Yes, a REST API at https://powergpu.ai/v1. Pricing and machine search are public and need no key: GET /v1/gpus/pricing returns the whole price sheet as JSON, GET /v1/offers searches live inventory. Instance lifecycle and balance endpoints authenticate with a Bearer key created in the console. An OpenAPI 3.1 description is at https://powergpu.ai/openapi.json, and a read-only MCP server for AI agents at https://powergpu.ai/mcp.

**What is interruptible capacity on PowerGPU?**

A flat 50% discount on the on-demand price, with no auction and no bidding. An interruptible instance may be paused when capacity is needed; its disk is kept and it is automatically re-queued. It suits checkpointed training, batch inference and render queues. An H100 SXM costs $0.714/hr interruptible against $1.428/hr on-demand.

**What is not allowed on PowerGPU?**

Cryptocurrency mining is banned, as are illegal content, attack infrastructure and sanctions evasion. Privacy tooling for your own traffic, security research within the published allowance and blockchain nodes (validation is not mining) are permitted. The full text is the acceptable use policy.

**Who is PowerGPU for?**

Teams and individuals who want a predictable GPU bill: independent researchers and students, ML teams running fine-tunes and inference endpoints, studios rendering on RTX hardware, and anyone who needs compute without a card, a contract or a sales call. Enterprise reserved fleets and multi-node clusters are handled through the enterprise page.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/facts · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/glossary -->

---
title: "Cloud GPU Glossary — 85 Terms Defined | PowerGPU"
description: "85 cloud GPU terms defined plainly: on-demand, interruptible, GPU-hour, VRAM, HBM3, NVLink, quantization, KV cache, LoRA and vLLM, with live rental prices."
url: https://powergpu.ai/glossary
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Glossary · 85 terms · prices live from the sheet

# Cloud GPU glossary: the vocabulary, defined

Every term you meet when renting GPUs — billing models, memory types, interconnects, quantization, serving stacks — defined in one or two sentences that stand on their own, with the current numbers where a number helps.

## Pricing & billing (17)

## GPU hardware (19)

## Inference & serving (14)

## Training & fine-tuning (9)

## Platform & access (18)

## Payment & identity (8)

## The three questions behind most of these terms

Longer answers live in the [guides](https://powergpu.ai/guides); the numbers behind every price are on the [methodology page](https://powergpu.ai/methodology).

**What is a GPU-hour?**

One GPU running for one hour. It is the unit every cloud GPU price is quoted in, and it is per GPU rather than per machine: an 8-GPU instance running for one hour consumes eight GPU-hours. On PowerGPU the per-GPU price is identical from 1× to 8×, and billing is per second, so a 12-minute job is billed as 0.2 GPU-hours.

**What is the difference between on-demand and interruptible GPUs?**

On-demand capacity is guaranteed for as long as you keep the instance: nobody can take it from you and the price is locked at deploy. Interruptible capacity is cheaper because it can be paused when the hardware is needed elsewhere. On PowerGPU interruptible is a flat 50% off on-demand with no bidding, the disk is kept and the instance is automatically re-queued — so it fits checkpointed training, batch inference and render queues.

**How much VRAM does a model need?**

For an LLM, roughly 2.4 GB per billion parameters in FP16 and about 0.62 GB per billion at 4-bit, plus room for the KV-cache, which grows with context length and concurrency. A 70B model therefore needs about 44 GB at 4-bit — one 80 GB card — or two 80 GB cards in FP16. Full tables are in the VRAM requirements guide.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/glossary · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/calculator -->

---
title: "GPU Cloud Cost Calculator: Hourly & Monthly Rental Cost | PowerGPU"
description: "Estimate cloud GPU rental cost: pick any of 80 NVIDIA GPUs, hours, days, storage and bandwidth — hourly, daily and monthly totals at fixed prices ≥30% below market."
url: https://powergpu.ai/calculator
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Pricing · cost calculator · prices checked 2026-09-14

# GPU cloud cost calculator: hourly, daily, monthly

Pick a GPU and a schedule; the estimate uses the same fixed rates the console bills — every on-demand price is the public marketplace median × 0.70, rounded down. Storage at $0.08/GB/mo and bandwidth at $0.01/GB are the only other lines.

Estimated cost

$1,037.16 / month

- GPU time $1.428 × 1 × 720 h **$1,028.16**
- Storage 100 GB × $0.08 **$8.00**
- Bandwidth 100 GB × $0.01 **$1.00**

Per hour instance **$1.428**

Per day GPU only **$34.27**

GPU-hours / month **720**

Same GPU-hours at the public marketplace median

*Market median **$1,469.81***

*PowerGPU **$1,028.16***

You keep **$441.65** (−30%)

[Deploy this configuration](https://cloud.powergpu.ai/?gpu=h100-sxm)

[Full price sheet](https://powergpu.ai/pricing)

Billed per second in practice — this estimate is an upper bound. Prices are fixed at deploy; no setup fees, no minimums.

## Four real workloads, costed

Each card is a preset at today's prices. Load one into the calculator and adjust it to your run.

### Fine-tune an 8B model with QLoRA

1× RTX 4090 On-demand 2-hour run 50 GB NVMe

**$0.79** for the run

GPU time at market median: ~~$0.94~~ (−30%)

### A working week of experiments

1× RTX 4090 On-demand 8 h/day × 5 days 50 GB NVMe 20 GB transfer

**$13.95** for the week

GPU time at market median: ~~$18.70~~ (−30%)

### A 72-hour training run

8× H100 SXM Interruptible −50% 72 hours nonstop 500 GB NVMe

**$415** for the run

GPU time at market median: ~~$1,176~~ (−65%)

### A production inference endpoint

1× L40S Reserved −35% 24/7 · 30 days 200 GB NVMe 2 TB transfer

**$276** per month

GPU time at market median: ~~$530~~ (−54%)

## How the estimate is built

Three lines, one rule. The worked example follows whatever you set in the calculator.

Worked example — 1× H100 SXM, on-demand, 24 h × 30 d

The rule behind every GPU price: [market median × 0.70, rounded down](https://powergpu.ai/pricing). This configuration: market median $2.04 × 0.70 → $1.428

- GPU timePrice × GPUs × hours. An 8× machine costs exactly 8× one GPU — no multi-GPU premium, no interconnect surcharge.
- StorageBills on allocated size while the disk exists — including while the instance is stopped. Destroy the disk and the line stops.
- BandwidthOne flat rate in and out, every region. No egress tiers, no inter-zone fees, no surprise at the end of the month.
- Billing modesInterruptible halves the GPU line for jobs that checkpoint. Reserved locks −35% for 3+ months. On-demand is guaranteed until you stop it.

## Monthly cost of popular GPUs (730 hours)

What one GPU costs if it never stops — per-second billing means any schedule costs proportionally less.

| GPU | VRAM | On-demand / mo | Interruptible / mo | Reserved / mo | Market median / mo |
| --- | --- | --- | --- | --- | --- |
| [B200](https://powergpu.ai/gpu/b200) | 192 GB | $3,960 | $1,980 | $2,574 | $5,658 |
| [H200](https://powergpu.ai/gpu/h200) | 141 GB | $2,037 | $1,018 | $1,324 | $2,911 |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1,042 | $521 | $677 | $1,490 |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1,363 | $681 | $885 | $1,948 |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $409 | $204 | $266 | $585 |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $375 | $188 | $244 | $537 |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $759 | $380 | $493 | $1,085 |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | $341 | $170 | $221 | $488 |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $320 | $160 | $208 | $459 |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $239 | $119 | $155 | $341 |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $79 | $39 | $51 | $113 |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB | $52 | $26 | $34 | $75 |

All 80 models with hourly rates: [the price sheet](https://powergpu.ai/pricing). Market medians: public marketplace pricing feed, snapshot 2026-09-14.

## GPU cost questions

Billing mechanics in depth: [billing docs](https://powergpu.ai/docs/billing).

**How is the monthly GPU cost calculated?**

GPU price per hour × number of GPUs × hours per day × days per month, plus storage (allocated GB × $0.08/GB/month, prorated) and bandwidth (GB transferred × $0.01/GB). Billing is actually per second, so the estimate is an upper bound for any schedule you enter.

**What does an H100 cost per month?**

Running non-stop (730 h), an H100 SXM costs $1,042 on-demand, $521 interruptible or $677 reserved on our sheet — versus about $1,490 at the public marketplace median.

**What does an RTX 4090 cost per hour and per month?**

$0.327/hr on-demand ($239/month at 730 h) or $0.163/hr interruptible. Eight hours a day for 20 days is $52.32 on-demand.

**Are there hidden fees the calculator does not show?**

No. GPU time, storage and bandwidth are the entire fee schedule — no setup fee, no minimum, no per-request charges, no support tiers, no multi-GPU premium. Public IPs and port mappings are included.

**Can I share or bookmark an estimate?**

Yes. Every setting is written to the page address as you change it, so the URL in your browser — or the “Copy estimate link” button — reproduces the exact configuration for a colleague.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/calculator · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/enterprise -->

---
title: "Enterprise GPU Cloud — Reserved Fleets, InfiniBand Clusters | PowerGPU"
description: "GPU fleets for teams that outgrow self-serve: reserved H100/H200/B200 capacity, InfiniBand clusters, volume discounts on public prices, 99.9% SLA, named support."
url: https://powergpu.ai/enterprise
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Enterprise

# Enterprise GPU cloud: hyperscaler scale, marketplace prices

Reserved fleets and InfiniBand clusters built from the same public price sheet as a single RTX 4090 — H100 SXM from **$0.928** /hr reserved, with volume tiers past 32 GPUs. No opaque quotes: the sheet is the quote.

### Sheet-based quotes

Every enterprise price is public sheet × published tier discount. Procurement can audit the maths in one spreadsheet cell.

### Capacity guarantees

Reserved fleets are fenced hardware with a named region and topology, contractually held for your term.

### Sub-accounts

Per-team balances, API keys and spend reporting under one master account — cost centres without spreadsheet archaeology.

### Named support

A dedicated engineer, shared incident channel and 15-minute 24/7 response on enterprise reservations.

## Fleet economics at a glance

Reserved rate × count × 730 h — what a standing fleet actually costs per month.

| Fleet | Per GPU-hr (reserved) | Monthly | Typical hyperscaler list |
| --- | --- | --- | --- |
| **32 × H100 SXM** | $0.928 | $21,678 | $106,746+ |
| **64 × H200** | $1.814 | $84,750 | $417,266+ |
| **64 × B200** | $3.526 | $164,735 | $811,059+ |

Hyperscaler column: public list prices for comparable SXM instances, September 3, 2026 — routinely 3–5× our reserved rate. Volume tiers (−5% at 32 GPUs, −10% at 128) apply on top.

## How an engagement runs

1. Ticket with your shape GPU model, count, term, region constraints, interconnect needs. From any account, in the console.
2. Sheet-derived quote Within a business day: rates, topology, delivery date, SLA riders. Nothing in it you cannot recompute.
3. Fleet live in days Standard pods in 1–3 business days, larger topologies under two weeks, with acceptance benchmarks (NCCL, storage, thermals).

Privacy posture, enterprise-grade by default

Crypto settlement, no KYC, no stored IP addresses, single-tenant hardware on reservations and full-disk encryption — the same defaults every account gets, documented on the [security page](https://powergpu.ai/security).

Compliance documentation for datacenter facilities (ISO 27001, SOC 2, Tier III) is available on request per site.

## Enterprise GPU cloud FAQ

**What counts as "enterprise" here?**

Anything past self-serve scale: 16+ GPUs sustained, multi-node clusters, custom regions, bespoke terms or invoicing needs. The pricing input never changes — the public sheet — but structure, capacity guarantees and support move to a named-engineer model.

**Do you do sales calls and custom quotes?**

A quote, yes; a discovery-call gauntlet, no. Open a ticket with your target fleet and term, and a quote built from the public sheet comes back within one business day — typically with volume tiers of −5% at 32 GPUs and −10% at 128 on top of reserved rates.

**Can we pay any other way than crypto?**

Payment remains crypto-only at every scale — USDT (TRC-20/ERC-20), BTC, ETH, XMR, LTC, TRX or SOL. For large commitments we support scheduled instalments and per-entity sub-accounts with separate balances and API keys.

**What about SLAs and support?**

The standard SLA is 99.9% monthly uptime on on-demand and reserved capacity with credit remedies (see /legal/sla). Enterprise reservations add a named engineer, a shared incident channel and a 15-minute response target, 24/7.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/enterprise · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/about -->

---
title: "About PowerGPU — Fixed-Price GPU Cloud, 30% Below Market"
description: "PowerGPU exists because GPU pricing became a casino. We buy verified datacenter capacity and resell it at one public rule: market median −30%, re-checked weekly."
url: https://powergpu.ai/about
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

About PowerGPU

# About PowerGPU: GPU pricing became a casino.
We sell the boring version.

Auctions that re-price hourly, spot instances that vanish mid-epoch, "contact sales" walls in front of a price tag. Compute is infrastructure — it should be priced like electricity, not like concert tickets.

## The model, in one paragraph

We commit to capacity in bulk — whole racks, long terms — inside established Tier-III datacenters, and resell it by the second at a mechanical rule: **the public marketplace median for each GPU, × 0.70, rounded down, re-checked weekly.** Wholesale buying is the margin; the rule is the product. No auctions to win, no quotes to negotiate, no free tier subsidised by everyone else's bill.

The rule is falsifiable on purpose. The median source is public, the snapshot date (2026-09-14) is printed on every page, and the [API](https://powergpu.ai/api) serves the whole sheet as JSON — if we ever drift above median −30%, anyone can prove it in one curl.

The platform today

- **80 GPU models** — RTX 3060 to B300
- **4,800+ GPUs** deployable right now
- **32 regions** on five continents
- **H100 SXM** at $1.428/hr vs ~~$2.04~~ median
- **99.9% SLA**, billed per second

## What we believe about selling compute

### Prices are promises

A price that changes hourly is not a price, it is a bet. Ours is fixed at deploy and published for everyone — the same number for a student and a fleet buyer.

### Verified beats crowdsourced

Marketplace economics with garage-host reliability is a bad trade. Every machine we sell lives in a vetted datacenter with redundant power, cooling and network.

### Privacy is the default

Crypto settlement, no KYC, no stored IPs, no analytics trackers. We keep what runs the service — an email, a hash, a ledger — and nothing else.

### Infrastructure, not theatre

No sales calls, no "book a demo", no gamified credits. A price sheet, an API, a console that deploys in 30 seconds. The product is the product.

## Where the machines live

We do not run a marketplace and we do not own buildings. We lease dedicated rows in existing Tier-III facilities — the same buildings the big clouds' partners use — and run our own hardware, network and provisioning stack inside them.

- Facilities hold ISO 27001 / SOC 2 attestations and Tier-III (or equivalent) design certification.
- Power-cheap regions get the flagship density — that is why the fleet skews Nordic, central-US and APAC hubs.
- Hardware is burned in for 72 h and benchmarked before it ever appears as an offer; reliability scores on offer cards come from that telemetry.

Operating capacity to sell us? Datacenters with spare GPU rows can open a [ticket](https://cloud.powergpu.ai/app/support) — we buy in rack units, not in gaming rigs.

## Numbers we hold ourselves to

- **Price floor**: ≥30% under market median, every model, verified weekly
- **Deploy time**: < 60 s for cached templates (typically ~30 s)
- **Uptime**: 99.9% monthly on on-demand/reserved — [credits if we miss](https://powergpu.ai/legal/sla)
- **Support first answer**: < 2 h median, 24/7, humans
- **Data retention**: Destroyed instance = erased disk; [the privacy policy fits on one screen](https://powergpu.ai/legal/privacy)

Live health of all of it: [status page](https://powergpu.ai/status), no login required.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/about · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/security -->

---
title: "GPU Cloud Security — Isolation, Encryption, No Stored IPs | PowerGPU"
description: "How PowerGPU protects workloads: dedicated GPUs, KVM/container isolation, encryption at rest and in transit, Tier-III datacenters, Argon2id auth, no stored IPs."
url: https://powergpu.ai/security
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Security

# GPU cloud security posture, stated precisely

No vague "bank-grade" claims: this page lists exactly how tenancy, data, network and accounts are protected — and what we deliberately do not collect, which is most of what could leak.

## Workload isolation

### Dedicated silicon

A rented GPU is passed through whole — no MIG slices, no time-sharing, no co-tenants on the die. CPU cores, RAM and disk quotas are reserved, not oversubscribed.

### Container hardening

Non-privileged containers under a hardened runtime: user namespaces, seccomp and cgroup walls, read-only host surfaces. Your image's processes see your instance and nothing else.

### VM boundary on demand

[Full VMs](https://powergpu.ai/products/vms) add KVM hardware virtualisation with VFIO passthrough — kernel-level isolation for compliance-sensitive workloads, at the same price sheet.

## Data protection

- **At rest** — instance disks and volumes encrypted (AES-256-XTS) with per-resource keys; destroy = key discard = cryptographic erasure.
- **In transit** — TLS 1.2+ on the console, API and every HTTP-exposed port; instance-to-instance traffic stays on the private fabric per region.
- **Secrets** — deploy-time env vars stored encrypted, masked in logs and UI; registry credentials likewise.
- **No sneak backups** — we never snapshot customer disks on our own initiative; what you do not snapshot does not exist twice.

## Account security

- Passwords hashed with **Argon2id**; sessions are httpOnly, SameSite, server-revocable.
- API keys stored as SHA-256 hashes, shown once, revocable in seconds — [key docs](https://powergpu.ai/docs/api-keys).
- Password change signs out every other session; bot traffic is filtered without fingerprinting humans.

## Facilities

Every machine lives in an established colocation facility — never residential hosts, never "a rig in a spare room":

- Tier-III (or equivalent) design: redundant power, cooling and carriers;
- ISO 27001 / SOC 2 attested operations, 24/7 physical access control;
- per-site compliance documentation available under NDA for [enterprise](https://powergpu.ai/enterprise) reservations.

## What we do not collect

The cheapest data breach is the one that cannot happen:

- No IP address retention — not in logs, not in the database;
- no payment PII — crypto settlement carries no card, name or address;
- no third-party trackers, no analytics scripts, no ad pixels — check the page source;
- no sale or sharing of usage data, full stop — [policy](https://powergpu.ai/legal/privacy).

## Disclosure & questions

Vulnerability reports: console ticket titled **Security** — routed straight to engineers, answered within 24 h, rewarded when it is real.

**Is my workload isolated from other customers?**

Yes — GPUs are dedicated (no MIG sharing, no time-slicing), containers run under hardened runtimes with their own namespaces and cgroups, and VMs add full KVM hardware virtualisation with VFIO GPU passthrough. Reserved capacity is single-tenant down to the chassis.

**What happens to my data when I destroy an instance?**

The disk is cryptographically erased: instance disks are encrypted at rest with per-instance keys, and destroy discards the key before the blocks are recycled. Volumes follow the same model when deleted.

**How do I report a vulnerability?**

Open a console ticket titled "Security" (any account, free) — it routes to the security engineers directly, first response inside 24 h. Good-faith research on your own resources is explicitly welcomed in our acceptable use policy; bounties are paid in crypto, naturally.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/security · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases -->

---
title: "GPU Cloud Use Cases: Training, Inference, Rendering & More | PowerGPU"
description: "GPU cloud use cases: LLM training and inference, fine-tuning, image and video generation, rendering, computer vision — each with matched GPUs and real cost math."
url: https://powergpu.ai/use-cases
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use cases

# GPU cloud use cases: right GPU, right mode, right price

Eight workload playbooks. Each one pairs the job with the best-value cards from our catalogue, the billing mode that fits its failure tolerance, and honest cost math you can rerun yourself.

- [**LLM training** Pre-train and continue-train transformer models on HBM GPUs with NVLink and InfiniBand clusters. H100 SXM from 30%+ under market Top pick: **H100 SXM** $1.428/hr](https://powergpu.ai/use-cases/llm-training)
- [**LLM inference** Serve 7B–70B models with vLLM or TensorRT-LLM at the lowest $ per million tokens. Best $/token in class Top pick: **RTX 5090** $0.439/hr](https://powergpu.ai/use-cases/llm-inference)
- [**Fine-tuning** LoRA, QLoRA and full fine-tunes — from a single 24 GB card to 8× A100 nodes. QLoRA 70B on one GPU Top pick: **A100 SXM4** $0.560/hr](https://powergpu.ai/use-cases/fine-tuning)
- [**Image generation** SDXL, Flux and ComfyUI pipelines — interactive sessions or thousand-image batch runs. Flux dev ~2 s/image on 4090 Top pick: **RTX 4090** $0.327/hr](https://powergpu.ai/use-cases/image-generation)
- [**Video generation** Wan, HunyuanVideo and image-to-video models need VRAM headroom — rent it by the second. 32–141 GB VRAM on tap Top pick: **RTX 5090** $0.439/hr](https://powergpu.ai/use-cases/video-generation)
- [**3D rendering** Blender, Octane, Redshift and V-Ray on RTX hardware — per-second billing fits render farms perfectly. Per-second render billing Top pick: **RTX PRO 6000 WS** $1.040/hr](https://powergpu.ai/use-cases/rendering)
- [**Computer vision** Train YOLO and detection models, run batch inference over image and video archives. YOLO11 epochs from $0.09 Top pick: **RTX 4090** $0.327/hr](https://powergpu.ai/use-cases/computer-vision)
- [**Scientific computing** FP64, huge memory bandwidth and MPI clusters for simulation, genomics and quantitative research. Up to 4.8 TB/s per GPU Top pick: **H200** $2.791/hr](https://powergpu.ai/use-cases/hpc)

## Not sure where your job fits?

Two shortcuts: the [VRAM sizing guide](https://powergpu.ai/guides/llm-vram-requirements) answers "which card can even run this?", and the [price sheet](https://powergpu.ai/pricing) answers "what will it cost per hour?". Everything deploys from the same [console](https://cloud.powergpu.ai/) in about 30 seconds.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/llm-training -->

---
title: "Cloud GPUs for LLM Training — H100/H200/B200 from $0.714/hr | PowerGPU"
description: "Train LLMs on HBM GPUs with NVLink and InfiniBand: H100 SXM at $1.428/hr fixed, $0.714/hr interruptible. Checkpoint patterns, cluster pods, real cost math."
url: https://powergpu.ai/use-cases/llm-training
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · LLM training

# LLM training on cloud GPUs: the same silicon, no auction

Pre-training and continued pre-training live on HBM parts: H100 SXM at **$1.428** /GPU-hr fixed (market median $2.04), NVLink in-node, InfiniBand across nodes, and interruptible capacity at half price for every checkpointed epoch.

## The training cards, ranked

Full comparison in the [H100 vs H200 vs B200 guide](https://powergpu.ai/guides/h100-vs-h200-vs-b200).

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.560 | $0.280 | Proven 80 GB HBM2e workhorse — unbeatable $/step for ≤13B experiments and LoRA-heavy labs. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| (Better) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1.428 | $0.714 | The default: FP8 transformer engine, 3.35 TB/s, deep supply in every major region. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |
| (Best) | [B200](https://powergpu.ai/gpu/b200) | 192 GB | $5.425 | $2.712 | 192 GB HBM3e and Blackwell FP4/FP8 — shortest wall-clock when the deadline is the budget. | [Deploy](https://cloud.powergpu.ai/?gpu=b200) |

## Cost of a real run, line by line

Continued pre-training, 8× H100 SXM, 72 hours, interruptible with volume checkpoints:

- **GPU time**: 8 × 72 h × $0.714 · **=**: $411.26
- **Checkpoint volume 500 GB**: 3 days × $40.00/mo · **=**: $4.00
- **Dataset ingress 400 GB**: 400 × $0.01 · **=**: $4.00
- **Total**:  · **≈**: **$419**

The market-median equivalent of the GPU line alone: about $588 at spot auctions you have to babysit — or $1,176 on-demand.

## Patterns that make it cheap

- **Checkpoint to a volume** every N steps — interruptions restart free, from disk.
- **Tune on 1×, train on 8×** — debug the config at one-eighth the burn rate.
- **Pack with FP8/BF16** — the H100/H200 transformer engine is the discount nobody uses.
- **Destroy, keep the volume** — datasets stay warm at $0.08/GB/mo, GPUs bill zero.

Templates: [PyTorch, axolotl](https://powergpu.ai/templates) (FSDP-ready) — or your own image with [a full VM](https://powergpu.ai/products/vms) for exotic stacks.

## LLM training GPUs: FAQ

Sizing math: [VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements).

**Which GPU should I train an LLM on?**

For serious runs: H100 SXM ($1.428/hr) is the price-performance default, H200 adds 141 GB for longer context and bigger micro-batches, B200 leads when wall-clock time is the constraint. For sub-13B experiments, 8× A100 nodes at $0.560/GPU-hr are hard to beat.

**On-demand or interruptible for training?**

Interruptible, almost always — training checkpoints anyway, and −50% compounds over hundreds of GPU-hours. A 72-hour run on 8× H100 costs about $411 interruptible vs $823 on-demand. Keep on-demand for the final, deadline-bound run.

**How do multi-node runs connect?**

Single machines scale to 8× with NVLink. Past that, clusters connect 8-GPU nodes over InfiniBand with NCCL pre-tuned — see the clusters page for pod pricing and all-reduce figures.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/llm-training · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/llm-inference -->

---
title: "Cloud GPUs for LLM Inference — Lowest $/Token with vLLM | PowerGPU"
description: "Serve LLMs with vLLM on fixed-price GPUs: RTX 5090 at $0.439/hr for 7B–14B, L40S for 32B, H100 for 70B. $/token math, tensor parallelism, serverless scale-to-zero."
url: https://powergpu.ai/use-cases/llm-inference
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · LLM inference

# LLM inference GPUs: tokens priced like a utility bill

Fixed GPU prices turn $/token into arithmetic: pick the smallest card that holds your model + KV-cache, saturate it with vLLM continuous batching, divide. A RTX 5090 at **$0.439** /hr serves a 8B model for pennies per million tokens.

## The serving cards, by model size

VRAM decides; price-per-VRAM ranks the candidates.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7 — the $/token king for 7B–14B FP16 and 32B 4-bit chat models. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | 48 GB and server-grade cooling for 24/7 endpoints; 32B class and long-context 14B. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| (Best) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB HBM3 for 70B quantized on one card — or FP8 for maximum throughput per dollar. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |

Bigger than 70B? Tensor-parallel across 2–8× cards on one machine — the per-GPU price never changes. Sizing tables: [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## $/million tokens, computed honestly

Throughput varies with context and batch mix, so we publish the formula, not a marketing number:

$/M tokens = GPU $/hr ÷ (tokens/s × 3600) × 1,000,000

A RTX 5090 at $0.439/hr sustaining ~2,800 tok/s of Llama-3.1-8B under continuous batching → **$0.044** /M tokens. The same card interruptible halves it.

Benchmark your own model in 10 minutes — the [serving guide](https://powergpu.ai/guides/serve-llm-vllm) ships the exact vllm bench command.

## Serving patterns

- **Reserved + vLLM** for the baseline load — −35% on the card that never sleeps.
- **Serverless burst** on top — scale-to-zero workers absorb the spikes.
- **Model library volume** mounted read-only — new workers skip the 40 GB download.
- **Quantize first** — AWQ/GPTQ 4-bit doubles the models a card can hold, rarely hurts chat quality.

Templates: [vLLM, Ollama, TGWUI](https://powergpu.ai/templates) — all OpenAI-compatible out of the box.

## LLM inference GPUs: FAQ

**What is the cheapest way to serve a 7B–14B model?**

A single RTX 5090 at $0.439/hr running vLLM. At typical chat throughput that lands well under $0.15 per million output tokens — an order of magnitude below API-provider pricing for comparable open models.

**When do I need an 80 GB card?**

Roughly at 70B: a 4-bit 70B needs ~44 GB plus KV-cache, so a single H100 PCIE (80 GB) or 2× L40S with tensor parallelism. FP16 70B wants 2× 80 GB. Below 32B, 32–48 GB cards are the value zone.

**Instances or serverless for serving?**

Steady saturating traffic → a reserved instance (−35%) you keep hot. Spiky or overnight-idle traffic → serverless with scale-to-zero. Same per-second prices; the difference is who pays for idle.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/llm-inference · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/fine-tuning -->

---
title: "Cloud GPUs for Fine-Tuning LLMs — QLoRA from $0 | PowerGPU"
description: "Fine-tuning GPUs at fixed prices: QLoRA an 8B for ~$0, 70B overnight on one 80 GB card, full-parameter on 8× A100. Live prices, axolotl template, walkthrough."
url: https://powergpu.ai/use-cases/fine-tuning
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · fine-tuning

# Fine-tune LLMs on cloud GPUs: your data, an evening's budget

Adapter methods moved fine-tuning from cluster territory to single-card territory: QLoRA a 8B model for about **$0.24** on an interruptible RTX 4090, or a 70B overnight on one 80 GB card. Per-second billing means the meter stops with the last step.

## The fine-tuning ladder

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7 — QLoRA up to ~34B on one card, twice the memory bandwidth of the 4090. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.560 | $0.280 | 80 GB HBM2e — the classic LoRA/full-FT node; 8× SXM4 scales without drama. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| (Best) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB + FP8 — full fine-tunes finish in roughly half the A100 wall-clock. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |

Budget corner: a 24 GB [RTX 4090](https://powergpu.ai/gpu/rtx-4090) at $0.163/hr interruptible handles 8B–13B QLoRA beautifully.

## What fits where (QLoRA, 4-bit base)

| Base model | VRAM needed | Cheapest single card | ≈ per training hour |
| --- | --- | --- | --- |
| Llama 3.1 **8B** | ~11 GB | [RTX 3090 24 GB](https://powergpu.ai/gpu/rtx-3090) | $0.054 |
| Qwen 2.5 **14B** | ~17 GB | [RTX 4090 24 GB](https://powergpu.ai/gpu/rtx-4090) | $0.163 |
| Qwen 2.5 **32B** | ~26 GB | [RTX 5090 32 GB](https://powergpu.ai/gpu/rtx-5090) | $0.219 |
| Llama 3.1 **70B** | ~48 GB | [H100 PCIe 80 GB](https://powergpu.ai/gpu/h100-pcie) | $0.933 |

Interruptible rates shown — fine-tunes checkpoint, so pay half. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## A clean fine-tuning loop

1. Stage data on a volume Dataset + output dir live on a $0.08/GB/mo volume, not the disposable instance.
2. One YAML, one command axolotl template: axolotl train qlora.yml — checkpoints stream to the volume.
3. Merge, serve, destroy Merge adapters, point a [vLLM instance](https://powergpu.ai/use-cases/llm-inference) at the volume, destroy the trainer. Billing: zero.

## Fine-tuning GPUs: FAQ

The full copy-paste run lives in the [QLoRA walkthrough](https://powergpu.ai/guides/fine-tune-llm-qlora).

**What does a typical fine-tune cost?**

A QLoRA pass over 10k instruction pairs on Llama-3.1-8B takes ~1.5 h on one RTX 4090 — about $0.24 interruptible. A 70B QLoRA on one 80 GB card is an overnight run in the tens of dollars. Full-parameter 8B on 8× A100: low hundreds.

**LoRA, QLoRA or full fine-tune?**

QLoRA first: 4-bit base + trainable adapters fits 8B in under 12 GB and 70B in ~48 GB, and quality is usually within a point of full tuning for instruction tasks. Go full-parameter only when the domain shift is large — then rent A100/H100 nodes for hours, not weeks.

**Which template should I start from?**

axolotl — one YAML covers LoRA/QLoRA/full, DeepSpeed and FSDP, and the walkthrough guide is written against it. Kohya covers the image-LoRA side.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/fine-tuning · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/image-generation -->

---
title: "Stable Diffusion & Flux Cloud GPUs — ComfyUI from $0.327/hr | PowerGPU"
description: "Run ComfyUI, Flux and SDXL on fixed-price GPUs: RTX 4090 at $0.327/hr ≈ 5,004 Flux images per dollar. One-click templates, LoRA training, batch pipelines."
url: https://powergpu.ai/use-cases/image-generation
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · image generation

# Image generation GPUs: 5,004 images per dollar

ComfyUI and Forge boot in ~30 seconds on cards the community actually tunes for: RTX 4090 at **$0.327** /hr. Generate interactively, batch overnight on interruptible at half price, and stop paying the second the queue empties.

## The image cards, ranked by images-per-dollar

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | 24 GB at the lowest price that runs everything — the batch-farm workhorse. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | The community default: fastest ecosystem support, ~2.2 s Flux dev images. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7 headroom for full-precision Flux, video nodes and monster upscale chains. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Serving images to users? The 48 GB [L40S](https://powergpu.ai/gpu/l40s) on a [serverless endpoint](https://powergpu.ai/products/serverless) handles concurrent workflows with server-grade cooling.

## Interactive vs batch: two bills

- **An evening of prompting**: 3 h interactive ComfyUI on a 4090, on-demand · $0.98
- **A 10,000-image batch**: ~6.1 h on a 4090, interruptible queue · $0.99
- **A style LoRA**: ~45 min Kohya training run · $0.12
- **Model library, always warm**: 120 GB volume (checkpoints, LoRAs, VAEs) · $9.60/mo

## Set up once, generate forever

- **ComfyUI template** ships with Manager — install nodes from the browser.
- **Models on a volume** — 40 GB of checkpoints load in seconds on every new instance.
- **Forge for A1111 muscle memory** — same extensions, lower VRAM, faster attention.
- **API mode** — every ComfyUI workflow exports as JSON and runs headless for pipelines.

Moving pictures? The [video generation playbook](https://powergpu.ai/use-cases/video-generation) continues from here.

## Stable Diffusion & Flux GPUs: FAQ

**What does an AI image actually cost to generate?**

Flux dev at ~2.2 s/image on an RTX 4090 works out to about $0.00020 per image on-demand — roughly 5,004 images per dollar. Batch on interruptible and it doubles. SDXL is 2–3× faster still.

**Which GPU for ComfyUI with Flux?**

Flux dev FP8 wants ~17 GB, so 24 GB cards are the entry point: the RTX 4090 at $0.327/hr is the community default. Full-precision Flux, heavy upscaler chains or video nodes benefit from the 32 GB RTX 5090.

**How do I train a style LoRA?**

The Kohya template: 20–40 images, captions, ~30–60 minutes on a 4090 — a couple of dollars end to end. Keep datasets and outputs on a volume so trainer instances stay disposable.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/image-generation · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/video-generation -->

---
title: "Cloud GPUs for AI Video Generation — Wan, HunyuanVideo, LTX | PowerGPU"
description: "Rent VRAM-heavy GPUs for AI video (Wan, HunyuanVideo, LTX): RTX 5090 32 GB from $0.439/hr, H100 80 GB, H200 141 GB. Per-clip cost math, ComfyUI workflows."
url: https://powergpu.ai/use-cases/video-generation
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · video generation

# Video generation GPUs: VRAM for video, rented by the clip

Video diffusion is the most VRAM-hungry workload in the catalogue — and the best case for renting: generate on an 80 GB H100 PCIE at **$1.867** /hr only while the queue runs, then destroy it. Nobody buys a $25k card for weekend clips.

## The video cards, by headroom

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7 — Wan 14B at 720p with offloading; the affordable way in. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB HBM3 — full-quality Wan/Hunyuan without offload gymnastics, 2–3× faster clips. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |
| (Best) | [H200](https://powergpu.ai/gpu/h200) | 141 GB | $2.791 | $1.395 | 141 GB — long clips, high resolutions and batch generation without a single OOM. | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |

Middle path: the 48 GB [L40S](https://powergpu.ai/gpu/l40s) at $0.514/hr runs most 720p pipelines at full precision.

## A batch night, budgeted

60 five-second clips for a storyboard, overnight on interruptible:

- **GPU time**: ~6 min/clip × 60 on H100 PCIE interruptible · $5.60
- **Model volume**: 200 GB (Wan + Hunyuan + LoRAs), one month · $16.00
- **Download the results**: ~3 GB out · $0.03
- **Total for the night**:  · **≈ $21.63**

Interruptions cost nothing here: each clip is an independent queue item that re-runs from the volume.

## Make it not hurt

- **Models on a volume** — video checkpoints are 20–60 GB; download once, mount forever.
- **Prototype at 480p on the 5090**, final-render at 720p+ on 80 GB — same workflow JSON.
- **Queue as items, not marathons** — per-clip jobs love interruptible pricing.
- **Watch VRAM, not GPU%** — video pipelines OOM before they saturate compute.

## AI video GPUs: FAQ

**How much VRAM do video models need?**

More than image models by an order of magnitude of activations: Wan 2.1 14B wants 24–32 GB for 720p clips with offloading, comfortable at 48 GB; HunyuanVideo prefers 48–80 GB. The 32 GB RTX 5090 is the realistic entry point, 80 GB cards the comfortable one.

**What does a clip cost to generate?**

A 5-second 720p Wan 2.1 clip takes roughly 4–8 minutes on an H100 PCIE — about $0.19 at $1.867/hr. Batch overnight on interruptible and the per-clip cost halves.

**Which template do I start from?**

ComfyUI — current video models (Wan, Hunyuan, LTX, image-to-video pipelines) all ship ComfyUI workflows first. Put models on a volume: video checkpoints are 20–60 GB each.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/video-generation · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/rendering -->

---
title: "Cloud GPU Rendering for Blender, Octane & Redshift | PowerGPU"
description: "Cloud GPU rendering billed by the second: RTX 4090 frames from $0.010, 96 GB RTX PRO 6000 for production scenes. Headless Blender, per-frame queues, no farm markup."
url: https://powergpu.ai/use-cases/rendering
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · 3D rendering

# GPU cloud rendering: a render farm that bills like a stopwatch

Classic farms charge per GHz-hour with queue fees and priority tiers. Here it is one number: RTX 4090 at **$0.327** /hr — about $0.010 per 3.5-minute Cycles frame on interruptible — with OptiX, your exact Blender version, and zero markup on idle.

## The render cards, ranked

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | OptiX monster for scenes under 24 GB — the best price per sample in the catalogue. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | $0.467 | $0.233 | 48 GB ECC for production scenes, studio drivers, server-grade thermals. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-6000ada) |
| (Best) | [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | 96 GB GDDR7 — city-scale scenes, volumes and 8K texture sets fully in memory. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |

## A 720-frame shot, quoted

- **Frames**: 720 × ~3.5 min on RTX 4090, interruptible queue · $6.85
- **Parallelised**: Same queue across 8 single-GPU instances · ~5.3 h wall-clock, same cost
- **Scene volume 80 GB**: assets + EXR output, one month · $6.40
- **Download EXRs 40 GB**: 40 × $0.01 · $0.40

Per-second billing means parallelising is free: 8 instances for 1 hour cost what 1 instance costs for 8.

## Farm patterns

- **Frames as queue items** — blender -b scene.blend -f N per item; interruptions re-render one frame, not the shot.
- **Assets on a read-only volume** — 8 workers mount one scene library.
- **Interactive lookdev on a VM** — [VNC desktop](https://powergpu.ai/products/vms) on a workstation card, then switch to headless for the farm.
- **Pin your Blender version** — templates are tagged; renders are reproducible.

## GPU cloud rendering: FAQ

Full setup in the [Blender cloud guide](https://powergpu.ai/guides/blender-cloud-rendering).

**Which renderers work on rented GPUs?**

Anything CUDA/OptiX: Blender Cycles, Octane, Redshift, V-Ray GPU, Arnold GPU. The Blender template renders.blend files headless out of the box; Octane/Redshift run inside a full VM with your own licences and a VNC desktop if you want one.

**What does a frame cost?**

A Cycles frame that takes 3.5 minutes on an RTX 4090 costs about $0.010 interruptible. A 720-frame shot: ~$7 — and per-frame queue items make interruptions free re-runs.

**Why a workstation card instead of a 4090?**

VRAM and ECC: heavy production scenes (fur, volumes, 8K textures) blow past 24 GB. The RTX PRO 6000 WS (96 GB) and RTX 6000Ada (48 GB) hold entire scenes in memory with studio drivers — still far below hyperscaler workstation pricing.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/rendering · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/computer-vision -->

---
title: "Computer Vision Cloud GPUs — YOLO Training from $0.02/epoch | PowerGPU"
description: "Train YOLO, detection and segmentation models or sweep image archives: RTX 4090 at $0.327/hr, budget cards from $0.103/hr. NVDEC pipelines, per-second billing."
url: https://powergpu.ai/use-cases/computer-vision
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · computer vision

# Computer vision GPUs: detection budgets measured in epochs

Vision workloads are bursty by nature — a training sprint, a giant batch job, then nothing. Per-second billing fits perfectly: a YOLO11 epoch for about **$0.02**, an archive sweep for the price of coffee, zero idle spend between sprints.

## The vision cards, ranked

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | 24 GB at rock-bottom pricing — big batches for augmentation-heavy training. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Better) | [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | $0.161 | $0.080 | ECC + blower cooling for week-long training queues; the reliability pick. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a5000) |
| (Best) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Fastest epochs per dollar — the default for YOLO, segmentation and ViT fine-tunes. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |

Batch-inference fleets: [Tesla T4](https://powergpu.ai/gpu/tesla-t4) from $0.103/hr — TensorRT detectors barely notice the smaller card.

## Two typical jobs, costed

- **Train YOLO11m, 100 epochs**: 50k images @ 640px, RTX 4090 interruptible · ≈ $1.96
- **Sweep 1M archived photos**: TensorRT detector, 4 × Tesla T4 in parallel · ≈ $0.18
- **Dataset volume 150 GB**: images + labels + runs, one month · $12.00

## Pipelines that scale down to zero

- **Ultralytics/MMDetection in the PyTorch template** — pip install and train.
- **Export to TensorRT** before batch runs — 3–5× throughput on the same card.
- **GPU video decode** — NVDEC keeps CPU out of the hot path for camera streams.
- **Weights on a volume**, instances disposable — retrain Fridays, pay Fridays only.

## Computer vision GPUs: FAQ

**Which GPU for training YOLO models?**

A RTX 4090 trains YOLO11m on a 50k-image dataset at roughly 7 minutes per epoch — about $0.02/epoch interruptible. 24 GB fits big batch sizes at 640px; step up to multi-GPU only past a few hundred thousand images.

**What about batch inference over an archive?**

Cheap cards shine: a Tesla T4 at $0.103/hr pushes hundreds of frames/s with a TensorRT-exported detector. A million images costs a few dollars — spread the queue over several instances and it finishes over lunch.

**Do you support video pipelines?**

Yes — NVDEC/NVENC are exposed in containers and VMs, so decode → detect → encode runs entirely on GPU (DeepStream, PyAV, ffmpeg builds in the PyTorch template).

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/computer-vision · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/use-cases/hpc -->

---
title: "HPC Cloud GPUs — FP64, HBM Bandwidth, MPI Clusters | PowerGPU"
description: "HPC on rented GPUs: H200 with 4.8 TB/s HBM3e at $2.791/hr, A100/H100 FP64 tensor cores, InfiniBand MPI clusters. Simulation, genomics, quant research."
url: https://powergpu.ai/use-cases/hpc
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Use case · scientific computing

# Scientific computing GPUs: cluster-grade compute, no allocation committee

CFD, molecular dynamics, genomics, quant backtests: rent the same Hopper silicon national labs queue for — H200 with 4.8 TB/s of HBM3e at **$2.791** /hr, by the second, starting now instead of next quarter.

## The HPC cards, ranked

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.560 | $0.280 | FP64 tensor cores at the lowest HBM price — the budget line for double precision. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| (Better) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1.428 | $0.714 | ~34 TFLOPS FP64, 3.35 TB/s — the general-purpose simulation default. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |
| (Best) | [H200](https://powergpu.ai/gpu/h200) | 141 GB | $2.791 | $1.395 | Same compute, 141 GB and 4.8 TB/s — memory-bound codes see the biggest jump. | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |

Legacy budget: [Tesla V100](https://powergpu.ai/gpu/tesla-v100) at $0.130/hr still does honest FP64 for coursework and small models.

## What runs here today

- **Molecular dynamics** — GROMACS, AMBER, LAMMPS with CUDA builds in the CUDA-dev template.
- **CFD & FEA** — AmgX-accelerated solvers, lattice-Boltzmann codes, in full VMs when licences demand them.
- **Genomics** — Parabricks-style pipelines: a 30× WGS alignment in under an hour on one H100.
- **Quant research** — Monte-Carlo sweeps and backtests that parallelise embarrassingly well on interruptible.

Licensing note: bring-your-own-licence software runs in [VMs](https://powergpu.ai/products/vms) with hardware dongle passthrough available on reserved machines.

## A simulation week, budgeted

- **Parameter sweep**: 40 runs × 3 h on A100 SXM4, interruptible · $33.60
- **Hero run**: 36 h on 8× H100 SXM, on-demand · $411
- **Result volume 1 TB**: trajectories + fields, one month · $80

The sweep on interruptible costs less than the storage — that is the shape per-second pricing gives research budgets.

## HPC GPUs: FAQ

**Which GPUs have real FP64 performance?**

The datacenter line: H100/H200 (~34 TFLOPS FP64 tensor), A100 (~19.5 with tensor cores), V100 (~7.8). Consumer RTX cards throttle FP64 to 1/64th of FP32 — fine for ML, wrong for double-precision simulation. If your solver is FP64-bound, rent Hopper or Ampere datacenter parts.

**Can I run MPI jobs across machines?**

Yes — clusters connect 8× SXM nodes over InfiniBand with CUDA-aware MPI and NCCL/UCX preinstalled on request. Single-node 8× machines cover a surprising share of mid-size simulations by themselves.

**Why H200 for memory-bound codes?**

Bandwidth: 4.8 TB/s of HBM3e per GPU — stencil kernels, CFD and lattice codes that starve on GDDR parts often see near-linear speedups just from feeding the cores. 141 GB also keeps bigger domains resident.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/use-cases/hpc · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/products/on-demand -->

---
title: "On-Demand Cloud GPUs — Fixed Price, Guaranteed Capacity | PowerGPU"
description: "Guaranteed GPU instances at fixed prices ≥30% below market median: H100 at $1.428/hr, RTX 4090 at $0.327/hr. Per-second billing, deploy in 30s, stop anytime."
url: https://powergpu.ai/products/on-demand
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Products · on-demand

# On-demand cloud GPUs.
Fixed price, guaranteed resources

The default way to rent a GPU: deploy in ~30 seconds, keep it as long as you want, pay per second at a price that was fixed ≥30% under the market median before you arrived. No auctions to win, no evictions to survive.

### Guaranteed capacity

Once running, the instance is yours until you stop it. On-demand workloads are never preempted — that is what interruptible is for, at half price.

### Price fixed at deploy

Weekly market re-checks only change prices for new deploys. A training run started today finishes at today's rate.

### Per-second, no minimums

A 90-second experiment costs 90 seconds. No hourly rounding, no daily minimum, no idle fees when stopped (storage only).

### 1× to 8× per machine

Scale within a chassis at the same per-GPU price — NVLink on SXM hardware, no multi-GPU premium anywhere.

## Today's on-demand rates, spot-checked

Full sheet on the [pricing page](https://powergpu.ai/pricing) — here is the shape of it.

| GPU | VRAM | Market median | On-demand | Est. / month |  |
| --- | --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $2.04 | $1.428 (−30%) | $1,042 | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |
| [H200](https://powergpu.ai/gpu/h200) | 141 GB | $3.99 | $2.791 (−30%) | $2,037 | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.80 | $0.560 (−30%) | $409 | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.74 | $0.514 (−30%) | $375 | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.63 | $0.439 (−30%) | $320 | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.47 | $0.327 (−30%) | $239 | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |

## Built for stateful, interactive work

On-demand is the right mode when losing the machine mid-job costs more than the discount you would get on interruptible:

- **Serving & APIs** — vLLM endpoints, ComfyUI backends, anything with users on it.
- **Interactive sessions** — Jupyter, SSH development, dataset exploration.
- **Un-checkpointed jobs** — long renders, one-shot ETL, anything you cannot resume.
- **Deadline work** — when "it re-queued overnight" is not an acceptable answer.

Fault-tolerant batch jobs should look at [interruptible (−50%)](https://powergpu.ai/products/interruptible); permanent workloads at [reserved (−35%)](https://powergpu.ai/products/reserved).

*deploy — on-demand*

```
$ powergpu launch --gpu h100-sxm --template vllm \
    --env MODEL=meta-llama/Llama-3.1-8B-Instruct
✓ capacity reserved on m-1d29c04a (eu-central-1)
✓ instance i-52ab77c1 running (24.1s)
# endpoint: https://i-52ab77c1.powergpu.ai:8000/v1
# billing: $1.428/hr · per second · stop anytime
```

## On-demand GPUs: FAQ

More detail in the [instance lifecycle docs](https://powergpu.ai/docs/instances).

**What does "on-demand" guarantee exactly?**

The GPU is yours from start to stop: no preemption, no eviction, no bidding. The price you deploy at is contractually your price for the life of the instance, and capacity is reserved to your account the moment the deploy succeeds.

**How fast is a deploy really?**

Container templates cold-start in about 30 seconds — image pull is the variable (popular templates are pre-cached on hosts). Full VMs take 2–4 minutes. You can watch the state machine live in the console.

**Can I stop an instance and keep my data?**

Yes. Stop ends GPU billing that second; the disk stays, billed at $0.08/GB/mo, and the instance restarts on the same data. Destroy ends everything, including storage billing.

**Is on-demand available on every GPU model?**

All 80 models in the catalogue, from flagship B200s to GTX-class cards — same rules, same per-second billing, 1× to 8× per machine.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/products/on-demand · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/products/interruptible -->

---
title: "Spot GPU Instances Without Auctions — Flat 50% Off | PowerGPU"
description: "Interruptible (spot) GPU instances at a flat 50% off, no bidding: H100 at $0.714/hr, RTX 4090 at $0.163/hr. Auto-requeue, disk kept, per-second billing."
url: https://powergpu.ai/products/interruptible
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Products · interruptible

# Interruptible (spot) GPUs: half price, zero auctions

Interruptible instances cost a flat 50% of on-demand — H100 SXM at **$0.714** /hr instead of $1.428. No bids to place, no eviction markets to monitor: your job may pause under capacity pressure, your data never disappears.

### Flat −50%, always

Not "up to". Not "from". Every model, every region: exactly half the on-demand price, fixed like everything else here.

### Disk survives

Interruption = stop, not destroy. Checkpoints, datasets and outputs stay on the instance disk, billing drops to storage-only.

### Auto-requeue

Interrupted instances rejoin the capacity queue automatically and restart from disk — overnight queues drain themselves.

### No babysitting

Nothing to outbid at 2am. The console shows interruption events; webhooks tell your pipeline. That is the whole operational surface.

## The math on real jobs

Same silicon as on-demand — the only difference is the eviction clause.

| GPU | On-demand | Interruptible | 100 GPU-hrs cost | You save |  |
| --- | --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | $1.428 | $0.714 | $71.40 | −$71.40 | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm&type=spot) |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | $0.560 | $0.280 | $28.00 | −$28.00 | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4&type=spot) |
| [L40S](https://powergpu.ai/gpu/l40s) | $0.514 | $0.257 | $25.70 | −$25.70 | [Deploy](https://cloud.powergpu.ai/?gpu=l40s&type=spot) |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | $0.439 | $0.219 | $21.90 | −$22.00 | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090&type=spot) |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | $0.327 | $0.163 | $16.30 | −$16.40 | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090&type=spot) |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | $0.108 | $0.054 | $5.40 | −$5.40 | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090&type=spot) |

Versus the marketplaces: their spot is an auction — win it, and a higher bid can still evict you an hour later. Ours is a queue.

## Made for fault-tolerant work

- **Checkpointed training** — resume from the last step; pay half for every step.
- **Batch queues** — image generation, transcription, embedding jobs, render frames.
- **Hyperparameter sweeps** — dozens of short runs, none of them precious.
- **CI & evaluation** — nightly benchmark suites that just re-run if paused.

If an interruption would hurt, use [on-demand](https://powergpu.ai/products/on-demand) — or mix: serve on-demand, train interruptible, one balance pays both.

*train — interruptible*

```
$ powergpu launch --gpu a100-sxm4 --type interruptible \
    --template axolotl --volume ckpts:50GB
✓ instance i-b3e19d02 running · $0.280/hr
# 03:12 interrupted — checkpoint step 4200 on volume
# 03:57 capacity free — auto-restarted, resumed 4200
✓ run finished · 9.4 GPU-hrs · $2.63
```

## Interruptible GPUs: FAQ

Checkpoint patterns live in the [instance docs](https://powergpu.ai/docs/instances).

**How often do interruptions actually happen?**

Only when an on-demand deploy needs your slot — there is no bidding war constantly re-pricing you out. Fleet-wide, interruptible instances run a median of many hours between interruptions; deep-supply consumer cards (RTX 30/40 series) are interrupted least.

**What happens when my instance is interrupted?**

It is stopped, not destroyed: the container/VM state on disk survives and GPU billing ends instantly. You get a console event and an optional webhook, and the instance auto-requeues for capacity unless you disable that. When capacity frees, it restarts from your disk.

**Is the discount really a flat 50%?**

Yes — interruptible = on-demand × 0.50 on every model, every region, all the time. It is not a fluctuating spot market: the price is as fixed as our on-demand price, just half of it.

**How should I design a job for interruptible?**

Checkpoint to the mounted volume (every N steps for training, per-item for batch queues), make startup idempotent, and let auto-requeue do the rest. The docs include ready patterns for PyTorch, axolotl and render farms.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/products/interruptible · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/products/reserved -->

---
title: "Reserved GPU Instances — Commit 3 Months, Save 35% More | PowerGPU"
description: "Reserved GPU capacity at 35% off on-demand: H100 SXM at $0.928/hr locked for the term, capacity guaranteed. 3, 6 or 12-month commitments, billed monthly."
url: https://powergpu.ai/products/reserved
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Products · reserved

# Reserved GPU capacity: −35% and a held seat

For workloads that run every day, reservations stack a further 35% discount on top of prices already 30% under market — H100 SXM drops to **$0.928** /hr, locked for the term, with capacity guaranteed even when a region runs hot.

### Deepest rate

On-demand × 0.65 — roughly 55% under the market median once the 30% floor is stacked in. The cheapest legitimate way to run a permanent GPU.

### Capacity held

Reserved capacity is fenced from the on-demand pool. Your deploys never queue, even during launch-week GPU droughts.

### Rate locked

Market up or down, your price is signed for the term. Budget a quarter to the dollar.

### Billed monthly

No giant upfront invoice: the commitment draws from your crypto-funded balance month by month.

## Reserved rates on the big cards

The 3-month figure is the full commitment cost per GPU — compare it with four weeks of your current invoice.

| GPU | On-demand | Reserved | Per month (730 h) | 3-month commit |  |
| --- | --- | --- | --- | --- | --- |
| [B200](https://powergpu.ai/gpu/b200) | $5.425 | $3.526 (−35%) | $2,574 | $7,722 | [Reserve](https://cloud.powergpu.ai/app/support) |
| [H200](https://powergpu.ai/gpu/h200) | $2.791 | $1.814 (−35%) | $1,324 | $3,973 | [Reserve](https://cloud.powergpu.ai/app/support) |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | $1.428 | $0.928 (−35%) | $677 | $2,032 | [Reserve](https://cloud.powergpu.ai/app/support) |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | $0.560 | $0.364 (−35%) | $266 | $797 | [Reserve](https://cloud.powergpu.ai/app/support) |
| [L40S](https://powergpu.ai/gpu/l40s) | $0.514 | $0.334 (−35%) | $244 | $731 | [Reserve](https://cloud.powergpu.ai/app/support) |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | $0.439 | $0.285 (−35%) | $208 | $624 | [Reserve](https://cloud.powergpu.ai/app/support) |

6-month terms take a further −5%, 12-month −10% — quoted per request from the console.

## When reserving beats everything

- **Production inference** — endpoints with steady traffic 24/7.
- **Training programmes** — research groups with a standing compute budget.
- **Render pipelines** — studios with a predictable nightly queue.
- **Migration off hyperscalers** — an H100 reserved here costs less than a quarter of typical list price.

Rule of thumb: if your utilisation is above ~65%, reserved beats on-demand; below that, stay [on-demand](https://powergpu.ai/products/on-demand) or go [interruptible](https://powergpu.ai/products/interruptible).

Need more than 16 GPUs?

Fleet reservations, InfiniBand clusters, custom regions and invoicing conversations live on the [enterprise page](https://powergpu.ai/enterprise). Everything is still crypto-settled and still priced from the public sheet.

Median time from ticket to running fleet: 3 business days.

## Reserved GPUs: FAQ

Terms live in the [terms of service](https://powergpu.ai/legal/terms); capacity questions in a [console ticket](https://cloud.powergpu.ai/app/support).

**How does a reservation work?**

You commit to a GPU model and quantity for 3, 6 or 12 months and pay the reserved rate (−35% off on-demand) for the committed hours, drawn from your credit balance month by month — not all upfront. The capacity is held for you: deploys against a reservation never wait.

**What if I stop using it halfway through?**

The commitment bills for its term whether instances run or not — that is what buys the discount and the held capacity. You can upgrade a reservation to a bigger one at any time; we credit the remainder.

**Does the reserved price ever change mid-term?**

No. The rate is locked at signature for the entire term, even if market prices — and therefore our on-demand prices — move. Renewal re-prices at the then-current sheet.

**Can I reserve multi-node clusters?**

Yes — reservations are how most InfiniBand cluster capacity is sold. See the clusters page or open a ticket from the console with your target topology; provisioning takes 1–3 days.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/products/reserved · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/products/serverless -->

---
title: "Serverless GPU Endpoints — Autoscale, Scale to Zero | PowerGPU"
description: "Autoscaling serverless GPU endpoints for vLLM, ComfyUI or any container: scale to zero, cached weights, per-second billing at fixed prices ≥30% below market."
url: https://powergpu.ai/products/serverless
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Products · serverless

# Serverless GPU endpoints that scale to zero

Ship a model, get a URL, pay per second of actual GPU work. Workers scale with queue depth — from zero to dozens and back — at the same fixed prices as everything else: an L40S worker-second costs exactly $0.514/3600.

### Queue-based autoscale

Workers spawn when requests-per-worker crosses your threshold and drain when the queue empties. You set min/max; the default min is zero.

### Cached weights

Model files persist on regional NVMe cache, so scale-up loads from local disk — seconds, not a 40 GB download.

### Standard interfaces

vLLM endpoints speak the OpenAI API; anything else is plain HTTP on your container's port, behind TLS we terminate for you.

### Per-second economics

Active seconds × fixed price. A burst of 10,000 requests on RTX 5090 workers costs the same whether it takes one worker an hour or twelve workers five minutes.

## From weights to URL in one command

Endpoints are described declaratively — model, GPU class, scaling window — and managed like any other resource from the console, CLI or API.

- **LLM chat & completion** — vLLM template, OpenAI-compatible, streaming included.
- **Image generation** — ComfyUI workflows exposed as a JSON API.
- **Transcription, embeddings, rerankers** — Whisper and friends on cheap cards.
- **Your container** — anything that serves HTTP; we handle TLS, scaling and health checks.

*serverless — deploy*

```
$ powergpu endpoint create --name llama8b \
    --template vllm --gpu l40s --min 0 --max 8 \
    --env MODEL=meta-llama/Llama-3.1-8B-Instruct
✓ endpoint ep-31c8a2 ready
# https://llama8b-31c8a2.powergpu.ai/v1/chat/completions
$ powergpu endpoint stats llama8b
# workers 0→3 (last hour) · p50 342ms · 41,208 req · $1.87
```

## Which GPU class for which model

Serverless supports the same catalogue; these three cover most endpoints.

| Worker GPU | VRAM | Good for | Active price | 1M tokens ≈ |
| --- | --- | --- | --- | --- |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | 7B–14B chat, embeddings, Whisper | $0.439/hr | $0.05–0.15 |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | 14B–32B models, SDXL/Flux, long context | $0.514/hr | $0.10–0.30 |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | 70B quantized, video models, big batches | $0.560/hr | $0.25–0.60 |

Token costs are illustrative vLLM throughput ranges; benchmark your model with the [serving guide](https://powergpu.ai/guides/serve-llm-vllm).

## Serverless GPUs: FAQ

Endpoint reference in the [docs](https://powergpu.ai/docs) and [API](https://powergpu.ai/api).

**How is serverless billed?**

Per second of active GPU time at the same fixed per-model prices as instances, plus the flat storage rate for your model cache. Scale-to-zero means an idle endpoint costs only its cached weights — a 16 GB model cache is about $1.28/month.

**What are cold starts like?**

Weights stay cached on NVMe next to the GPUs, so cold starts are load-to-VRAM, not download-from-internet: typically 4–15 s for 7B–13B class models, longer for 70B. Set min-workers to 1 to remove them entirely — you pay for that worker only while it exists.

**What can I deploy on it?**

The vLLM and ComfyUI templates work out of the box (OpenAI-compatible and REST respectively), or bring any Docker image that answers HTTP on a port. Autoscaling watches queue depth per worker.

**When should I use instances instead?**

Steady, saturating traffic is cheaper on a reserved instance you keep busy. Serverless wins for spiky traffic, many small models, or products that sleep at night — pay for requests, not for idle.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/products/serverless · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/products/clusters -->

---
title: "GPU Clusters — Multi-Node H100/H200/B200 with InfiniBand | PowerGPU"
description: "Multi-node GPU clusters with NVLink and InfiniBand: 16 to 512× H100, H200 or B200 at fixed per-GPU prices ≥30% below market. NCCL-tuned, Slurm-ready, live in days."
url: https://powergpu.ai/products/clusters
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Products · clusters

# GPU clusters for distributed training, priced like the sheet

Multi-node pods of 8× SXM machines stitched with InfiniBand — 16 to 512 GPUs — at the same public per-GPU prices as single instances. A 64× H100 SXM pod at reserved rates: **$59.39** /hour, fabric included.

### Real fabric

NVLink/NVSwitch in-node, InfiniBand rail-optimised between nodes, NCCL pre-tuned. All-reduce numbers ship with the quote.

### No cluster premium

Per-GPU price × GPU count. Fabric, head nodes and scratch storage are included — compare that line with any hyperscaler quote.

### Your scheduler

Slurm installed and configured, or bare SSH and your own orchestration. Root on every node either way.

### Days, not quarters

Standard pods provision in 1–3 business days. No committee, no "capacity review call" — a ticket and a quote.

## Standard pod configurations

Built from 8× SXM nodes; any multiple works. Reserved rates shown — burst pods at on-demand when capacity allows.

| Pod | GPUs | VRAM total | Interconnect | $/hour (reserved) |
| --- | --- | --- | --- | --- |
| **2 × 8× H100 SXM** | 16 | 1.3 TB | NVLink + IB HDR | $14.85 |
| **8 × 8× H100 SXM** | 64 | 5.1 TB | NVLink + IB NDR | $59.39 |
| **8 × 8× H200** | 64 | 9.0 TB | NVLink + IB NDR | $116.10 |
| **8 × 8× B200** | 64 | 12.3 TB | NVLink 5 + IB NDR | $225.66 |
| **64 × 8× H100 SXM** | 512 | 41 TB | Rail-optimised NDR | $475 |

Per-GPU maths: reserved rate ([public sheet](https://powergpu.ai/pricing)) × GPU count. Nothing else is added.

## What teams run on pods

- **Pre-training & continued pre-training** — FSDP/DeepSpeed/Megatron across nodes.
- **Large fine-tunes** — 70B+ full-parameter runs that outgrow one machine.
- **RLHF / RLVR pipelines** — actor, critic and reward fleets on one fabric.
- **Batch scale-outs** — thousand-GPU-hour render or simulation campaigns.

Single-node 8× machines (no fabric needed) deploy instantly from the [console](https://cloud.powergpu.ai/) like any instance — clusters are for when one chassis stops being enough.

*slurm — 64× h100*

```
$ sinfo
PARTITION AVAIL NODES STATE  NODELIST
gpu*      up        8  idle   pgx-[01-08]
$ sbatch --nodes 8 --gpus-per-node 8 train.sh
✓ Submitted batch job 1129
# NCCL all-reduce 256MB: 47 GB/s bus BW (rail-optimised NDR)
```

## GPU clusters: FAQ

Start with a [console ticket](https://cloud.powergpu.ai/app/support): model, GPU count, region preference, term. A quote with topology comes back within a business day.

**What interconnect do clusters use?**

Inside a node: NVLink/NVSwitch on SXM platforms. Between nodes: InfiniBand NDR/HDR (400/200 Gb/s per port, rail-optimised) on flagship pods; 100 GbE RoCE on smaller A100/L40S pods. Topology details ship with every quote.

**How fast can I get a cluster?**

Standard blocks (2–8 nodes of 8× H100/H200) are usually provisioned in 1–3 business days from a console ticket. Larger or unusual topologies are quoted with a delivery date — typically under two weeks.

**How are clusters priced?**

Per GPU-hour from the same public sheet — reserved rates for committed terms, on-demand for burst blocks when available. A 64× H100 SXM pod at reserved rates runs $59/hour all-in; InfiniBand fabric and management nodes are included, not line items.

**What software do they come with?**

Ubuntu with the NVIDIA driver stack, CUDA, NCCL tuned for the fabric, Slurm or plain SSH per your preference, and shared NVMe scratch. Bring your own images — clusters are yours at root.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/products/clusters · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/products/vms -->

---
title: "GPU Virtual Machines (GPU VPS) — Full Root, Your Kernel | PowerGPU"
description: "Full KVM virtual machines with dedicated GPUs: root access, custom kernels, GUI desktops. RTX 4090 VMs from $0.327/hr, per-second billing, same fixed price sheet."
url: https://powergpu.ai/products/vms
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Products · virtual machines

# GPU virtual machines: a whole machine, not just a container

Full KVM virtual machines with the GPU passed straight through: your kernel, your drivers, systemd, reboots, even a desktop. Same price sheet as containers — an RTX 4090 VM is still **$0.327** /hr, billed per second.

### True isolation

Hardware virtualisation with GPU passthrough (VFIO) — your workload never shares a kernel with anyone.

### Root, kernel, all of it

Load kernel modules, run nested Docker, patch drivers, mount FUSE — everything a bare server allows.

### Desktop-capable

Optional XFCE + VNC/RDP for interactive creative work on workstation GPUs like the RTX PRO 6000 WS.

### Snapshots & images

Snapshot a configured VM into a private image and stamp out copies — golden images for teams.

## Container or VM? Thirty-second decision

|  | Container instance | Virtual machine |
| --- | --- | --- |
| Cold start | ~30 seconds | 2–4 minutes |
| Kernel control | Host kernel (pinned NVIDIA stack) | Yours — any kernel, any module |
| Nested Docker | Via template flag | Native |
| GUI desktop | App-level (ComfyUI, Jupyter) | Full XFCE via VNC/RDP |
| Price | Same sheet | Same sheet |

Rule: start with a container; switch to a VM the first time the container says no.

*vm — launch*

```
$ powergpu launch --gpu rtx-4090 --vm ubuntu-24.04 \
    --disk 200 --ports 22,3389
✓ VM i-77e0c9d3 booted (2m41s)
$ ssh root@i-77e0c9d3.powergpu.ai
root@i-77e0c9d3:~# uname -r
6.8.0-49-generic   # your kernel, swap it if you like
```

## GPU VMs: FAQ

Boot images and cloud-init details in the [instance docs](https://powergpu.ai/docs/instances).

**How is a VM different from a container instance here?**

A VM is a full KVM virtual machine: your own kernel, systemd, root, reboots, custom drivers and GUI sessions. Container instances share the host kernel and start faster (~30 s vs 2–4 min) — use VMs when you need kernel control, a desktop, or software that refuses to live in Docker.

**Can I run a graphical desktop?**

Yes — the Ubuntu VM image ships with an optional XFCE desktop and self-hosted VNC/RDP over the instance's mapped ports. Studios use it for interactive Blender/DaVinci sessions on workstation cards.

**Do VMs cost more than containers?**

No — same GPU sheet, same $0.08/GB/mo storage. The only practical difference is the slower start and a slightly larger disk footprint for the OS.

**Can I bring my own image?**

Cloud-init compatible qcow2 images can be imported from a URL. Snapshots of a running VM export the same way, so migration in and out is symmetric.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/products/vms · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/products/volumes -->

---
title: "GPU Cloud Storage Volumes — NVMe at $0.08/GB/mo | PowerGPU"
description: "Persistent NVMe network volumes for GPU instances: $0.08/GB/month flat, per-second billing, survive instance destruction, read-only sharing for serving fleets."
url: https://powergpu.ai/products/volumes
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Products · volumes

# GPU cloud volumes: storage that outlives the instance

Network NVMe volumes at a flat **$0.08** /GB/month, billed per second. Keep datasets, checkpoints and model libraries warm while instances come and go — attach to any machine in the region in seconds.

### Instance-independent

Destroy the expensive GPU the second a job ends; the volume keeps your state for the next one. The core per-second cost pattern.

### NVMe throughput

Multi-GB/s reads put 70B weights in VRAM in tens of seconds — no S3 download tax at every deploy.

### Read-only fan-out

One writable attachment, many readers: a serving fleet mounts the same model library without copies.

### One flat rate

$0.08/GB/mo everywhere. No IOPS tiers, no throughput classes, no per-request fees.

## What storage really costs here

Per-second billing means you can also think of it hourly — both columns are the same number.

| Scenario | Size | Per month | Per hour | Notes |
| --- | --- | --- | --- | --- |
| **A LoRA workspace** | 50 GB | $4.00 | $0.0055 | checkpoints + datasets for a fine-tune |
| **A model library** | 500 GB | $40.00 | $0.0548 | ~6 quantized 70B models ready to serve |
| **A render vault** | 2,000 GB | $160.00 | $0.2192 | scene files + frames for a studio pipeline |
| **A dataset lake** | 10,000 GB | $800.00 | $1.0959 | LAION-scale shards staged next to the GPUs |

## The pattern that saves the most money

GPU time is the expensive line; storage is cents. Volumes let you stop paying for the former without losing your work:

- Keep **datasets & weights** on a volume; deploy GPUs against it only while computing.
- Point **interruptible training** checkpoints at a volume — interruptions become free restarts.
- Build a **team model library** once; every new instance mounts it read-only in seconds.
- Snapshot instance disks to volumes before destroying — [one command](https://powergpu.ai/docs/volumes).

*volumes*

```
$ powergpu volume create --name models --size 500 --region eu-west-1
✓ vol-8c31f2 · 500 GB · $40.00/mo
$ powergpu launch --gpu l40s --template vllm --volume models:/models:ro
✓ instance i-4fd02b11 running · library mounted read-only
```

## GPU volumes: FAQ

Mount options and snapshots in the [storage docs](https://powergpu.ai/docs/volumes).

**How are volumes billed?**

Allocated size × $0.08/GB/month, per second, whether attached or not. A 100 GB volume costs about $0.0110 per hour — delete it and billing stops that second.

**Can several instances share one volume?**

One read-write attachment at a time, plus unlimited read-only attachments in the same region — the standard pattern for serving fleets reading one model library while a trainer writes checkpoints elsewhere.

**How fast are they?**

NVMe-backed over the datacenter fabric: multi-GB/s sequential reads on datacenter hosts — enough to load a 70B model in well under a minute — with local instance NVMe still there for scratch I/O.

**Do volumes survive instance destruction?**

Yes — that is the point. Destroy instances freely; volumes persist until you delete them, and reattach to any new instance in their region.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/products/volumes · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/legal/terms -->

---
title: "Terms of service | PowerGPU"
description: "The PowerGPU terms of service: accounts, credits and crypto payments, service modes, acceptable use, reservations, liability and termination — in plain language."
url: https://powergpu.ai/legal/terms
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Legal · Terms of service

# Terms of service

Written to be read. Section titles are honest summaries; where a clause is unusual for a cloud provider (crypto-only payment, privacy defaults), it is called out rather than buried.

- Effective **2026-09-03**
- Version **1.0**
- Reading time **4 min**

In plain language

- You prepay in crypto; usage draws the balance down per second. Unused credits never expire and are refundable in the original coin on request.
- On-demand and reserved capacity is guaranteed; interruptible capacity can be paused (never destroyed) — that is what buys its 50% discount.
- Your data, models and outputs stay yours. Destroy = cryptographic erasure. Keep your own copies of anything you cannot lose.
- No mining, no attacks, no illegal content. Serious abuse is stopped first; everything else gets 48 hours to talk.
- Our liability is capped at what you paid us in the last three months (or $100). Notices go through a console ticket.

This summary helps you read the document; the numbered sections below are the binding text.

## 1 · Who we are, what this covers

"PowerGPU", "we" and "us" mean the operator of powergpu.ai and cloud.powergpu.ai (the "Service") — rental of GPU compute, storage and network by the second. Creating an account or using the Service means you accept these terms, the [Acceptable Use Policy](https://powergpu.ai/legal/aup) (part of these terms) and the [Privacy Policy](https://powergpu.ai/legal/privacy). If you use the Service for an organisation, you confirm you may bind it.

## 2 · Accounts

- One human per account; you are responsible for what happens under your credentials and API keys.
- Keep the email reachable — it is our only channel for service notices (balance exhaustion, interruptions, ticket answers).
- You must be at least 18 and not barred from receiving services under applicable sanctions law. We may refuse or close accounts used to evade previous enforcement.

## 3 · Credits & payment (crypto only)

- The Service is prepaid: you top up a USD-denominated credit balance using the supported cryptocurrencies. We never ask for card or bank details — a request for them is not from us.
- Credits are valued at the USD amount selected at top-up; the coin conversion is fixed by the payment window shown before you send. Network fees are yours; confirmation times are the network's.
- Credits do not expire. Unused credits are refundable on request in the original coin at the USD value remaining, minus network fees, except credits granted by us as goodwill.
- Usage draws the balance down per second as metered by the platform. Our metering is authoritative; obvious metering errors will be corrected in your favour when found.
- At zero balance we stop (never destroy) running instances; storage may draw the balance slightly negative until settled.

## 4 · The service modes

- **On-demand** — capacity is yours from successful deploy until you stop or destroy it, at the price shown at deploy.
- **Interruptible** — half the on-demand rate; we may pause (stop) the instance when capacity is required, with the recovery behaviour described in the docs. Interruption is not a fault.
- **Reserved** — committed capacity for an agreed term and rate; commitments bill for the term whether used or not and are non-cancellable, though upgradeable.
- Prices for new deployments follow the published sheet, which changes with the weekly market re-check. Running instances keep their deploy-time price.

## 5 · Your content

- What you run and store stays yours. We claim no rights over your data, models or outputs.
- We access instance contents only when strictly necessary to operate the Service, at your request (support), or when the AUP compels investigation — and we log such access internally.
- Destroying an instance or volume cryptographically erases it. Keep your own copies of anything you cannot lose; per-second infrastructure is designed to be disposable.

## 6 · Acceptable use

The [AUP](https://powergpu.ai/legal/aup) is short and enforced: illegal content and attacks are banned, abuse that endangers the platform or other customers is stopped first and discussed second. Cryptocurrency mining is not permitted — the economics only work by abusing interruptible capacity, so we exclude it for everyone.

## 7 · Availability

On-demand and reserved capacity carry the [SLA](https://powergpu.ai/legal/sla) (99.9% monthly, remedied in credits). Interruptible capacity, by design, carries none. Maintenance windows are announced on [/status](https://powergpu.ai/status).

## 8 · Liability

- The Service is provided "as is"; to the maximum extent the law allows, our aggregate liability for any claim is capped at the greater of $100 or the amount you paid us in the three months before the event.
- We are not liable for indirect or consequential loss — lost profits, lost models, lost training time — nor for the behaviour of cryptocurrency networks.
- Nothing here limits liability that cannot lawfully be limited.

## 9 · Suspension & termination

- You can leave any time: destroy resources, request a refund of remaining credits, done.
- We may suspend resources immediately for AUP violations, security emergencies or legal compulsion, and will say why unless legally prevented.
- Closed-account data follows the retention rules in the [privacy policy](https://powergpu.ai/legal/privacy).

## 10 · Changes, law, contact

- Material changes to these terms are announced in-console 14 days ahead; continued use is acceptance.
- These terms are governed by the law of the operator's place of establishment, excluding conflict rules; disputes go to the courts there unless mandatory consumer law says otherwise.
- All notices to us: a ticket in the [console support desk](https://cloud.powergpu.ai/app/support) — the authenticated channel is the official one.

Version 1.0, effective 2026-09-03. Material changes are announced in-console 14 days ahead; previous versions are available on request through a support ticket.

Questions about this document?

Legal questions go through the same authenticated channel as everything else — a support ticket in the console, answered by a human.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Key facts

- Crypto only · no card · no KYC
- No IP retention, no trackers
- Per-second billing, credits never expire
- 99.9% SLA, remedied in credits

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/legal/terms · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/legal/privacy -->

---
title: "Privacy policy | PowerGPU"
description: "PowerGPU privacy policy: we store an email, a password hash and a usage ledger — no names, no payment PII, no IP retention, no trackers. Short because the data is."
url: https://powergpu.ai/legal/privacy
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Legal · Privacy policy

# Privacy policy

Most privacy policies are long because the data practices are. Ours fits on one screen because we designed the collection out: crypto payment, no KYC, no IP retention, no trackers.

- Effective **2026-09-03**
- Version **1.0**
- Reading time **3 min**

In plain language

- We store an email, a password hash, your balance/usage ledger, instance metadata and tickets — nothing else.
- No name, address or ID documents (crypto, no KYC); no IP addresses in logs or the database; no trackers or analytics.
- We never read your instance disks, volumes or traffic in normal operations; the narrow exceptions are listed in section 4.
- Data leaves us only for transactional email delivery, payment infrastructure, or valid legal process.
- Export or delete everything yourself from the console; deletion completes after a 90-day accounting window.

This summary helps you read the document; the numbered sections below are the binding text.

## 1 · What we store

| Data | Why | Kept until |
| --- | --- | --- |
| Email address | sign-in, service notices, ticket answers | account deletion |
| Password (Argon2id hash) | authentication — we never see the password itself | account deletion |
| Balance & usage ledger | metering, statements, dispute resolution | account deletion + 90 days |
| Instance/volume metadata | operating your resources (not their contents) | resource destruction + 90 days |
| Top-up records (coin, amount, tx reference) | crediting and troubleshooting payments | account deletion + 90 days |
| Support tickets | answering you, quality | account deletion |
| API key hashes & last-use time | authentication, your own audit | key revocation |

## 2 · What we deliberately do not collect

- **No identity data** — no name, address, phone, ID documents. Payment is cryptocurrency; there is no KYC process.
- **No IP retention** — connection addresses are used transiently to serve requests and are not written to logs or the database.
- **No trackers** — no analytics scripts, no advertising pixels, no fingerprinting. The only cookies are the session cookie and a CSRF token on the console.
- **No content inspection** — we do not read instance disks, volumes or traffic in the normal course of operations (see §4 for the narrow exceptions).

## 3 · How the data is used

To run the Service you asked for: authenticate you, meter usage, credit payments, answer tickets and send the few operational emails that matter (password reset, instances stopped at zero balance, top-up receipts, ticket answers). No marketing lists, no profiling, no sale or rental of data — ever.

## 4 · When data leaves us

- **Processors** — a transactional email provider delivers the emails above (recipient and message only); payment infrastructure sees deposit addresses and amounts, never your account data.
- **Legal compulsion** — we answer valid legal process with what we hold, which §1 and §2 keep minimal, and we push back on overbroad requests. We cannot hand over what we never collected.
- **Abuse response** — investigating a concrete AUP violation may require inspecting the specific resource involved; such access is logged and limited to the incident.

## 5 · Your controls

- **Export** — your ledger and account data are downloadable from the console (or via the API) in machine-readable form.
- **Deletion** — deleting the account (Settings) destroys resources, erases content cryptographically and removes personal data after the 90-day accounting window.
- **Correction** — the email address is editable in Settings; everything else we hold is operational data you generate.
- Questions or requests beyond the console's buttons: a [support ticket](https://cloud.powergpu.ai/app/support) reaches the people responsible directly.

## 6 · Legal bases & transfers

Where GDPR-style laws apply, processing rests on contract performance (running the Service) and legitimate interest (security, abuse prevention). Data is processed in datacenters under contractual safeguards; the storage footprint in §1 is identical in every region.

## 7 · Changes

Material changes are announced in-console 14 days ahead. The changelog of this document lives at the bottom of the page source, dated.

Version 1.0, effective 2026-09-03. Material changes are announced in-console 14 days ahead; previous versions are available on request through a support ticket.

Questions about this document?

Legal questions go through the same authenticated channel as everything else — a support ticket in the console, answered by a human.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Key facts

- Crypto only · no card · no KYC
- No IP retention, no trackers
- Per-second billing, credits never expire
- 99.9% SLA, remedied in credits

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/legal/privacy · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/legal/aup -->

---
title: "Acceptable use policy | PowerGPU"
description: "What may and may not run on PowerGPU: the banned list (attacks, illegal content, spam, mining), the research allowance, and how enforcement works."
url: https://powergpu.ai/legal/aup
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Legal · Acceptable use policy

# Acceptable use policy

Privacy-first hosting is not anything-goes hosting. This page draws the line precisely, so the 99% of customers training models and rendering frames never feel it.

- Effective **2026-09-03**
- Version **1.0**
- Reading time **2 min**

In plain language

- Banned outright: illegal content, attacks and abuse infrastructure, cryptocurrency mining, sanctions evasion.
- Explicitly fine: training and serving models, lawful scraping, security research on your own resources, reselling capacity to users who follow this policy.
- Do not interfere with metering, isolation or scheduling — for example anti-interruption tricks on interruptible instances.
- Active harm is suspended immediately; anything else gets a notice and 48 hours to respond.
- Report abuse hosted here with a ticket titled “Abuse” — first response within 24 hours.

This summary helps you read the document; the numbered sections below are the binding text.

## 1 · Never allowed

- **Illegal content** — CSAM (zero tolerance, reported to the relevant authorities), content you have no right to process, material illegal where we or the facility operate.
- **Attacks** — DoS/DDoS in any direction, unauthorised access attempts, credential stuffing, malware command-and-control, exploit hosting aimed at third parties.
- **Abuse infrastructure** — spam origination, phishing sites, carding, botnet nodes, proxies/VPN exits operated to launder abusive traffic.
- **Cryptocurrency mining** — including "just overnight". Mining monetises interruptible capacity into uselessness for everyone; it is excluded platform-wide regardless of coin or intent.
- **Sanctions evasion** — use by or for sanctioned parties and destinations.

## 2 · Explicitly fine

Because "AI policy" pages often blur it: training and serving models (including open-weights LLMs and image/video models), scraping you are lawfully entitled to do, security research *against your own resources*, penetration-testing targets you own or have written authorisation for, Tor/VPN clients for your own egress privacy, and blockchain *nodes* (validation is not mining). When in doubt, ask first via a [ticket](https://cloud.powergpu.ai/app/support) — a written yes beats guessing.

## 3 · Platform health rules

- No deliberate interference with metering, isolation or capacity scheduling (e.g. anti-interruption tricks on interruptible instances).
- Outbound network abuse controls apply (port-scan storms, mail floods); genuine research traffic that trips them is unblocked with one ticket.
- Resale of capacity is fine (agencies, render farms, notebooks-as-a-service) as long as your end users stay within this AUP — you answer for them.

## 4 · How enforcement works

1. **Active harm** (attacks in progress, CSAM) — immediate suspension of the offending resources, then notice.
2. **Everything else** — notice first with 48 h to respond; most cases are misunderstandings and end there.
3. Repeated or bad-faith violations end the account; remaining legitimate credits are refunded per the [terms](https://powergpu.ai/legal/terms), unless the law forbids it.
4. Reports of abuse hosted here: open a ticket titled **Abuse** (any account, free to create) with URLs/indicators — first response within 24 h.

Version 1.0, effective 2026-09-03. Material changes are announced in-console 14 days ahead; previous versions are available on request through a support ticket.

Questions about this document?

Legal questions go through the same authenticated channel as everything else — a support ticket in the console, answered by a human.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Key facts

- Crypto only · no card · no KYC
- No IP retention, no trackers
- Per-second billing, credits never expire
- 99.9% SLA, remedied in credits

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/legal/aup · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/legal/sla -->

---
title: "Service level agreement (SLA) | PowerGPU"
description: "The PowerGPU SLA: 99.9% monthly uptime on on-demand and reserved capacity, automatic-on-request credit remedies up to 50%, clear definitions and exclusions."
url: https://powergpu.ai/legal/sla
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Legal · Service level agreement

# Service level agreement

99.9% monthly uptime on on-demand and reserved capacity, remedied in credits when missed. Interruptible capacity is excluded by design — its discount is the compensation.

- Effective **2026-09-03**
- Version **1.0**
- Reading time **2 min**

In plain language

- On-demand and reserved instances are backed by a monthly uptime commitment; interruptible capacity is excluded by design.
- Downtime = minutes an instance is unreachable on its mapped ports because of our platform, or a region rejecting all deploys.
- Credits: 10% below the target, 25% below 99.0%, 50% below 95.0% of that instance’s monthly charges.
- Claim within 30 days with a ticket titled “SLA”; we verify against telemetry and credit your balance.
- Excluded: your own workload failures, AUP suspensions, zero-balance stops, announced maintenance (≤4 h/month), force majeure.

This summary helps you read the document; the numbered sections below are the binding text.

## 1 · Definitions

- **Downtime** — minutes in which a running on-demand/reserved instance is unreachable on its mapped ports due to a failure of our platform (host, fabric, edge), or in which a region rejects all deploys.
- **Monthly uptime** — (minutes in month − downtime) ÷ minutes in month, per instance.
- Downtime starts when our monitoring detects the failure or you open a ticket about it, whichever is first, and ends at restoration.

## 2 · Credit schedule

| Monthly uptime of the instance | Credit (of that instance's charges that month) |
| --- | --- |
| < 99.9% – ≥ 99.0% | 10% |
| < 99.0% – ≥ 95.0% | 25% |
| < 95.0% | 50% |

Claim with a ticket titled **SLA** within 30 days, naming the instance(s) and window; we verify against platform telemetry and apply the credit to your balance. Credits are the exclusive remedy for availability misses.

## 3 · Exclusions

- Interruptible instances (interruption is a feature, not downtime);
- anything inside your workload — crashed processes, OOM kills, misconfigured images;
- suspensions under the [AUP](https://powergpu.ai/legal/aup) or zero-balance auto-stops;
- announced maintenance windows (published on [/status](https://powergpu.ai/status) ≥72 h ahead, ≤4 h/month);
- force majeure and failures of networks we do not operate (your ISP, cryptocurrency networks).

## 4 · Enterprise riders

[Enterprise reservations](https://powergpu.ai/enterprise) can attach stronger riders — named-engineer support with 15-minute response, per-fleet uptime targets, custom maintenance calendars — specified in the reservation quote and prevailing over this page where they differ.

## 5 · Transparency

Live component and per-region status, plus the full incident history this SLA is judged against, are public at [powergpu.ai/status](https://powergpu.ai/status) — no login, no "operational" theatre over unlisted incidents: if it impacted customers, it is listed.

Version 1.0, effective 2026-09-03. Material changes are announced in-console 14 days ahead; previous versions are available on request through a support ticket.

Questions about this document?

Legal questions go through the same authenticated channel as everything else — a support ticket in the console, answered by a human.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Key facts

- Crypto only · no card · no KYC
- No IP retention, no trackers
- Per-second billing, credits never expire
- 99.9% SLA, remedied in credits

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/legal/sla · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/b300 -->

---
title: "Rent B300 — $6.737/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the B300 (288 GB HBM3e) from $3.368/hr interruptible or $6.737/hr on-demand — fixed, ≥30% below market. 34 GPUs in 4 regions, per-second billing."
url: https://powergpu.ai/gpu/b300
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter flagship · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA B300 — 288 GB, $6.737/hr on-demand

- VRAM 288 GB HBM3e
- FP16 tensor 2,800 TFLOPS
- PowerScore 1972 RTX 3090 = 100
- Configs 1–2× NVLink
- Online now 34 4 regions

[On-demand (guaranteed) $6.737 /GPU-hr ≈ $4,918/mo · market ~~$9.63~~ (−30%)](https://cloud.powergpu.ai/?gpu=b300) [Interruptible $3.368 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=b300&type=spot) [Reserved 3 mo $4.379 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter flagship · Blackwell architecture

The B300 is Blackwell Ultra: 288 GB of HBM3e per GPU and the highest FP4 inference throughput NVIDIA ships. Teams rent it for frontier-scale serving and for training runs where even a 192 GB shard is too small. Supply is thin everywhere; here it is priced by the same public rule as an RTX 3060.

This is training-grade silicon: HBM3e memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.

## NVIDIA B300 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA B300 · **Architecture**: Blackwell (2025)
- **VRAM**: 288 GB HBM3e · **Memory bandwidth**: 8,000 GB/s
- **FP16 tensor perf.**: 2,800 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: — · **TDP**: 1100 W
- **PowerScore ((RTX 3090 = 100))**: 1972 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 2× · NVLink · **Max instance storage**: 16,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## B300 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the B300 ($9.63/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $6.737, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $6.737 | $161.69 | $4,918 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $3.368 | $80.83 | $2,459 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $4.379 | $105.10 | $3,197 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 2× machine costs exactly 2× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a B300 (288 GB VRAM)

With 288 GB of HBM3e, a single card holds a **~105B-parameter LLM in FP16** or up to **~405B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 2× GPUs on one machine with NVLink for bigger models or bigger batches — the per-GPU price stays $6.737.

- One-click template: [vLLM on a B300](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch NGC on a B300](https://powergpu.ai/templates/pytorch-ngc)
- One-click template: [Axolotl — Fine Tuning on a B300](https://powergpu.ai/templates/axolotl-fine-tuning)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## B300 availability by region

34 × B300 across 4 machines, live from inventory:

- Seattle, WA
- Frankfurt
- Montréal
- Tokyo

## B300 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **B300** (this card) | 288 GB | 2,800 | $6.737 | $2.41‰ |
| [B200](https://powergpu.ai/gpu/b200) | 192 GB | 2,250 | $5.425 | $2.41‰ |
| [H200](https://powergpu.ai/gpu/h200) | 141 GB | 990 | $2.791 | $2.82‰ |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | 990 | $1.428 | $1.44‰ |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | 312 | $0.560 | $1.79‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [B200 vs B300](https://powergpu.ai/compare/b200-vs-b300) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a B300: frequently asked questions

**How much does it cost to rent an NVIDIA B300 per hour?**

$6.737 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $9.63. Interruptible capacity costs $3.368/hr and a 3-month reservation $4.379/hr. Around $4,918/month if you keep one running non-stop, billed per second.

**What can a B300 with 288 GB VRAM run?**

In LLM terms, roughly a 105B-parameter model in FP16 or up to ~405B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 2× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the B300 available to rent right now?**

Yes — 34 GPUs across 4 machines in 4 regions are listed as we render this page. Configurations go from 1× to 2× with NVLink on multi-GPU chassis. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a B300?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by B300, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu b300 --template pytorch.

**Why is the B300 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/b300 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/b200 -->

---
title: "Rent B200 — $5.425/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the B200 (192 GB HBM3e) from $2.712/hr interruptible or $5.425/hr on-demand — fixed, ≥30% below market. 32 GPUs in 4 regions, per-second billing."
url: https://powergpu.ai/gpu/b200
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter flagship · Blackwell · launched 2024 · prices checked 2026-09-14

# Rent NVIDIA B200 — 192 GB, $5.425/hr on-demand

- VRAM 192 GB HBM3e
- FP16 tensor 2,250 TFLOPS
- PowerScore 1585 RTX 3090 = 100
- Configs 1–8× NVLink
- Online now 32 4 regions

[On-demand (guaranteed) $5.425 /GPU-hr ≈ $3,960/mo · market ~~$7.75~~ (−30%)](https://cloud.powergpu.ai/?gpu=b200) [Interruptible $2.712 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=b200&type=spot) [Reserved 3 mo $3.526 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter flagship · Blackwell architecture

The B200 is the Blackwell workhorse: 192 GB HBM3e, 8 TB/s and a second-generation Transformer Engine with FP4. Against an H100 it trains roughly 2–3× faster per GPU and serves large models several times faster, so a B200 hour frequently costs less per token than an H100 hour despite the higher rate.

This is training-grade silicon: HBM3e memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.

## NVIDIA B200 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA B200 · **Architecture**: Blackwell (2024)
- **VRAM**: 192 GB HBM3e · **Memory bandwidth**: 8,000 GB/s
- **FP16 tensor perf.**: 2,250 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: — · **TDP**: 1000 W
- **PowerScore ((RTX 3090 = 100))**: 1585 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · NVLink · **Max instance storage**: 16,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## B200 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the B200 ($7.75/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $5.425, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $5.425 | $130.20 | $3,960 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $2.712 | $65.09 | $1,980 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $3.526 | $84.62 | $2,574 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a B200 (192 GB VRAM)

With 192 GB of HBM3e, a single card holds a **~72B-parameter LLM in FP16** or up to **~235B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine with NVLink for bigger models or bigger batches — the per-GPU price stays $5.425.

- Recommended for [llm training](https://powergpu.ai/use-cases/llm-training) — H100 SXM from 30%+ under market
- One-click template: [vLLM on a B200](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch NGC on a B200](https://powergpu.ai/templates/pytorch-ngc)
- One-click template: [Axolotl — Fine Tuning on a B200](https://powergpu.ai/templates/axolotl-fine-tuning)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## B200 availability by region

32 × B200 across 9 machines, live from inventory:

- Ashburn, VA
- Tokyo
- Dallas, TX
- Singapore

## B200 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **B200** (this card) | 192 GB | 2,250 | $5.425 | $2.41‰ |
| [B300](https://powergpu.ai/gpu/b300) | 288 GB | 2,800 | $6.737 | $2.41‰ |
| [H200](https://powergpu.ai/gpu/h200) | 141 GB | 990 | $2.791 | $2.82‰ |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | 990 | $1.428 | $1.44‰ |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | 312 | $0.560 | $1.79‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H200 vs B200](https://powergpu.ai/compare/h200-vs-b200) — specs, price per hour, which to rent
- [H100 SXM vs B200](https://powergpu.ai/compare/h100-sxm-vs-b200) — specs, price per hour, which to rent
- [B200 vs B300](https://powergpu.ai/compare/b200-vs-b300) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a B200: frequently asked questions

**How much does it cost to rent an NVIDIA B200 per hour?**

$5.425 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $7.75. Interruptible capacity costs $2.712/hr and a 3-month reservation $3.526/hr. Around $3,960/month if you keep one running non-stop, billed per second.

**What can a B200 with 192 GB VRAM run?**

In LLM terms, roughly a 72B-parameter model in FP16 or up to ~235B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the B200 available to rent right now?**

Yes — 32 GPUs across 9 machines in 4 regions are listed as we render this page. Configurations go from 1× to 8× with NVLink on multi-GPU chassis. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a B200?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by B200, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu b200 --template pytorch.

**Why is the B200 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/b200 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/h200 -->

---
title: "Rent H200 — $2.791/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the H200 (141 GB HBM3e) from $1.395/hr interruptible or $2.791/hr on-demand — fixed, ≥30% below market. 50 GPUs in 6 regions, per-second billing."
url: https://powergpu.ai/gpu/h200
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter flagship · Hopper · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA H200 — 141 GB, $2.791/hr on-demand

- VRAM 141 GB HBM3e
- FP16 tensor 990 TFLOPS
- PowerScore 697 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 50 6 regions

[On-demand (guaranteed) $2.791 /GPU-hr ≈ $2,037/mo · market ~~$3.99~~ (−30%)](https://cloud.powergpu.ai/?gpu=h200) [Interruptible $1.395 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=h200&type=spot) [Reserved 3 mo $1.814 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter flagship · Hopper architecture

The H200 is an H100 with 141 GB of HBM3e at 4.8 TB/s — the same Hopper compute, 76% more memory and 43% more bandwidth. That is exactly what long-context inference, MoE serving and big micro-batches want; it is the card to rent when 80 GB shards force awkward parallelism.

This is training-grade silicon: HBM3e memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.

## NVIDIA H200 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA H200 · **Architecture**: Hopper (2023)
- **VRAM**: 141 GB HBM3e · **Memory bandwidth**: 4,800 GB/s
- **FP16 tensor perf.**: 990 TFLOPS · **FP32 perf.**: 67.0 TFLOPS
- **CUDA cores**: 16,896 · **TDP**: 700 W
- **PowerScore ((RTX 3090 = 100))**: 697 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 16,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## H200 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the H200 ($3.99/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $2.791, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $2.791 | $66.98 | $2,037 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $1.395 | $33.48 | $1,018 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $1.814 | $43.54 | $1,324 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a H200 (141 GB VRAM)

With 141 GB of HBM3e, a single card holds a **~49B-parameter LLM in FP16** or up to **~141B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $2.791.

- Recommended for [llm training](https://powergpu.ai/use-cases/llm-training) — H100 SXM from 30%+ under market
- Recommended for [scientific computing](https://powergpu.ai/use-cases/hpc) — Up to 4.8 TB/s per GPU
- One-click template: [vLLM on a H200](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch NGC on a H200](https://powergpu.ai/templates/pytorch-ngc)
- One-click template: [Axolotl — Fine Tuning on a H200](https://powergpu.ai/templates/axolotl-fine-tuning)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## H200 availability by region

50 × H200 across 7 machines, live from inventory:

- Stockholm
- Los Angeles, CA
- Mumbai
- Frankfurt
- London
- Chicago, IL

## H200 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **H200** (this card) | 141 GB | 990 | $2.791 | $2.82‰ |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | 990 | $1.428 | $1.44‰ |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | 312 | $0.560 | $1.79‰ |
| [B200](https://powergpu.ai/gpu/b200) | 192 GB | 2,250 | $5.425 | $2.41‰ |
| [B300](https://powergpu.ai/gpu/b300) | 288 GB | 2,800 | $6.737 | $2.41‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H100 SXM vs H200](https://powergpu.ai/compare/h100-sxm-vs-h200) — specs, price per hour, which to rent
- [H200 vs B200](https://powergpu.ai/compare/h200-vs-b200) — specs, price per hour, which to rent
- [H200 vs H200 NVL](https://powergpu.ai/compare/h200-vs-h200-nvl) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a H200: frequently asked questions

**How much does it cost to rent an NVIDIA H200 per hour?**

$2.791 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $3.99. Interruptible capacity costs $1.395/hr and a 3-month reservation $1.814/hr. Around $2,037/month if you keep one running non-stop, billed per second.

**What can a H200 with 141 GB VRAM run?**

In LLM terms, roughly a 49B-parameter model in FP16 or up to ~141B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the H200 available to rent right now?**

Yes — 50 GPUs across 7 machines in 6 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a H200?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by H200, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu h200 --template pytorch.

**Why is the H200 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/h200 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/h200-nvl -->

---
title: "Rent H200 NVL — $2.650/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the H200 NVL (141 GB HBM3e) from $1.325/hr interruptible or $2.650/hr on-demand — fixed, ≥30% below market. 27 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/h200-nvl
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter flagship · Hopper · launched 2024 · prices checked 2026-09-14

# Rent NVIDIA H200 NVL — 141 GB, $2.650/hr on-demand

- VRAM 141 GB HBM3e
- FP16 tensor 835 TFLOPS
- PowerScore 588 RTX 3090 = 100
- Configs 1–8× NVLink
- Online now 27 2 regions

[On-demand (guaranteed) $2.650 /GPU-hr ≈ $1,935/mo · market ~~$3.79~~ (−30%)](https://cloud.powergpu.ai/?gpu=h200-nvl) [Interruptible $1.325 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=h200-nvl&type=spot) [Reserved 3 mo $1.722 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter flagship · Hopper architecture

The H200 NVL is the PCIe form of the H200: 141 GB HBM3e in a standard server slot, bridgeable in pairs or quads with NVLink. It brings H200-class memory to hosts without an SXM baseboard, which is why it rents for less per hour than the SXM part.

This is training-grade silicon: HBM3e memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.

## NVIDIA H200 NVL specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA H200 NVL · **Architecture**: Hopper (2024)
- **VRAM**: 141 GB HBM3e · **Memory bandwidth**: 4,800 GB/s
- **FP16 tensor perf.**: 835 TFLOPS · **FP32 perf.**: 60.0 TFLOPS
- **CUDA cores**: 16,896 · **TDP**: 600 W
- **PowerScore ((RTX 3090 = 100))**: 588 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · NVLink · **Max instance storage**: 16,000 GB NVMe
- **Network up to**: 1,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## H200 NVL price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the H200 NVL ($3.79/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $2.650, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $2.650 | $63.60 | $1,935 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $1.325 | $31.80 | $967 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $1.722 | $41.33 | $1,257 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a H200 NVL (141 GB VRAM)

With 141 GB of HBM3e, a single card holds a **~49B-parameter LLM in FP16** or up to **~141B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine with NVLink for bigger models or bigger batches — the per-GPU price stays $2.650.

- One-click template: [vLLM on a H200 NVL](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch NGC on a H200 NVL](https://powergpu.ai/templates/pytorch-ngc)
- One-click template: [Axolotl — Fine Tuning on a H200 NVL](https://powergpu.ai/templates/axolotl-fine-tuning)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## H200 NVL availability by region

27 × H200 NVL across 2 machines, live from inventory:

- Ashburn, VA
- Tokyo

## H200 NVL vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **H200 NVL** (this card) | 141 GB | 835 | $2.650 | $3.17‰ |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | 756 | $1.867 | $2.47‰ |
| [H100 NVL](https://powergpu.ai/gpu/h100-nvl) | 80 GB | 835 | $1.811 | $2.17‰ |
| [A40](https://powergpu.ai/gpu/a40) | 48 GB | 144 | $0.765 | $5.31‰ |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | 362 | $0.514 | $1.42‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H200 vs H200 NVL](https://powergpu.ai/compare/h200-vs-h200-nvl) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a H200 NVL: frequently asked questions

**How much does it cost to rent an NVIDIA H200 NVL per hour?**

$2.650 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $3.79. Interruptible capacity costs $1.325/hr and a 3-month reservation $1.722/hr. Around $1,935/month if you keep one running non-stop, billed per second.

**What can a H200 NVL with 141 GB VRAM run?**

In LLM terms, roughly a 49B-parameter model in FP16 or up to ~141B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the H200 NVL available to rent right now?**

Yes — 27 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8× with NVLink on multi-GPU chassis. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a H200 NVL?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by H200 NVL, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu h200-nvl --template pytorch.

**Why is the H200 NVL cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/h200-nvl · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/h100-pcie -->

---
title: "Rent H100 PCIE — $1.867/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the H100 PCIE (80 GB HBM2e) from $0.933/hr interruptible or $1.867/hr on-demand — fixed, ≥30% below market. 73 GPUs in 8 regions, per-second billing."
url: https://powergpu.ai/gpu/h100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter flagship · Hopper · launched 2022 · prices checked 2026-09-14

# Rent NVIDIA H100 PCIE — 80 GB, $1.867/hr on-demand

- VRAM 80 GB HBM2e
- FP16 tensor 756 TFLOPS
- PowerScore 532 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 73 8 regions

[On-demand (guaranteed) $1.867 /GPU-hr ≈ $1,363/mo · market ~~$2.67~~ (−30%)](https://cloud.powergpu.ai/?gpu=h100-pcie) [Interruptible $0.933 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=h100-pcie&type=spot) [Reserved 3 mo $1.213 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter flagship · Hopper architecture

The H100 PCIe is the Hopper card for standard servers: 80 GB HBM2e at 2 TB/s, 350 W and the FP8 Transformer Engine, with NVLink limited to bridged pairs. For single-GPU fine-tuning, quantized 70B inference and FP8 serving it delivers most of the SXM experience at a lower hourly rate — the value Hopper pick.

This is training-grade silicon: HBM2e memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.

## NVIDIA H100 PCIE specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA H100 PCIE · **Architecture**: Hopper (2022)
- **VRAM**: 80 GB HBM2e · **Memory bandwidth**: 2,000 GB/s
- **FP16 tensor perf.**: 756 TFLOPS · **FP32 perf.**: 51.0 TFLOPS
- **CUDA cores**: 14,592 · **TDP**: 350 W
- **PowerScore ((RTX 3090 = 100))**: 532 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 16,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## H100 PCIE price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the H100 PCIE ($2.67/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $1.867, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $1.867 | $44.81 | $1,363 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.933 | $22.39 | $681 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $1.213 | $29.11 | $885 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a H100 PCIE (80 GB VRAM)

With 80 GB of HBM2e, a single card holds a **~32B-parameter LLM in FP16** or up to **~123B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $1.867.

- Recommended for [llm inference](https://powergpu.ai/use-cases/llm-inference) — Best $/token in class
- Recommended for [fine-tuning](https://powergpu.ai/use-cases/fine-tuning) — QLoRA 70B on one GPU
- Recommended for [video generation](https://powergpu.ai/use-cases/video-generation) — 32–141 GB VRAM on tap
- One-click template: [vLLM on a H100 PCIE](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch NGC on a H100 PCIE](https://powergpu.ai/templates/pytorch-ngc)
- One-click template: [Axolotl — Fine Tuning on a H100 PCIE](https://powergpu.ai/templates/axolotl-fine-tuning)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## H100 PCIE availability by region

73 × H100 PCIE across 10 machines, live from inventory:

- Seattle, WA
- Stockholm
- Tokyo
- Mumbai
- Chicago, IL
- Frankfurt
- Montréal
- Los Angeles, CA

## H100 PCIE vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **H100 PCIE** (this card) | 80 GB | 756 | $1.867 | $2.47‰ |
| [H100 NVL](https://powergpu.ai/gpu/h100-nvl) | 80 GB | 835 | $1.811 | $2.17‰ |
| [H200 NVL](https://powergpu.ai/gpu/h200-nvl) | 141 GB | 835 | $2.650 | $3.17‰ |
| [A40](https://powergpu.ai/gpu/a40) | 48 GB | 144 | $0.765 | $5.31‰ |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | 362 | $0.514 | $1.42‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H100 SXM vs H100 PCIE](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie) — specs, price per hour, which to rent
- [H100 PCIE vs A100 PCIE](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie) — specs, price per hour, which to rent
- [L40S vs H100 PCIE](https://powergpu.ai/compare/l40s-vs-h100-pcie) — specs, price per hour, which to rent
- [RTX PRO 6000 WS vs H100 PCIE](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a H100 PCIE: frequently asked questions

**How much does it cost to rent an NVIDIA H100 PCIE per hour?**

$1.867 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $2.67. Interruptible capacity costs $0.933/hr and a 3-month reservation $1.213/hr. Around $1,363/month if you keep one running non-stop, billed per second.

**What can a H100 PCIE with 80 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~123B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the H100 PCIE available to rent right now?**

Yes — 73 GPUs across 10 machines in 8 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a H100 PCIE?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by H100 PCIE, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu h100-pcie --template pytorch.

**Why is the H100 PCIE cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/h100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/h100-nvl -->

---
title: "Rent H100 NVL — $1.811/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the H100 NVL (80 GB HBM3) from $0.905/hr interruptible or $1.811/hr on-demand — fixed, ≥30% below market. 7 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/h100-nvl
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter flagship · Hopper · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA H100 NVL — 80 GB, $1.811/hr on-demand

- VRAM 80 GB HBM3
- FP16 tensor 835 TFLOPS
- PowerScore 588 RTX 3090 = 100
- Configs 1–4× NVLink
- Online now 7 2 regions

[On-demand (guaranteed) $1.811 /GPU-hr ≈ $1,322/mo · market ~~$2.59~~ (−30%)](https://cloud.powergpu.ai/?gpu=h100-nvl) [Interruptible $0.905 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=h100-nvl&type=spot) [Reserved 3 mo $1.177 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter flagship · Hopper architecture

The H100 NVL is the PCIe Hopper card tuned for inference: higher clocks and more bandwidth than the plain H100 PCIe, sold in NVLink-bridged pairs for serving 70B-class models across two cards. Rent it for Hopper serving throughput on PCIe hosts.

This is training-grade silicon: HBM3 memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.

## NVIDIA H100 NVL specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA H100 NVL · **Architecture**: Hopper (2023)
- **VRAM**: 80 GB HBM3 · **Memory bandwidth**: 3,900 GB/s
- **FP16 tensor perf.**: 835 TFLOPS · **FP32 perf.**: 60.0 TFLOPS
- **CUDA cores**: 14,592 · **TDP**: 400 W
- **PowerScore ((RTX 3090 = 100))**: 588 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 4× · NVLink · **Max instance storage**: 12,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## H100 NVL price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the H100 NVL ($2.59/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $1.811, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $1.811 | $43.46 | $1,322 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.905 | $21.72 | $661 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $1.177 | $28.25 | $859 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a H100 NVL (80 GB VRAM)

With 80 GB of HBM3, a single card holds a **~32B-parameter LLM in FP16** or up to **~123B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine with NVLink for bigger models or bigger batches — the per-GPU price stays $1.811.

- One-click template: [vLLM on a H100 NVL](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch NGC on a H100 NVL](https://powergpu.ai/templates/pytorch-ngc)
- One-click template: [Axolotl — Fine Tuning on a H100 NVL](https://powergpu.ai/templates/axolotl-fine-tuning)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## H100 NVL availability by region

7 × H100 NVL across 3 machines, live from inventory:

- Tokyo
- Stockholm

## H100 NVL vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **H100 NVL** (this card) | 80 GB | 835 | $1.811 | $2.17‰ |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | 756 | $1.867 | $2.47‰ |
| [H200 NVL](https://powergpu.ai/gpu/h200-nvl) | 141 GB | 835 | $2.650 | $3.17‰ |
| [A40](https://powergpu.ai/gpu/a40) | 48 GB | 144 | $0.765 | $5.31‰ |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | 362 | $0.514 | $1.42‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H100 NVL vs H100 SXM](https://powergpu.ai/compare/h100-nvl-vs-h100-sxm) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a H100 NVL: frequently asked questions

**How much does it cost to rent an NVIDIA H100 NVL per hour?**

$1.811 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $2.59. Interruptible capacity costs $0.905/hr and a 3-month reservation $1.177/hr. Around $1,322/month if you keep one running non-stop, billed per second.

**What can a H100 NVL with 80 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~123B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the H100 NVL available to rent right now?**

Yes — 7 GPUs across 3 machines in 2 regions are listed as we render this page. Configurations go from 1× to 4× with NVLink on multi-GPU chassis. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a H100 NVL?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by H100 NVL, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu h100-nvl --template pytorch.

**Why is the H100 NVL cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/h100-nvl · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/h100-sxm -->

---
title: "Rent H100 SXM — $1.428/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the H100 SXM (80 GB HBM3) from $0.714/hr interruptible or $1.428/hr on-demand — fixed, ≥30% below market. 54 GPUs in 7 regions, per-second billing."
url: https://powergpu.ai/gpu/h100-sxm
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter flagship · Hopper · launched 2022 · prices checked 2026-09-14

# Rent NVIDIA H100 SXM — 80 GB, $1.428/hr on-demand

- VRAM 80 GB HBM3
- FP16 tensor 990 TFLOPS
- PowerScore 697 RTX 3090 = 100
- Configs 1–8× NVLink
- Online now 54 7 regions

[On-demand (guaranteed) $1.428 /GPU-hr ≈ $1,042/mo · market ~~$2.04~~ (−30%)](https://cloud.powergpu.ai/?gpu=h100-sxm) [Interruptible $0.714 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=h100-sxm&type=spot) [Reserved 3 mo $0.928 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter flagship · Hopper architecture

The H100 SXM is the reference training GPU of its generation: 80 GB HBM3 at 3.35 TB/s, 700 W and full 900 GB/s NVLink on 8-GPU HGX baseboards. It is the deepest pool on the sheet, which keeps interruptible slots available, and the price-performance default for pre-training, RLHF and full fine-tunes.

This is training-grade silicon: HBM3 memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.

## NVIDIA H100 SXM specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA H100 SXM · **Architecture**: Hopper (2022)
- **VRAM**: 80 GB HBM3 · **Memory bandwidth**: 3,350 GB/s
- **FP16 tensor perf.**: 990 TFLOPS · **FP32 perf.**: 67.0 TFLOPS
- **CUDA cores**: 16,896 · **TDP**: 700 W
- **PowerScore ((RTX 3090 = 100))**: 697 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · NVLink · **Max instance storage**: 16,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## H100 SXM price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the H100 SXM ($2.04/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $1.428, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $1.428 | $34.27 | $1,042 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.714 | $17.14 | $521 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.928 | $22.27 | $677 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a H100 SXM (80 GB VRAM)

With 80 GB of HBM3, a single card holds a **~32B-parameter LLM in FP16** or up to **~123B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine with NVLink for bigger models or bigger batches — the per-GPU price stays $1.428.

- Recommended for [llm training](https://powergpu.ai/use-cases/llm-training) — H100 SXM from 30%+ under market
- Recommended for [scientific computing](https://powergpu.ai/use-cases/hpc) — Up to 4.8 TB/s per GPU
- One-click template: [vLLM on a H100 SXM](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch NGC on a H100 SXM](https://powergpu.ai/templates/pytorch-ngc)
- One-click template: [Axolotl — Fine Tuning on a H100 SXM](https://powergpu.ai/templates/axolotl-fine-tuning)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## H100 SXM availability by region

54 × H100 SXM across 12 machines, live from inventory:

- Amsterdam
- Stockholm
- Mumbai
- Montréal
- London
- Chicago, IL
- Singapore

## H100 SXM vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **H100 SXM** (this card) | 80 GB | 990 | $1.428 | $1.44‰ |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | 312 | $0.560 | $1.79‰ |
| [H200](https://powergpu.ai/gpu/h200) | 141 GB | 990 | $2.791 | $2.82‰ |
| [B200](https://powergpu.ai/gpu/b200) | 192 GB | 2,250 | $5.425 | $2.41‰ |
| [B300](https://powergpu.ai/gpu/b300) | 288 GB | 2,800 | $6.737 | $2.41‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H100 SXM vs A100 SXM4](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4) — specs, price per hour, which to rent
- [H100 SXM vs H200](https://powergpu.ai/compare/h100-sxm-vs-h200) — specs, price per hour, which to rent
- [H100 SXM vs B200](https://powergpu.ai/compare/h100-sxm-vs-b200) — specs, price per hour, which to rent
- [H100 SXM vs H100 PCIE](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie) — specs, price per hour, which to rent
- [H100 NVL vs H100 SXM](https://powergpu.ai/compare/h100-nvl-vs-h100-sxm) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a H100 SXM: frequently asked questions

**How much does it cost to rent an NVIDIA H100 SXM per hour?**

$1.428 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $2.04. Interruptible capacity costs $0.714/hr and a 3-month reservation $0.928/hr. Around $1,042/month if you keep one running non-stop, billed per second.

**What can a H100 SXM with 80 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~123B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the H100 SXM available to rent right now?**

Yes — 54 GPUs across 12 machines in 7 regions are listed as we render this page. Configurations go from 1× to 8× with NVLink on multi-GPU chassis. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a H100 SXM?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by H100 SXM, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu h100-sxm --template pytorch.

**Why is the H100 SXM cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/h100-sxm · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-pro-6000-s -->

---
title: "Rent RTX PRO 6000 S — $1.073/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX PRO 6000 S (48 GB GDDR7) from $0.536/hr interruptible or $1.073/hr on-demand — fixed, ≥30% below market. 99 GPUs in 7 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-pro-6000-s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX PRO 6000 S — 48 GB, $1.073/hr on-demand

- VRAM 48 GB GDDR7
- FP16 tensor 450 TFLOPS
- PowerScore 317 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 99 7 regions

[On-demand (guaranteed) $1.073 /GPU-hr ≈ $783/mo · market ~~$1.53~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-s) [Interruptible $0.536 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-s&type=spot) [Reserved 3 mo $0.697 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Blackwell architecture

The RTX PRO 6000 Server Edition is the passively cooled, rack-mounted sibling of the workstation card: the same Blackwell silicon tuned for 24/7 datacenter duty. Rent it for inference fleets that want Blackwell throughput with ECC memory and server thermals.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX PRO 6000 S specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX PRO 6000 S · **Architecture**: Blackwell (2025)
- **VRAM**: 48 GB GDDR7 · **Memory bandwidth**: 1,792 GB/s
- **FP16 tensor perf.**: 450 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: 24,064 · **TDP**: 600 W
- **PowerScore ((RTX 3090 = 100))**: 317 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX PRO 6000 S price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX PRO 6000 S ($1.53/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $1.073, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $1.073 | $25.75 | $783 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.536 | $12.86 | $391 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.697 | $16.73 | $509 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX PRO 6000 S (48 GB VRAM)

With 48 GB of GDDR7, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $1.073.

- One-click template: [Linux Desktop on a RTX PRO 6000 S](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX PRO 6000 S](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX PRO 6000 S](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX PRO 6000 S availability by region

99 × RTX PRO 6000 S across 8 machines, live from inventory:

- Chicago, IL
- Dallas, TX
- Ashburn, VA
- Frankfurt
- Tokyo
- Singapore
- São Paulo

## RTX PRO 6000 S vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX PRO 6000 S** (this card) | 48 GB | 450 | $1.073 | $2.38‰ |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | 505 | $1.040 | $2.06‰ |
| [RTX PRO 6000 Max-Q](https://powergpu.ai/gpu/rtx-pro-6000-max-q) | 96 GB | 420 | $0.980 | $2.33‰ |
| [RTX PRO 5000](https://powergpu.ai/gpu/rtx-pro-5000) | 32 GB | 280 | $0.560 | $2.00‰ |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | 364 | $0.467 | $1.28‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX PRO 6000 S: frequently asked questions

**How much does it cost to rent an NVIDIA RTX PRO 6000 S per hour?**

$1.073 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $1.53. Interruptible capacity costs $0.536/hr and a 3-month reservation $0.697/hr. Around $783/month if you keep one running non-stop, billed per second.

**What can a RTX PRO 6000 S with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX PRO 6000 S available to rent right now?**

Yes — 99 GPUs across 8 machines in 7 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX PRO 6000 S?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX PRO 6000 S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-pro-6000-s --template pytorch.

**Why is the RTX PRO 6000 S cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-pro-6000-s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-pro-6000-ws -->

---
title: "Rent RTX PRO 6000 WS — $1.040/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX PRO 6000 WS (96 GB GDDR7) from $0.520/hr interruptible or $1.040/hr on-demand — fixed, ≥30% below market. 74 GPUs in 9 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-pro-6000-ws
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX PRO 6000 WS — 96 GB, $1.040/hr on-demand

- VRAM 96 GB GDDR7
- FP16 tensor 505 TFLOPS
- PowerScore 356 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 74 9 regions

[On-demand (guaranteed) $1.040 /GPU-hr ≈ $759/mo · market ~~$1.49~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) [Interruptible $0.520 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws&type=spot) [Reserved 3 mo $0.676 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Blackwell architecture

The RTX PRO 6000 Blackwell Workstation Edition puts 96 GB of GDDR7 and 24,064 CUDA cores behind studio drivers. It is the largest-VRAM single card outside the HBM line — city-scale render scenes, 70B 4-bit models on one GPU, video diffusion without offloading — at a fraction of H100 pricing.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX PRO 6000 WS specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX PRO 6000 WS · **Architecture**: Blackwell (2025)
- **VRAM**: 96 GB GDDR7 · **Memory bandwidth**: 1,792 GB/s
- **FP16 tensor perf.**: 505 TFLOPS · **FP32 perf.**: 125.0 TFLOPS
- **CUDA cores**: 24,064 · **TDP**: 600 W
- **PowerScore ((RTX 3090 = 100))**: 356 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX PRO 6000 WS price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX PRO 6000 WS ($1.49/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $1.040, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $1.040 | $24.96 | $759 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.520 | $12.48 | $380 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.676 | $16.22 | $493 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX PRO 6000 WS (96 GB VRAM)

With 96 GB of GDDR7, a single card holds a **~32B-parameter LLM in FP16** or up to **~141B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $1.040.

- Recommended for [3d rendering](https://powergpu.ai/use-cases/rendering) — Per-second render billing
- One-click template: [Linux Desktop on a RTX PRO 6000 WS](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX PRO 6000 WS](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX PRO 6000 WS](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX PRO 6000 WS availability by region

74 × RTX PRO 6000 WS across 9 machines, live from inventory:

- Querétaro
- Los Angeles, CA
- São Paulo
- Montréal
- Tokyo
- London
- Amsterdam
- Johannesburg
- Madrid

## RTX PRO 6000 WS vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX PRO 6000 WS** (this card) | 96 GB | 505 | $1.040 | $2.06‰ |
| [RTX PRO 6000 S](https://powergpu.ai/gpu/rtx-pro-6000-s) | 48 GB | 450 | $1.073 | $2.38‰ |
| [RTX PRO 6000 Max-Q](https://powergpu.ai/gpu/rtx-pro-6000-max-q) | 96 GB | 420 | $0.980 | $2.33‰ |
| [RTX PRO 5000](https://powergpu.ai/gpu/rtx-pro-5000) | 32 GB | 280 | $0.560 | $2.00‰ |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | 364 | $0.467 | $1.28‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX PRO 6000 WS vs RTX 6000Ada](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada) — specs, price per hour, which to rent
- [RTX PRO 6000 WS vs H100 PCIE](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie) — specs, price per hour, which to rent
- [RTX PRO 6000 WS vs RTX 5090](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX PRO 6000 WS: frequently asked questions

**How much does it cost to rent an NVIDIA RTX PRO 6000 WS per hour?**

$1.040 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $1.49. Interruptible capacity costs $0.520/hr and a 3-month reservation $0.676/hr. Around $759/month if you keep one running non-stop, billed per second.

**What can a RTX PRO 6000 WS with 96 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~141B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX PRO 6000 WS available to rent right now?**

Yes — 74 GPUs across 9 machines in 9 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX PRO 6000 WS?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX PRO 6000 WS, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-pro-6000-ws --template pytorch.

**Why is the RTX PRO 6000 WS cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-pro-6000-ws · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-pro-6000-max-q -->

---
title: "Rent RTX PRO 6000 Max-Q — $0.980/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX PRO 6000 Max-Q (96 GB GDDR7) from $0.490/hr interruptible or $0.980/hr on-demand — fixed, ≥30% below market. 74 GPUs in 6 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-pro-6000-max-q
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX PRO 6000 Max-Q — 96 GB, $0.980/hr on-demand

- VRAM 96 GB GDDR7
- FP16 tensor 420 TFLOPS
- PowerScore 296 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 74 6 regions

[On-demand (guaranteed) $0.980 /GPU-hr ≈ $715/mo · market ~~$1.40~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-max-q) [Interruptible $0.490 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-max-q&type=spot) [Reserved 3 mo $0.637 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Blackwell architecture

The RTX PRO 6000 Max-Q is the 300 W variant of the Blackwell workstation flagship: the same 96 GB of GDDR7 at roughly half the power, so hosts pack more of them per chassis. Slightly lower clocks, noticeably lower price — the efficient way to rent 96 GB.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX PRO 6000 Max-Q specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX PRO 6000 Max-Q · **Architecture**: Blackwell (2025)
- **VRAM**: 96 GB GDDR7 · **Memory bandwidth**: 1,792 GB/s
- **FP16 tensor perf.**: 420 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: 24,064 · **TDP**: 300 W
- **PowerScore ((RTX 3090 = 100))**: 296 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX PRO 6000 Max-Q price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX PRO 6000 Max-Q ($1.40/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.980, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.980 | $23.52 | $715 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.490 | $11.76 | $358 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.637 | $15.29 | $465 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX PRO 6000 Max-Q (96 GB VRAM)

With 96 GB of GDDR7, a single card holds a **~32B-parameter LLM in FP16** or up to **~141B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.980.

- One-click template: [Linux Desktop on a RTX PRO 6000 Max-Q](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX PRO 6000 Max-Q](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX PRO 6000 Max-Q](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX PRO 6000 Max-Q availability by region

74 × RTX PRO 6000 Max-Q across 7 machines, live from inventory:

- Chicago, IL
- Seattle, WA
- Singapore
- Mumbai
- Stockholm
- Santiago

## RTX PRO 6000 Max-Q vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX PRO 6000 Max-Q** (this card) | 96 GB | 420 | $0.980 | $2.33‰ |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | 505 | $1.040 | $2.06‰ |
| [RTX PRO 6000 S](https://powergpu.ai/gpu/rtx-pro-6000-s) | 48 GB | 450 | $1.073 | $2.38‰ |
| [RTX PRO 5000](https://powergpu.ai/gpu/rtx-pro-5000) | 32 GB | 280 | $0.560 | $2.00‰ |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | 364 | $0.467 | $1.28‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX PRO 6000 Max-Q: frequently asked questions

**How much does it cost to rent an NVIDIA RTX PRO 6000 Max-Q per hour?**

$0.980 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $1.40. Interruptible capacity costs $0.490/hr and a 3-month reservation $0.637/hr. Around $715/month if you keep one running non-stop, billed per second.

**What can a RTX PRO 6000 Max-Q with 96 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~141B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX PRO 6000 Max-Q available to rent right now?**

Yes — 74 GPUs across 7 machines in 6 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX PRO 6000 Max-Q?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX PRO 6000 Max-Q, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-pro-6000-max-q --template pytorch.

**Why is the RTX PRO 6000 Max-Q cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-pro-6000-max-q · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/a40 -->

---
title: "Rent A40 — $0.765/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the A40 (48 GB GDDR6) from $0.382/hr interruptible or $0.765/hr on-demand — fixed, ≥30% below market. 12 GPUs in 0 regions, per-second billing."
url: https://powergpu.ai/gpu/a40
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

GPU · NVIDIA · prices checked 2026-09-14

# Rent NVIDIA A40 — 48 GB, $0.765/hr on-demand

- VRAM 48 GB GDDR6
- FP16 tensor 144 TFLOPS
- PowerScore 101 RTX 3090 = 100
- Configs 1–1× PCIe 3.0
- Online now 12 0 regions

[On-demand (guaranteed) $0.765 /GPU-hr ≈ $558/mo · market ~~$1.09~~ (−30%)](https://cloud.powergpu.ai/?gpu=a40) [Interruptible $0.382 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=a40&type=spot) [Reserved 3 mo $0.497 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

GPU · NVIDIA architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA A40 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA A40 · **Architecture**: NVIDIA
- **VRAM**: 48 GB GDDR6 · **Memory bandwidth**: —
- **FP16 tensor perf.**: 144 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: — · **TDP**: —
- **PowerScore ((RTX 3090 = 100))**: 101 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 1× · **Max instance storage**: 0 GB NVMe
- **Network up to**: 0 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## A40 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the A40 ($1.09/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.765, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.765 | $18.36 | $558 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.382 | $9.17 | $279 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.497 | $11.93 | $363 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 1× machine costs exactly 1× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a A40 (48 GB VRAM)

With 48 GB of GDDR6, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 1× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.765.

- One-click template: [Ollama on a A40](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a A40](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a A40](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## A40 availability by region

12 × A40 across 0 machines, live from inventory:

## A40 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **A40** (this card) | 48 GB | 144 | $0.765 | $5.31‰ |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | 362 | $0.514 | $1.42‰ |
| [A800 PCIE](https://powergpu.ai/gpu/a800-pcie) | 80 GB | 240 | $0.466 | $1.94‰ |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB | 312 | $0.374 | $1.20‰ |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | 181 | $0.235 | $1.30‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a A40: frequently asked questions

**How much does it cost to rent an NVIDIA A40 per hour?**

$0.765 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $1.09. Interruptible capacity costs $0.382/hr and a 3-month reservation $0.497/hr. Around $558/month if you keep one running non-stop, billed per second.

**What can a A40 with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 1× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the A40 available to rent right now?**

Yes — 12 GPUs across 0 machines in 0 regions are listed as we render this page. Configurations go from 1× to 1×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a A40?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by A40, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu a40 --template pytorch.

**Why is the A40 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/a40 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/a100-sxm4 -->

---
title: "Rent A100 SXM4 — $0.560/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the A100 SXM4 (80 GB HBM2e) from $0.280/hr interruptible or $0.560/hr on-demand — fixed, ≥30% below market. 95 GPUs in 7 regions, per-second billing."
url: https://powergpu.ai/gpu/a100-sxm4
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Ampere · launched 2020 · prices checked 2026-09-14

# Rent NVIDIA A100 SXM4 — 80 GB, $0.560/hr on-demand

- VRAM 80 GB HBM2e
- FP16 tensor 312 TFLOPS
- PowerScore 220 RTX 3090 = 100
- Configs 1–8× NVLink
- Online now 95 7 regions

[On-demand (guaranteed) $0.560 /GPU-hr ≈ $409/mo · market ~~$0.80~~ (−30%)](https://cloud.powergpu.ai/?gpu=a100-sxm4) [Interruptible $0.280 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=a100-sxm4&type=spot) [Reserved 3 mo $0.364 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Ampere architecture

The A100 SXM4 (80 GB) adds 600 GB/s NVLink and higher sustained clocks to the Ampere flagship, in 4- and 8-GPU HGX baseboards. For ≤13B pre-training, LoRA farms and FP64 HPC it still wins on total job cost against newer cards — and interruptible A100 supply is deep.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA A100 SXM4 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA A100 SXM4 · **Architecture**: Ampere (2020)
- **VRAM**: 80 GB HBM2e · **Memory bandwidth**: 2,039 GB/s
- **FP16 tensor perf.**: 312 TFLOPS · **FP32 perf.**: 19.5 TFLOPS
- **CUDA cores**: 6,912 · **TDP**: 400 W
- **PowerScore ((RTX 3090 = 100))**: 220 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · NVLink · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## A100 SXM4 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the A100 SXM4 ($0.80/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.560, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.560 | $13.44 | $409 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.280 | $6.72 | $204 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.364 | $8.74 | $266 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a A100 SXM4 (80 GB VRAM)

With 80 GB of HBM2e, a single card holds a **~32B-parameter LLM in FP16** or up to **~123B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine with NVLink for bigger models or bigger batches — the per-GPU price stays $0.560.

- Recommended for [fine-tuning](https://powergpu.ai/use-cases/fine-tuning) — QLoRA 70B on one GPU
- Recommended for [scientific computing](https://powergpu.ai/use-cases/hpc) — Up to 4.8 TB/s per GPU
- One-click template: [vLLM on a A100 SXM4](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a A100 SXM4](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a A100 SXM4](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## A100 SXM4 availability by region

95 × A100 SXM4 across 8 machines, live from inventory:

- Dallas, TX
- Singapore
- São Paulo
- London
- Stockholm
- Mumbai
- Jakarta

## A100 SXM4 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **A100 SXM4** (this card) | 80 GB | 312 | $0.560 | $1.79‰ |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | 990 | $1.428 | $1.44‰ |
| [H200](https://powergpu.ai/gpu/h200) | 141 GB | 990 | $2.791 | $2.82‰ |
| [B200](https://powergpu.ai/gpu/b200) | 192 GB | 2,250 | $5.425 | $2.41‰ |
| [B300](https://powergpu.ai/gpu/b300) | 288 GB | 2,800 | $6.737 | $2.41‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H100 SXM vs A100 SXM4](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4) — specs, price per hour, which to rent
- [A100 SXM4 vs A100 PCIE](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie) — specs, price per hour, which to rent
- [RTX 5090 vs A100 SXM4](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a A100 SXM4: frequently asked questions

**How much does it cost to rent an NVIDIA A100 SXM4 per hour?**

$0.560 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.80. Interruptible capacity costs $0.280/hr and a 3-month reservation $0.364/hr. Around $409/month if you keep one running non-stop, billed per second.

**What can a A100 SXM4 with 80 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~123B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the A100 SXM4 available to rent right now?**

Yes — 95 GPUs across 8 machines in 7 regions are listed as we render this page. Configurations go from 1× to 8× with NVLink on multi-GPU chassis. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a A100 SXM4?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by A100 SXM4, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu a100-sxm4 --template pytorch.

**Why is the A100 SXM4 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/a100-sxm4 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-pro-5000 -->

---
title: "Rent RTX PRO 5000 — $0.560/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX PRO 5000 (32 GB GDDR7) from $0.280/hr interruptible or $0.560/hr on-demand — fixed, ≥30% below market. 53 GPUs in 6 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-pro-5000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX PRO 5000 — 32 GB, $0.560/hr on-demand

- VRAM 32 GB GDDR7
- FP16 tensor 280 TFLOPS
- PowerScore 197 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 53 6 regions

[On-demand (guaranteed) $0.560 /GPU-hr ≈ $409/mo · market ~~$0.80~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-pro-5000) [Interruptible $0.280 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-pro-5000&type=spot) [Reserved 3 mo $0.364 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Blackwell architecture

The RTX PRO 5000 Blackwell is the mid-range professional card: Blackwell tensor cores, ECC GDDR7 and a 300 W envelope. A sensible rental for inference of 14B–32B models, batch image generation and viewport-heavy rendering that does not need the 96 GB flagship.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX PRO 5000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX PRO 5000 · **Architecture**: Blackwell (2025)
- **VRAM**: 32 GB GDDR7 · **Memory bandwidth**: 1,344 GB/s
- **FP16 tensor perf.**: 280 TFLOPS · **FP32 perf.**: 74.0 TFLOPS
- **CUDA cores**: 14,080 · **TDP**: 300 W
- **PowerScore ((RTX 3090 = 100))**: 197 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX PRO 5000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX PRO 5000 ($0.80/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.560, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.560 | $13.44 | $409 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.280 | $6.72 | $204 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.364 | $8.74 | $266 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX PRO 5000 (32 GB VRAM)

With 32 GB of GDDR7, a single card holds a **~13B-parameter LLM in FP16** or up to **~49B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.560.

- One-click template: [Linux Desktop on a RTX PRO 5000](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX PRO 5000](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX PRO 5000](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX PRO 5000 availability by region

53 × RTX PRO 5000 across 8 machines, live from inventory:

- London
- Los Angeles, CA
- Helsinki
- Singapore
- Montréal
- Tokyo

## RTX PRO 5000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX PRO 5000** (this card) | 32 GB | 280 | $0.560 | $2.00‰ |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | 364 | $0.467 | $1.28‰ |
| [RTX 5880Ada](https://powergpu.ai/gpu/rtx-5880ada) | 48 GB | 340 | $0.420 | $1.24‰ |
| [RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB | 190 | $0.369 | $1.94‰ |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | 155 | $0.281 | $1.81‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX PRO 5000: frequently asked questions

**How much does it cost to rent an NVIDIA RTX PRO 5000 per hour?**

$0.560 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.80. Interruptible capacity costs $0.280/hr and a 3-month reservation $0.364/hr. Around $409/month if you keep one running non-stop, billed per second.

**What can a RTX PRO 5000 with 32 GB VRAM run?**

In LLM terms, roughly a 13B-parameter model in FP16 or up to ~49B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX PRO 5000 available to rent right now?**

Yes — 53 GPUs across 8 machines in 6 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX PRO 5000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX PRO 5000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-pro-5000 --template pytorch.

**Why is the RTX PRO 5000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-pro-5000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/l40s -->

---
title: "Rent L40S — $0.514/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the L40S (48 GB GDDR6) from $0.257/hr interruptible or $0.514/hr on-demand — fixed, ≥30% below market. 107 GPUs in 8 regions, per-second billing."
url: https://powergpu.ai/gpu/l40s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA L40S — 48 GB, $0.514/hr on-demand

- VRAM 48 GB GDDR6
- FP16 tensor 362 TFLOPS
- PowerScore 255 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 107 8 regions

[On-demand (guaranteed) $0.514 /GPU-hr ≈ $375/mo · market ~~$0.74~~ (−30%)](https://cloud.powergpu.ai/?gpu=l40s) [Interruptible $0.257 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=l40s&type=spot) [Reserved 3 mo $0.334 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Ada Lovelace architecture

The L40S is NVIDIA's universal datacenter card for Ada: 48 GB GDDR6, 18,176 CUDA cores, FP8 support and passive cooling at 350 W. It is the go-to for 24/7 inference endpoints, SDXL/Flux serving and 32B-class LLMs — server-grade reliability at a consumer-adjacent price.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA L40S specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA L40S · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 48 GB GDDR6 · **Memory bandwidth**: 864 GB/s
- **FP16 tensor perf.**: 362 TFLOPS · **FP32 perf.**: 91.6 TFLOPS
- **CUDA cores**: 18,176 · **TDP**: 350 W
- **PowerScore ((RTX 3090 = 100))**: 255 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 5,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## L40S price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the L40S ($0.74/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.514, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.514 | $12.34 | $375 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.257 | $6.17 | $188 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.334 | $8.02 | $244 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a L40S (48 GB VRAM)

With 48 GB of GDDR6, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.514.

- Recommended for [llm inference](https://powergpu.ai/use-cases/llm-inference) — Best $/token in class
- Recommended for [image generation](https://powergpu.ai/use-cases/image-generation) — Flux dev ~2 s/image on 4090
- Recommended for [video generation](https://powergpu.ai/use-cases/video-generation) — 32–141 GB VRAM on tap
- One-click template: [vLLM on a L40S](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a L40S](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a L40S](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## L40S availability by region

107 × L40S across 10 machines, live from inventory:

- London
- Frankfurt
- Ashburn, VA
- Chicago, IL
- Sydney
- Los Angeles, CA
- Singapore
- Seattle, WA

## L40S vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **L40S** (this card) | 48 GB | 362 | $0.514 | $1.42‰ |
| [A800 PCIE](https://powergpu.ai/gpu/a800-pcie) | 80 GB | 240 | $0.466 | $1.94‰ |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB | 312 | $0.374 | $1.20‰ |
| [A40](https://powergpu.ai/gpu/a40) | 48 GB | 144 | $0.765 | $5.31‰ |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | 181 | $0.235 | $1.30‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX 4090 vs L40S](https://powergpu.ai/compare/rtx-4090-vs-l40s) — specs, price per hour, which to rent
- [L40S vs A100 PCIE](https://powergpu.ai/compare/l40s-vs-a100-pcie) — specs, price per hour, which to rent
- [L40S vs H100 PCIE](https://powergpu.ai/compare/l40s-vs-h100-pcie) — specs, price per hour, which to rent
- [RTX 5090 vs L40S](https://powergpu.ai/compare/rtx-5090-vs-l40s) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a L40S: frequently asked questions

**How much does it cost to rent an NVIDIA L40S per hour?**

$0.514 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.74. Interruptible capacity costs $0.257/hr and a 3-month reservation $0.334/hr. Around $375/month if you keep one running non-stop, billed per second.

**What can a L40S with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the L40S available to rent right now?**

Yes — 107 GPUs across 10 machines in 8 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a L40S?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by L40S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu l40s --template pytorch.

**Why is the L40S cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/l40s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4090d -->

---
title: "Rent RTX 4090D — $0.467/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4090D (24 GB GDDR6X) from $0.233/hr interruptible or $0.467/hr on-demand — fixed, ≥30% below market. 7 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4090d
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer flagship · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA RTX 4090D — 24 GB, $0.467/hr on-demand

- VRAM 24 GB GDDR6X
- FP16 tensor 296 TFLOPS
- PowerScore 208 RTX 3090 = 100
- Configs 1–2× PCIe 4.0
- Online now 7 2 regions

[On-demand (guaranteed) $0.467 /GPU-hr ≈ $341/mo · market ~~$0.67~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4090d) [Interruptible $0.233 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4090d&type=spot) [Reserved 3 mo $0.303 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer flagship · Ada Lovelace architecture

The RTX 4090 D is the export-market variant of the 4090 with fewer CUDA cores (14,592) and the same 24 GB GDDR6X. In practice it lands within 5–10% of a 4090 on diffusion and inference, so rent it whenever it is cheaper or more available than the standard card.

The community favourite: consumer pricing with serious tensor throughput. Ideal for diffusion models, quantized LLMs and fine-tuning runs that fit in 24 GB. Supply is deep, so interruptible capacity is almost always available at half price.

## NVIDIA RTX 4090D specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4090D · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 24 GB GDDR6X · **Memory bandwidth**: 1,008 GB/s
- **FP16 tensor perf.**: 296 TFLOPS · **FP32 perf.**: 73.5 TFLOPS
- **CUDA cores**: 14,592 · **TDP**: 425 W
- **PowerScore ((RTX 3090 = 100))**: 208 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 2× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 750 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4090D price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4090D ($0.67/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.467, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.467 | $11.21 | $341 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.233 | $5.59 | $170 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.303 | $7.27 | $221 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 2× machine costs exactly 2× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4090D (24 GB VRAM)

With 24 GB of GDDR6X, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 2× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.467.

- One-click template: [ComfyUI on a RTX 4090D](https://powergpu.ai/templates/comfyui)
- One-click template: [Ollama on a RTX 4090D](https://powergpu.ai/templates/ollama)
- One-click template: [Kohya's GUI on a RTX 4090D](https://powergpu.ai/templates/kohya-s-gui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4090D availability by region

7 × RTX 4090D across 2 machines, live from inventory:

- Bucharest
- Miami, FL

## RTX 4090D vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4090D** (this card) | 24 GB | 296 | $0.467 | $1.58‰ |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | 330 | $0.327 | $0.99‰ |
| [RTX 4080](https://powergpu.ai/gpu/rtx-4080) | 16 GB | 195 | $0.168 | $0.86‰ |
| [RTX 4080S](https://powergpu.ai/gpu/rtx-4080s) | 16 GB | 208 | $0.127 | $0.61‰ |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | 176 | $0.103 | $0.59‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4090D: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4090D per hour?**

$0.467 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.67. Interruptible capacity costs $0.233/hr and a 3-month reservation $0.303/hr. Around $341/month if you keep one running non-stop, billed per second.

**What can a RTX 4090D with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 2× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4090D available to rent right now?**

Yes — 7 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 2×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4090D?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4090D, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4090d --template pytorch.

**Why is the RTX 4090D cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4090d · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-6000ada -->

---
title: "Rent RTX 6000Ada — $0.467/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 6000Ada (48 GB GDDR6) from $0.233/hr interruptible or $0.467/hr on-demand — fixed, ≥30% below market. 34 GPUs in 4 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-6000ada
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ada Lovelace · launched 2022 · prices checked 2026-09-14

# Rent NVIDIA RTX 6000Ada — 48 GB, $0.467/hr on-demand

- VRAM 48 GB GDDR6
- FP16 tensor 364 TFLOPS
- PowerScore 256 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 34 4 regions

[On-demand (guaranteed) $0.467 /GPU-hr ≈ $341/mo · market ~~$0.67~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-6000ada) [Interruptible $0.233 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-6000ada&type=spot) [Reserved 3 mo $0.303 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ada Lovelace architecture

The RTX 6000 Ada is the Ada Lovelace workstation flagship: 48 GB GDDR6 with ECC, 18,176 CUDA cores, 300 W. Renderers and CAD teams rent it for production scenes; ML teams for 32B-class inference and LoRA training where 48 GB and studio drivers matter more than HBM.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX 6000Ada specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 6000Ada · **Architecture**: Ada Lovelace (2022)
- **VRAM**: 48 GB GDDR6 · **Memory bandwidth**: 960 GB/s
- **FP16 tensor perf.**: 364 TFLOPS · **FP32 perf.**: 91.1 TFLOPS
- **CUDA cores**: 18,176 · **TDP**: 300 W
- **PowerScore ((RTX 3090 = 100))**: 256 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 6000Ada price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 6000Ada ($0.67/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.467, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.467 | $11.21 | $341 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.233 | $5.59 | $170 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.303 | $7.27 | $221 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 6000Ada (48 GB VRAM)

With 48 GB of GDDR6, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.467.

- Recommended for [3d rendering](https://powergpu.ai/use-cases/rendering) — Per-second render billing
- One-click template: [Linux Desktop on a RTX 6000Ada](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX 6000Ada](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX 6000Ada](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 6000Ada availability by region

34 × RTX 6000Ada across 6 machines, live from inventory:

- Tokyo
- Mumbai
- Seattle, WA
- Miami, FL

## RTX 6000Ada vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 6000Ada** (this card) | 48 GB | 364 | $0.467 | $1.28‰ |
| [RTX 5880Ada](https://powergpu.ai/gpu/rtx-5880ada) | 48 GB | 340 | $0.420 | $1.24‰ |
| [RTX PRO 5000](https://powergpu.ai/gpu/rtx-pro-5000) | 32 GB | 280 | $0.560 | $2.00‰ |
| [RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB | 190 | $0.369 | $1.94‰ |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | 155 | $0.281 | $1.81‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX PRO 6000 WS vs RTX 6000Ada](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada) — specs, price per hour, which to rent
- [RTX 6000Ada vs RTX A6000](https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX 6000Ada: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 6000Ada per hour?**

$0.467 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.67. Interruptible capacity costs $0.233/hr and a 3-month reservation $0.303/hr. Around $341/month if you keep one running non-stop, billed per second.

**What can a RTX 6000Ada with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 6000Ada available to rent right now?**

Yes — 34 GPUs across 6 machines in 4 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 6000Ada?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 6000Ada, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-6000ada --template pytorch.

**Why is the RTX 6000Ada cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-6000ada · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/a800-pcie -->

---
title: "Rent A800 PCIE — $0.466/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the A800 PCIE (80 GB GDDR6) from $0.233/hr interruptible or $0.466/hr on-demand — fixed, ≥30% below market. 10 GPUs in 0 regions, per-second billing."
url: https://powergpu.ai/gpu/a800-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

GPU · NVIDIA · prices checked 2026-09-14

# Rent NVIDIA A800 PCIE — 80 GB, $0.466/hr on-demand

- VRAM 80 GB GDDR6
- FP16 tensor 240 TFLOPS
- PowerScore 169 RTX 3090 = 100
- Configs 1–1× PCIe 3.0
- Online now 10 0 regions

[On-demand (guaranteed) $0.466 /GPU-hr ≈ $340/mo · market ~~$0.67~~ (−30%)](https://cloud.powergpu.ai/?gpu=a800-pcie) [Interruptible $0.233 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=a800-pcie&type=spot) [Reserved 3 mo $0.302 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

GPU · NVIDIA architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA A800 PCIE specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA A800 PCIE · **Architecture**: NVIDIA
- **VRAM**: 80 GB GDDR6 · **Memory bandwidth**: —
- **FP16 tensor perf.**: 240 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: — · **TDP**: —
- **PowerScore ((RTX 3090 = 100))**: 169 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 1× · **Max instance storage**: 0 GB NVMe
- **Network up to**: 0 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## A800 PCIE price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the A800 PCIE ($0.67/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.466, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.466 | $11.18 | $340 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.233 | $5.59 | $170 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.302 | $7.25 | $220 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 1× machine costs exactly 1× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a A800 PCIE (80 GB VRAM)

With 80 GB of GDDR6, a single card holds a **~32B-parameter LLM in FP16** or up to **~123B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 1× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.466.

- One-click template: [Ollama on a A800 PCIE](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a A800 PCIE](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a A800 PCIE](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## A800 PCIE availability by region

10 × A800 PCIE across 0 machines, live from inventory:

## A800 PCIE vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **A800 PCIE** (this card) | 80 GB | 240 | $0.466 | $1.94‰ |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | 362 | $0.514 | $1.42‰ |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB | 312 | $0.374 | $1.20‰ |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | 181 | $0.235 | $1.30‰ |
| [L4](https://powergpu.ai/gpu/l4) | 24 GB | 121 | $0.225 | $1.86‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a A800 PCIE: frequently asked questions

**How much does it cost to rent an NVIDIA A800 PCIE per hour?**

$0.466 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.67. Interruptible capacity costs $0.233/hr and a 3-month reservation $0.302/hr. Around $340/month if you keep one running non-stop, billed per second.

**What can a A800 PCIE with 80 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~123B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 1× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the A800 PCIE available to rent right now?**

Yes — 10 GPUs across 0 machines in 0 regions are listed as we render this page. Configurations go from 1× to 1×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a A800 PCIE?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by A800 PCIE, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu a800-pcie --template pytorch.

**Why is the A800 PCIE cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/a800-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5090 -->

---
title: "Rent RTX 5090 — $0.439/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5090 (32 GB GDDR7) from $0.219/hr interruptible or $0.439/hr on-demand — fixed, ≥30% below market. 922 GPUs in 22 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer flagship · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX 5090 — 32 GB, $0.439/hr on-demand

- VRAM 32 GB GDDR7
- FP16 tensor 419 TFLOPS
- PowerScore 295 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 922 22 regions

[On-demand (guaranteed) $0.439 /GPU-hr ≈ $320/mo · market ~~$0.63~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5090) [Interruptible $0.219 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5090&type=spot) [Reserved 3 mo $0.285 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer flagship · Blackwell architecture

The RTX 5090 is the Blackwell consumer flagship: 32 GB GDDR7 at 1.79 TB/s, 21,760 CUDA cores and native FP4 in fifth-generation tensor cores. It is the best dollars-per-token card on the sheet for 7B–32B models and the entry point for video diffusion; supply is the deepest in the catalogue.

The community favourite: consumer pricing with serious tensor throughput. Ideal for diffusion models, quantized LLMs and fine-tuning runs that fit in 32 GB. Supply is deep, so interruptible capacity is almost always available at half price.

## NVIDIA RTX 5090 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5090 · **Architecture**: Blackwell (2025)
- **VRAM**: 32 GB GDDR7 · **Memory bandwidth**: 1,792 GB/s
- **FP16 tensor perf.**: 419 TFLOPS · **FP32 perf.**: 104.8 TFLOPS
- **CUDA cores**: 21,760 · **TDP**: 575 W
- **PowerScore ((RTX 3090 = 100))**: 295 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5090 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5090 ($0.63/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.439, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.439 | $10.54 | $320 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.219 | $5.26 | $160 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.285 | $6.84 | $208 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5090 (32 GB VRAM)

With 32 GB of GDDR7, a single card holds a **~13B-parameter LLM in FP16** or up to **~49B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.439.

- Recommended for [llm inference](https://powergpu.ai/use-cases/llm-inference) — Best $/token in class
- Recommended for [fine-tuning](https://powergpu.ai/use-cases/fine-tuning) — QLoRA 70B on one GPU
- Recommended for [image generation](https://powergpu.ai/use-cases/image-generation) — Flux dev ~2 s/image on 4090
- Recommended for [video generation](https://powergpu.ai/use-cases/video-generation) — 32–141 GB VRAM on tap
- One-click template: [ComfyUI on a RTX 5090](https://powergpu.ai/templates/comfyui)
- One-click template: [Ollama on a RTX 5090](https://powergpu.ai/templates/ollama)
- One-click template: [Kohya's GUI on a RTX 5090](https://powergpu.ai/templates/kohya-s-gui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5090 availability by region

922 × RTX 5090 across 30 machines, live from inventory:

- São Paulo
- Amsterdam
- London
- Stockholm
- Ashburn, VA
- Tel Aviv
- Milan
- Vancouver
- Dallas, TX
- Paris
- Sydney
- Helsinki
- +10 more

## RTX 5090 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5090** (this card) | 32 GB | 419 | $0.439 | $1.05‰ |
| [RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB | 225 | $0.186 | $0.83‰ |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB | 176 | $0.131 | $0.74‰ |
| [RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB | 92 | $0.112 | $1.22‰ |
| [RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB | 123 | $0.112 | $0.91‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX 5090 vs RTX 4090](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090) — specs, price per hour, which to rent
- [RTX 5090 vs L40S](https://powergpu.ai/compare/rtx-5090-vs-l40s) — specs, price per hour, which to rent
- [RTX 5090 vs A100 SXM4](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4) — specs, price per hour, which to rent
- [RTX PRO 6000 WS vs RTX 5090](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX 5090: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5090 per hour?**

$0.439 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.63. Interruptible capacity costs $0.219/hr and a 3-month reservation $0.285/hr. Around $320/month if you keep one running non-stop, billed per second.

**What can a RTX 5090 with 32 GB VRAM run?**

In LLM terms, roughly a 13B-parameter model in FP16 or up to ~49B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5090 available to rent right now?**

Yes — 922 GPUs across 30 machines in 22 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5090?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5090, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5090 --template pytorch.

**Why is the RTX 5090 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5880ada -->

---
title: "Rent RTX 5880Ada — $0.420/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5880Ada (48 GB GDDR6) from $0.210/hr interruptible or $0.420/hr on-demand — fixed, ≥30% below market. 18 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5880ada
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ada Lovelace · launched 2024 · prices checked 2026-09-14

# Rent NVIDIA RTX 5880Ada — 48 GB, $0.420/hr on-demand

- VRAM 48 GB GDDR6
- FP16 tensor 340 TFLOPS
- PowerScore 239 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 18 2 regions

[On-demand (guaranteed) $0.420 /GPU-hr ≈ $307/mo · market ~~$0.60~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5880ada) [Interruptible $0.210 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5880ada&type=spot) [Reserved 3 mo $0.273 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ada Lovelace architecture

The RTX 5880 Ada sits between the RTX 5000 Ada and RTX 6000 Ada: 48 GB ECC GDDR6 with 14,080 CUDA cores. It offers RTX 6000 Ada memory capacity at a lower rate — a quiet bargain for VRAM-bound rendering and inference.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX 5880Ada specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5880Ada · **Architecture**: Ada Lovelace (2024)
- **VRAM**: 48 GB GDDR6 · **Memory bandwidth**: 960 GB/s
- **FP16 tensor perf.**: 340 TFLOPS · **FP32 perf.**: 69.3 TFLOPS
- **CUDA cores**: 14,080 · **TDP**: 285 W
- **PowerScore ((RTX 3090 = 100))**: 239 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5880Ada price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5880Ada ($0.60/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.420, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.420 | $10.08 | $307 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.210 | $5.04 | $153 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.273 | $6.55 | $199 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5880Ada (48 GB VRAM)

With 48 GB of GDDR6, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.420.

- One-click template: [Linux Desktop on a RTX 5880Ada](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX 5880Ada](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX 5880Ada](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5880Ada availability by region

18 × RTX 5880Ada across 2 machines, live from inventory:

- Singapore
- Seattle, WA

## RTX 5880Ada vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5880Ada** (this card) | 48 GB | 340 | $0.420 | $1.24‰ |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | 364 | $0.467 | $1.28‰ |
| [RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB | 190 | $0.369 | $1.94‰ |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | 155 | $0.281 | $1.81‰ |
| [RTX 4500Ada](https://powergpu.ai/gpu/rtx-4500ada) | 24 GB | 72 | $0.280 | $3.89‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 5880Ada: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5880Ada per hour?**

$0.420 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.60. Interruptible capacity costs $0.210/hr and a 3-month reservation $0.273/hr. Around $307/month if you keep one running non-stop, billed per second.

**What can a RTX 5880Ada with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5880Ada available to rent right now?**

Yes — 18 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5880Ada?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5880Ada, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5880ada --template pytorch.

**Why is the RTX 5880Ada cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5880ada · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/a100-pcie -->

---
title: "Rent A100 PCIE — $0.374/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the A100 PCIE (80 GB HBM2e) from $0.187/hr interruptible or $0.374/hr on-demand — fixed, ≥30% below market. 46 GPUs in 9 regions, per-second billing."
url: https://powergpu.ai/gpu/a100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA A100 PCIE — 80 GB, $0.374/hr on-demand

- VRAM 80 GB HBM2e
- FP16 tensor 312 TFLOPS
- PowerScore 220 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 46 9 regions

[On-demand (guaranteed) $0.374 /GPU-hr ≈ $273/mo · market ~~$0.54~~ (−30%)](https://cloud.powergpu.ai/?gpu=a100-pcie) [Interruptible $0.187 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=a100-pcie&type=spot) [Reserved 3 mo $0.243 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Ampere architecture

The A100 PCIe (80 GB) is the Ampere datacenter card in a standard slot: 80 GB HBM2e, FP64 tensor cores, MIG support, 300 W. Years of software maturity make it the safest choice for production inference and mid-size training, and it is usually the cheapest 80 GB card on the sheet.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA A100 PCIE specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA A100 PCIE · **Architecture**: Ampere (2021)
- **VRAM**: 80 GB HBM2e · **Memory bandwidth**: 1,935 GB/s
- **FP16 tensor perf.**: 312 TFLOPS · **FP32 perf.**: 19.5 TFLOPS
- **CUDA cores**: 6,912 · **TDP**: 300 W
- **PowerScore ((RTX 3090 = 100))**: 220 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## A100 PCIE price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the A100 PCIE ($0.54/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.374, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.374 | $8.98 | $273 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.187 | $4.49 | $137 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.243 | $5.83 | $177 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a A100 PCIE (80 GB VRAM)

With 80 GB of HBM2e, a single card holds a **~32B-parameter LLM in FP16** or up to **~123B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.374.

- One-click template: [vLLM on a A100 PCIE](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a A100 PCIE](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a A100 PCIE](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## A100 PCIE availability by region

46 × A100 PCIE across 10 machines, live from inventory:

- London
- Singapore
- Montréal
- Dallas, TX
- Madrid
- Ashburn, VA
- Mumbai
- Amsterdam
- Seattle, WA

## A100 PCIE vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **A100 PCIE** (this card) | 80 GB | 312 | $0.374 | $1.20‰ |
| [A800 PCIE](https://powergpu.ai/gpu/a800-pcie) | 80 GB | 240 | $0.466 | $1.94‰ |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | 181 | $0.235 | $1.30‰ |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | 362 | $0.514 | $1.42‰ |
| [L4](https://powergpu.ai/gpu/l4) | 24 GB | 121 | $0.225 | $1.86‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [H100 PCIE vs A100 PCIE](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie) — specs, price per hour, which to rent
- [A100 SXM4 vs A100 PCIE](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie) — specs, price per hour, which to rent
- [RTX 4090 vs A100 PCIE](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie) — specs, price per hour, which to rent
- [L40S vs A100 PCIE](https://powergpu.ai/compare/l40s-vs-a100-pcie) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a A100 PCIE: frequently asked questions

**How much does it cost to rent an NVIDIA A100 PCIE per hour?**

$0.374 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.54. Interruptible capacity costs $0.187/hr and a 3-month reservation $0.243/hr. Around $273/month if you keep one running non-stop, billed per second.

**What can a A100 PCIE with 80 GB VRAM run?**

In LLM terms, roughly a 32B-parameter model in FP16 or up to ~123B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the A100 PCIE available to rent right now?**

Yes — 46 GPUs across 10 machines in 9 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a A100 PCIE?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by A100 PCIE, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu a100-pcie --template pytorch.

**Why is the A100 PCIE cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/a100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5000ada -->

---
title: "Rent RTX 5000Ada — $0.369/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5000Ada (32 GB GDDR6) from $0.184/hr interruptible or $0.369/hr on-demand — fixed, ≥30% below market. 7 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5000ada
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA RTX 5000Ada — 32 GB, $0.369/hr on-demand

- VRAM 32 GB GDDR6
- FP16 tensor 190 TFLOPS
- PowerScore 134 RTX 3090 = 100
- Configs 1–4× PCIe 5.0
- Online now 7 2 regions

[On-demand (guaranteed) $0.369 /GPU-hr ≈ $269/mo · market ~~$0.53~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5000ada) [Interruptible $0.184 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5000ada&type=spot) [Reserved 3 mo $0.239 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ada Lovelace architecture

The RTX 5000 Ada brings 32 GB of ECC GDDR6 and 12,800 CUDA cores at 250 W. For studios it is the workstation sweet spot; for ML it runs 14B FP16 or 32B 4-bit models comfortably with professional-driver stability.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX 5000Ada specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5000Ada · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 32 GB GDDR6 · **Memory bandwidth**: 576 GB/s
- **FP16 tensor perf.**: 190 TFLOPS · **FP32 perf.**: 65.3 TFLOPS
- **CUDA cores**: 12,800 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 134 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5000Ada price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5000Ada ($0.53/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.369, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.369 | $8.86 | $269 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.184 | $4.42 | $134 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.239 | $5.74 | $174 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5000Ada (32 GB VRAM)

With 32 GB of GDDR6, a single card holds a **~13B-parameter LLM in FP16** or up to **~49B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.369.

- One-click template: [Linux Desktop on a RTX 5000Ada](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX 5000Ada](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX 5000Ada](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5000Ada availability by region

7 × RTX 5000Ada across 2 machines, live from inventory:

- Seattle, WA
- Montréal

## RTX 5000Ada vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5000Ada** (this card) | 32 GB | 190 | $0.369 | $1.94‰ |
| [RTX 5880Ada](https://powergpu.ai/gpu/rtx-5880ada) | 48 GB | 340 | $0.420 | $1.24‰ |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | 155 | $0.281 | $1.81‰ |
| [RTX 4500Ada](https://powergpu.ai/gpu/rtx-4500ada) | 24 GB | 72 | $0.280 | $3.89‰ |
| [RTX PRO 4500](https://powergpu.ai/gpu/rtx-pro-4500) | 24 GB | 200 | $0.273 | $1.37‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 5000Ada: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5000Ada per hour?**

$0.369 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.53. Interruptible capacity costs $0.184/hr and a 3-month reservation $0.239/hr. Around $269/month if you keep one running non-stop, billed per second.

**What can a RTX 5000Ada with 32 GB VRAM run?**

In LLM terms, roughly a 13B-parameter model in FP16 or up to ~49B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5000Ada available to rent right now?**

Yes — 7 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5000Ada?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5000Ada, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5000ada --template pytorch.

**Why is the RTX 5000Ada cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5000ada · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4090 -->

---
title: "Rent RTX 4090 — $0.327/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4090 (24 GB GDDR6X) from $0.163/hr interruptible or $0.327/hr on-demand — fixed, ≥30% below market. 461 GPUs in 21 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer flagship · Ada Lovelace · launched 2022 · prices checked 2026-09-14

# Rent NVIDIA RTX 4090 — 24 GB, $0.327/hr on-demand

- VRAM 24 GB GDDR6X
- FP16 tensor 330 TFLOPS
- PowerScore 232 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 461 21 regions

[On-demand (guaranteed) $0.327 /GPU-hr ≈ $239/mo · market ~~$0.47~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4090) [Interruptible $0.163 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4090&type=spot) [Reserved 3 mo $0.212 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer flagship · Ada Lovelace architecture

The RTX 4090 remains the community default for AI work: 24 GB GDDR6X, 16,384 CUDA cores, Ada tensor cores and the largest ecosystem of tuned kernels and ComfyUI workflows. Interruptible 4090s are the cheapest serious GPU-hours you can rent for Stable Diffusion, Flux and quantized LLMs.

The community favourite: consumer pricing with serious tensor throughput. Ideal for diffusion models, quantized LLMs and fine-tuning runs that fit in 24 GB. Supply is deep, so interruptible capacity is almost always available at half price.

## NVIDIA RTX 4090 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4090 · **Architecture**: Ada Lovelace (2022)
- **VRAM**: 24 GB GDDR6X · **Memory bandwidth**: 1,008 GB/s
- **FP16 tensor perf.**: 330 TFLOPS · **FP32 perf.**: 82.6 TFLOPS
- **CUDA cores**: 16,384 · **TDP**: 450 W
- **PowerScore ((RTX 3090 = 100))**: 232 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4090 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4090 ($0.47/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.327, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.327 | $7.85 | $239 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.163 | $3.91 | $119 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.212 | $5.09 | $155 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4090 (24 GB VRAM)

With 24 GB of GDDR6X, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.327.

- Recommended for [image generation](https://powergpu.ai/use-cases/image-generation) — Flux dev ~2 s/image on 4090
- Recommended for [3d rendering](https://powergpu.ai/use-cases/rendering) — Per-second render billing
- Recommended for [computer vision](https://powergpu.ai/use-cases/computer-vision) — YOLO11 epochs from $0.09
- One-click template: [ComfyUI on a RTX 4090](https://powergpu.ai/templates/comfyui)
- One-click template: [Ollama on a RTX 4090](https://powergpu.ai/templates/ollama)
- One-click template: [Kohya's GUI on a RTX 4090](https://powergpu.ai/templates/kohya-s-gui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4090 availability by region

461 × RTX 4090 across 30 machines, live from inventory:

- Madrid
- Milan
- Querétaro
- Vancouver
- Ashburn, VA
- Johannesburg
- Tokyo
- Paris
- Warsaw
- Dubai
- Singapore
- Tel Aviv
- +9 more

## RTX 4090 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4090** (this card) | 24 GB | 330 | $0.327 | $0.99‰ |
| [RTX 4090D](https://powergpu.ai/gpu/rtx-4090d) | 24 GB | 296 | $0.467 | $1.58‰ |
| [RTX 4080](https://powergpu.ai/gpu/rtx-4080) | 16 GB | 195 | $0.168 | $0.86‰ |
| [RTX 4080S](https://powergpu.ai/gpu/rtx-4080s) | 16 GB | 208 | $0.127 | $0.61‰ |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | 176 | $0.103 | $0.59‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX 5090 vs RTX 4090](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090) — specs, price per hour, which to rent
- [RTX 4090 vs RTX 3090](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090) — specs, price per hour, which to rent
- [RTX 4090 vs A100 PCIE](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie) — specs, price per hour, which to rent
- [RTX 4090 vs L40S](https://powergpu.ai/compare/rtx-4090-vs-l40s) — specs, price per hour, which to rent
- [RTX A6000 vs RTX 4090](https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090) — specs, price per hour, which to rent
- [RTX 5080 vs RTX 4090](https://powergpu.ai/compare/rtx-5080-vs-rtx-4090) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX 4090: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4090 per hour?**

$0.327 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.47. Interruptible capacity costs $0.163/hr and a 3-month reservation $0.212/hr. Around $239/month if you keep one running non-stop, billed per second.

**What can a RTX 4090 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4090 available to rent right now?**

Yes — 461 GPUs across 30 machines in 21 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4090?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4090, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4090 --template pytorch.

**Why is the RTX 4090 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-a6000 -->

---
title: "Rent RTX A6000 — $0.281/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX A6000 (48 GB GDDR6) from $0.140/hr interruptible or $0.281/hr on-demand — fixed, ≥30% below market. 12 GPUs in 5 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-a6000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ampere · launched 2020 · prices checked 2026-09-14

# Rent NVIDIA RTX A6000 — 48 GB, $0.281/hr on-demand

- VRAM 48 GB GDDR6
- FP16 tensor 155 TFLOPS
- PowerScore 109 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 12 5 regions

[On-demand (guaranteed) $0.281 /GPU-hr ≈ $205/mo · market ~~$0.40~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-a6000) [Interruptible $0.140 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-a6000&type=spot) [Reserved 3 mo $0.182 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ampere architecture

The RTX A6000 is the Ampere workstation flagship: 48 GB GDDR6 with ECC, 10,752 CUDA cores, NVLink bridge support, 300 W. It is the cheapest 48 GB card on the sheet — ideal for 32B-class inference, LoRA training and render jobs that outgrow 24 GB.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX A6000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX A6000 · **Architecture**: Ampere (2020)
- **VRAM**: 48 GB GDDR6 · **Memory bandwidth**: 768 GB/s
- **FP16 tensor perf.**: 155 TFLOPS · **FP32 perf.**: 38.7 TFLOPS
- **CUDA cores**: 10,752 · **TDP**: 300 W
- **PowerScore ((RTX 3090 = 100))**: 109 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX A6000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX A6000 ($0.40/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.281, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.281 | $6.74 | $205 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.140 | $3.36 | $102 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.182 | $4.37 | $133 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX A6000 (48 GB VRAM)

With 48 GB of GDDR6, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.281.

- One-click template: [Linux Desktop on a RTX A6000](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX A6000](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX A6000](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX A6000 availability by region

12 × RTX A6000 across 5 machines, live from inventory:

- Miami, FL
- Chicago, IL
- Osaka
- New York, NY
- Ashburn, VA

## RTX A6000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX A6000** (this card) | 48 GB | 155 | $0.281 | $1.81‰ |
| [RTX 4500Ada](https://powergpu.ai/gpu/rtx-4500ada) | 24 GB | 72 | $0.280 | $3.89‰ |
| [RTX PRO 4500](https://powergpu.ai/gpu/rtx-pro-4500) | 24 GB | 200 | $0.273 | $1.37‰ |
| [RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB | 190 | $0.369 | $1.94‰ |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB | 150 | $0.183 | $1.22‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX 6000Ada vs RTX A6000](https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000) — specs, price per hour, which to rent
- [RTX A6000 vs RTX 4090](https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX A6000: frequently asked questions

**How much does it cost to rent an NVIDIA RTX A6000 per hour?**

$0.281 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.40. Interruptible capacity costs $0.140/hr and a 3-month reservation $0.182/hr. Around $205/month if you keep one running non-stop, billed per second.

**What can a RTX A6000 with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX A6000 available to rent right now?**

Yes — 12 GPUs across 5 machines in 5 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX A6000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX A6000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-a6000 --template pytorch.

**Why is the RTX A6000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-a6000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4500ada -->

---
title: "Rent RTX 4500Ada — $0.280/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4500Ada (24 GB GDDR6) from $0.140/hr interruptible or $0.280/hr on-demand — fixed, ≥30% below market. 10 GPUs in 0 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4500ada
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ada Lovelace · prices checked 2026-09-14

# Rent NVIDIA RTX 4500Ada — 24 GB, $0.280/hr on-demand

- VRAM 24 GB GDDR6
- FP16 tensor 72 TFLOPS
- PowerScore 51 RTX 3090 = 100
- Configs 1–1× PCIe 3.0
- Online now 10 0 regions

[On-demand (guaranteed) $0.280 /GPU-hr ≈ $204/mo · market ~~$0.40~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4500ada) [Interruptible $0.140 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4500ada&type=spot) [Reserved 3 mo $0.182 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ada Lovelace architecture

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX 4500Ada specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4500Ada · **Architecture**: Ada Lovelace
- **VRAM**: 24 GB GDDR6 · **Memory bandwidth**: —
- **FP16 tensor perf.**: 72 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: — · **TDP**: —
- **PowerScore ((RTX 3090 = 100))**: 51 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 1× · **Max instance storage**: 0 GB NVMe
- **Network up to**: 0 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4500Ada price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4500Ada ($0.40/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.280, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.280 | $6.72 | $204 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.140 | $3.36 | $102 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.182 | $4.37 | $133 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 1× machine costs exactly 1× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4500Ada (24 GB VRAM)

With 24 GB of GDDR6, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 1× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.280.

- One-click template: [Linux Desktop on a RTX 4500Ada](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX 4500Ada](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX 4500Ada](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4500Ada availability by region

10 × RTX 4500Ada across 0 machines, live from inventory:

## RTX 4500Ada vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4500Ada** (this card) | 24 GB | 72 | $0.280 | $3.89‰ |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | 155 | $0.281 | $1.81‰ |
| [RTX PRO 4500](https://powergpu.ai/gpu/rtx-pro-4500) | 24 GB | 200 | $0.273 | $1.37‰ |
| [RTX 5000Ada](https://powergpu.ai/gpu/rtx-5000ada) | 32 GB | 190 | $0.369 | $1.94‰ |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB | 150 | $0.183 | $1.22‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4500Ada: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4500Ada per hour?**

$0.280 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.40. Interruptible capacity costs $0.140/hr and a 3-month reservation $0.182/hr. Around $204/month if you keep one running non-stop, billed per second.

**What can a RTX 4500Ada with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 1× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4500Ada available to rent right now?**

Yes — 10 GPUs across 0 machines in 0 regions are listed as we render this page. Configurations go from 1× to 1×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4500Ada?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4500Ada, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4500ada --template pytorch.

**Why is the RTX 4500Ada cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4500ada · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-pro-4500 -->

---
title: "Rent RTX PRO 4500 — $0.273/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX PRO 4500 (24 GB GDDR7) from $0.136/hr interruptible or $0.273/hr on-demand — fixed, ≥30% below market. 4 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-pro-4500
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX PRO 4500 — 24 GB, $0.273/hr on-demand

- VRAM 24 GB GDDR7
- FP16 tensor 200 TFLOPS
- PowerScore 141 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 4 2 regions

[On-demand (guaranteed) $0.273 /GPU-hr ≈ $199/mo · market ~~$0.39~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-pro-4500) [Interruptible $0.136 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-pro-4500&type=spot) [Reserved 3 mo $0.177 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Blackwell architecture

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX PRO 4500 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX PRO 4500 · **Architecture**: Blackwell (2025)
- **VRAM**: 24 GB GDDR7 · **Memory bandwidth**: 896 GB/s
- **FP16 tensor perf.**: 200 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: 10,496 · **TDP**: 200 W
- **PowerScore ((RTX 3090 = 100))**: 141 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX PRO 4500 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX PRO 4500 ($0.39/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.273, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.273 | $6.55 | $199 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.136 | $3.26 | $99 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.177 | $4.25 | $129 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX PRO 4500 (24 GB VRAM)

With 24 GB of GDDR7, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.273.

- One-click template: [Linux Desktop on a RTX PRO 4500](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX PRO 4500](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX PRO 4500](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX PRO 4500 availability by region

4 × RTX PRO 4500 across 2 machines, live from inventory:

- Singapore
- Dallas, TX

## RTX PRO 4500 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX PRO 4500** (this card) | 24 GB | 200 | $0.273 | $1.37‰ |
| [RTX 4500Ada](https://powergpu.ai/gpu/rtx-4500ada) | 24 GB | 72 | $0.280 | $3.89‰ |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | 155 | $0.281 | $1.81‰ |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB | 150 | $0.183 | $1.22‰ |
| [Q RTX 8000](https://powergpu.ai/gpu/q-rtx-8000) | 48 GB | 65 | $0.178 | $2.74‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX PRO 4500: frequently asked questions

**How much does it cost to rent an NVIDIA RTX PRO 4500 per hour?**

$0.273 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.39. Interruptible capacity costs $0.136/hr and a 3-month reservation $0.177/hr. Around $199/month if you keep one running non-stop, billed per second.

**What can a RTX PRO 4500 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX PRO 4500 available to rent right now?**

Yes — 4 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX PRO 4500?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX PRO 4500, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-pro-4500 --template pytorch.

**Why is the RTX PRO 4500 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-pro-4500 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/l40 -->

---
title: "Rent L40 — $0.235/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the L40 (48 GB GDDR6) from $0.117/hr interruptible or $0.235/hr on-demand — fixed, ≥30% below market. 4 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/l40
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Ada Lovelace · launched 2022 · prices checked 2026-09-14

# Rent NVIDIA L40 — 48 GB, $0.235/hr on-demand

- VRAM 48 GB GDDR6
- FP16 tensor 181 TFLOPS
- PowerScore 127 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 4 2 regions

[On-demand (guaranteed) $0.235 /GPU-hr ≈ $172/mo · market ~~$0.34~~ (−30%)](https://cloud.powergpu.ai/?gpu=l40) [Interruptible $0.117 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=l40&type=spot) [Reserved 3 mo $0.152 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Ada Lovelace architecture

The L40 is the datacenter Ada card built for visualisation and inference: 48 GB GDDR6, 18,176 CUDA cores, passively cooled at 300 W. It shares silicon with the L40S at slightly lower clocks, making it a cheaper 48 GB option for render farms and SDXL serving.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA L40 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA L40 · **Architecture**: Ada Lovelace (2022)
- **VRAM**: 48 GB GDDR6 · **Memory bandwidth**: 864 GB/s
- **FP16 tensor perf.**: 181 TFLOPS · **FP32 perf.**: 90.5 TFLOPS
- **CUDA cores**: 18,176 · **TDP**: 300 W
- **PowerScore ((RTX 3090 = 100))**: 127 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## L40 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the L40 ($0.34/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.235, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.235 | $5.64 | $172 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.117 | $2.81 | $85 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.152 | $3.65 | $111 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a L40 (48 GB VRAM)

With 48 GB of GDDR6, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.235.

- One-click template: [vLLM on a L40](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a L40](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a L40](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## L40 availability by region

4 × L40 across 2 machines, live from inventory:

- Montréal
- Paris

## L40 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **L40** (this card) | 48 GB | 181 | $0.235 | $1.30‰ |
| [L4](https://powergpu.ai/gpu/l4) | 24 GB | 121 | $0.225 | $1.86‰ |
| [A10](https://powergpu.ai/gpu/a10) | 24 GB | 125 | $0.168 | $1.34‰ |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | 125 | $0.130 | $1.04‰ |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | 65 | $0.103 | $1.58‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a L40: frequently asked questions

**How much does it cost to rent an NVIDIA L40 per hour?**

$0.235 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.34. Interruptible capacity costs $0.117/hr and a 3-month reservation $0.152/hr. Around $172/month if you keep one running non-stop, billed per second.

**What can a L40 with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the L40 available to rent right now?**

Yes — 4 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a L40?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by L40, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu l40 --template pytorch.

**Why is the L40 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/l40 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/l4 -->

---
title: "Rent L4 — $0.225/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the L4 (24 GB GDDR6) from $0.112/hr interruptible or $0.225/hr on-demand — fixed, ≥30% below market. 28 GPUs in 4 regions, per-second billing."
url: https://powergpu.ai/gpu/l4
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA L4 — 24 GB, $0.225/hr on-demand

- VRAM 24 GB GDDR6
- FP16 tensor 121 TFLOPS
- PowerScore 85 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 28 4 regions

[On-demand (guaranteed) $0.225 /GPU-hr ≈ $164/mo · market ~~$0.32~~ (−30%)](https://cloud.powergpu.ai/?gpu=l4) [Interruptible $0.112 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=l4&type=spot) [Reserved 3 mo $0.146 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Ada Lovelace architecture

The L4 is the efficient inference card: 24 GB GDDR6 at just 72 W, single-slot, Ada tensor cores and dual NVENC/NVDEC engines. It is the right rental for video transcoding pipelines, small-model APIs and batch embeddings where power and price matter more than peak FLOPS.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA L4 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA L4 · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 24 GB GDDR6 · **Memory bandwidth**: 300 GB/s
- **FP16 tensor perf.**: 121 TFLOPS · **FP32 perf.**: 30.3 TFLOPS
- **CUDA cores**: 7,680 · **TDP**: 72 W
- **PowerScore ((RTX 3090 = 100))**: 85 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## L4 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the L4 ($0.32/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.225, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.225 | $5.40 | $164 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.112 | $2.69 | $82 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.146 | $3.50 | $107 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a L4 (24 GB VRAM)

With 24 GB of GDDR6, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.225.

- One-click template: [vLLM on a L4](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a L4](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a L4](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## L4 availability by region

28 × L4 across 4 machines, live from inventory:

- São Paulo
- Montréal
- Amsterdam
- Frankfurt

## L4 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **L4** (this card) | 24 GB | 121 | $0.225 | $1.86‰ |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | 181 | $0.235 | $1.30‰ |
| [A10](https://powergpu.ai/gpu/a10) | 24 GB | 125 | $0.168 | $1.34‰ |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | 125 | $0.130 | $1.04‰ |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | 65 | $0.103 | $1.58‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [L4 vs Tesla T4](https://powergpu.ai/compare/l4-vs-tesla-t4) — specs, price per hour, which to rent
- [L4 vs A10](https://powergpu.ai/compare/l4-vs-a10) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a L4: frequently asked questions

**How much does it cost to rent an NVIDIA L4 per hour?**

$0.225 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.32. Interruptible capacity costs $0.112/hr and a 3-month reservation $0.146/hr. Around $164/month if you keep one running non-stop, billed per second.

**What can a L4 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the L4 available to rent right now?**

Yes — 28 GPUs across 4 machines in 4 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a L4?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by L4, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu l4 --template pytorch.

**Why is the L4 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/l4 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5080 -->

---
title: "Rent RTX 5080 — $0.186/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5080 (16 GB GDDR7) from $0.093/hr interruptible or $0.186/hr on-demand — fixed, ≥30% below market. 97 GPUs in 16 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5080
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX 5080 — 16 GB, $0.186/hr on-demand

- VRAM 16 GB GDDR7
- FP16 tensor 225 TFLOPS
- PowerScore 158 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 97 16 regions

[On-demand (guaranteed) $0.186 /GPU-hr ≈ $136/mo · market ~~$0.27~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5080) [Interruptible $0.093 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5080&type=spot) [Reserved 3 mo $0.120 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Blackwell architecture

The RTX 5080 pairs 16 GB of GDDR7 at 960 GB/s with 10,752 Blackwell CUDA cores. It is a fast card for SDXL, 7B–8B FP16 inference and video encoding workloads that fit in 16 GB — with FP4 support that 40-series cards lack.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 5080 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5080 · **Architecture**: Blackwell (2025)
- **VRAM**: 16 GB GDDR7 · **Memory bandwidth**: 960 GB/s
- **FP16 tensor perf.**: 225 TFLOPS · **FP32 perf.**: 56.3 TFLOPS
- **CUDA cores**: 10,752 · **TDP**: 360 W
- **PowerScore ((RTX 3090 = 100))**: 158 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5080 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5080 ($0.27/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.186, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.186 | $4.46 | $136 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.093 | $2.23 | $68 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.120 | $2.88 | $88 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5080 (16 GB VRAM)

With 16 GB of GDDR7, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.186.

- One-click template: [Ollama on a RTX 5080](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 5080](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 5080](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5080 availability by region

97 × RTX 5080 across 24 machines, live from inventory:

- Helsinki
- Madrid
- Tel Aviv
- Osaka
- Sydney
- Dallas, TX
- Johannesburg
- Milan
- Seoul
- Amsterdam
- Querétaro
- Los Angeles, CA
- +4 more

## RTX 5080 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5080** (this card) | 16 GB | 225 | $0.186 | $0.83‰ |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB | 176 | $0.131 | $0.74‰ |
| [RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB | 92 | $0.112 | $1.22‰ |
| [RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB | 123 | $0.112 | $0.91‰ |
| [RTX 5060](https://powergpu.ai/gpu/rtx-5060) | 8 GB | 74 | $0.063 | $0.85‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX 5080 vs RTX 4090](https://powergpu.ai/compare/rtx-5080-vs-rtx-4090) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX 5080: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5080 per hour?**

$0.186 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.27. Interruptible capacity costs $0.093/hr and a 3-month reservation $0.120/hr. Around $136/month if you keep one running non-stop, billed per second.

**What can a RTX 5080 with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5080 available to rent right now?**

Yes — 97 GPUs across 24 machines in 16 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5080?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5080, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5080 --template pytorch.

**Why is the RTX 5080 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5080 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-pro-4000 -->

---
title: "Rent RTX PRO 4000 — $0.183/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX PRO 4000 (20 GB GDDR7) from $0.091/hr interruptible or $0.183/hr on-demand — fixed, ≥30% below market. 62 GPUs in 8 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-pro-4000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX PRO 4000 — 20 GB, $0.183/hr on-demand

- VRAM 20 GB GDDR7
- FP16 tensor 150 TFLOPS
- PowerScore 106 RTX 3090 = 100
- Configs 1–4× PCIe 5.0
- Online now 62 8 regions

[On-demand (guaranteed) $0.183 /GPU-hr ≈ $134/mo · market ~~$0.26~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-pro-4000) [Interruptible $0.091 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-pro-4000&type=spot) [Reserved 3 mo $0.118 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Blackwell architecture

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX PRO 4000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX PRO 4000 · **Architecture**: Blackwell (2025)
- **VRAM**: 20 GB GDDR7 · **Memory bandwidth**: 672 GB/s
- **FP16 tensor perf.**: 150 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: 8,960 · **TDP**: 140 W
- **PowerScore ((RTX 3090 = 100))**: 106 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX PRO 4000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX PRO 4000 ($0.26/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.183, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.183 | $4.39 | $134 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.091 | $2.18 | $66 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.118 | $2.83 | $86 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX PRO 4000 (20 GB VRAM)

With 20 GB of GDDR7, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.183.

- One-click template: [Linux Desktop on a RTX PRO 4000](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX PRO 4000](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX PRO 4000](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX PRO 4000 availability by region

62 × RTX PRO 4000 across 8 machines, live from inventory:

- Montréal
- Seattle, WA
- Chicago, IL
- Dubai
- Milan
- New York, NY
- Tokyo
- Frankfurt

## RTX PRO 4000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX PRO 4000** (this card) | 20 GB | 150 | $0.183 | $1.22‰ |
| [Q RTX 8000](https://powergpu.ai/gpu/q-rtx-8000) | 48 GB | 65 | $0.178 | $2.74‰ |
| [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | 111 | $0.161 | $1.45‰ |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB | 107 | $0.128 | $1.20‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX PRO 4000: frequently asked questions

**How much does it cost to rent an NVIDIA RTX PRO 4000 per hour?**

$0.183 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.26. Interruptible capacity costs $0.091/hr and a 3-month reservation $0.118/hr. Around $134/month if you keep one running non-stop, billed per second.

**What can a RTX PRO 4000 with 20 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX PRO 4000 available to rent right now?**

Yes — 62 GPUs across 8 machines in 8 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX PRO 4000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX PRO 4000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-pro-4000 --template pytorch.

**Why is the RTX PRO 4000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-pro-4000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/q-rtx-8000 -->

---
title: "Rent Q RTX 8000 — $0.178/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Q RTX 8000 (48 GB GDDR6) from $0.089/hr interruptible or $0.178/hr on-demand — fixed, ≥30% below market. 5 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/q-rtx-8000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2018 · prices checked 2026-09-14

# Rent NVIDIA Q RTX 8000 — 48 GB, $0.178/hr on-demand

- VRAM 48 GB GDDR6
- FP16 tensor 65 TFLOPS
- PowerScore 46 RTX 3090 = 100
- Configs 1–2× PCIe 3.0
- Online now 5 2 regions

[On-demand (guaranteed) $0.178 /GPU-hr ≈ $130/mo · market ~~$0.25~~ (−30%)](https://cloud.powergpu.ai/?gpu=q-rtx-8000) [Interruptible $0.089 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=q-rtx-8000&type=spot) [Reserved 3 mo $0.115 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

The Quadro RTX 8000 is a Turing workstation card with 48 GB of GDDR6 and NVLink bridge support. Slow by modern tensor standards, but 48 GB at this price is useful for VRAM-bound rendering and for running 32B-class 4-bit models on a budget.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA Q RTX 8000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Q RTX 8000 · **Architecture**: Turing (2018)
- **VRAM**: 48 GB GDDR6 · **Memory bandwidth**: 672 GB/s
- **FP16 tensor perf.**: 65 TFLOPS · **FP32 perf.**: 16.3 TFLOPS
- **CUDA cores**: 4,608 · **TDP**: 295 W
- **PowerScore ((RTX 3090 = 100))**: 46 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 2× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Q RTX 8000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Q RTX 8000 ($0.25/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.178, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.178 | $4.27 | $130 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.089 | $2.14 | $65 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.115 | $2.76 | $84 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 2× machine costs exactly 2× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Q RTX 8000 (48 GB VRAM)

With 48 GB of GDDR6, a single card holds a **~14B-parameter LLM in FP16** or up to **~72B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 2× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.178.

- One-click template: [Ollama on a Q RTX 8000](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a Q RTX 8000](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a Q RTX 8000](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Q RTX 8000 availability by region

5 × Q RTX 8000 across 2 machines, live from inventory:

- Santiago
- Bucharest

## Q RTX 8000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Q RTX 8000** (this card) | 48 GB | 65 | $0.178 | $2.74‰ |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB | 150 | $0.183 | $1.22‰ |
| [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | 111 | $0.161 | $1.45‰ |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB | 107 | $0.128 | $1.20‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Q RTX 8000: frequently asked questions

**How much does it cost to rent an NVIDIA Q RTX 8000 per hour?**

$0.178 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.25. Interruptible capacity costs $0.089/hr and a 3-month reservation $0.115/hr. Around $130/month if you keep one running non-stop, billed per second.

**What can a Q RTX 8000 with 48 GB VRAM run?**

In LLM terms, roughly a 14B-parameter model in FP16 or up to ~72B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 2× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Q RTX 8000 available to rent right now?**

Yes — 5 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 2×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Q RTX 8000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Q RTX 8000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu q-rtx-8000 --template pytorch.

**Why is the Q RTX 8000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/q-rtx-8000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/a10 -->

---
title: "Rent A10 — $0.168/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the A10 (24 GB GDDR6) from $0.084/hr interruptible or $0.168/hr on-demand — fixed, ≥30% below market. 8 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/a10
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA A10 — 24 GB, $0.168/hr on-demand

- VRAM 24 GB GDDR6
- FP16 tensor 125 TFLOPS
- PowerScore 88 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 8 3 regions

[On-demand (guaranteed) $0.168 /GPU-hr ≈ $123/mo · market ~~$0.24~~ (−30%)](https://cloud.powergpu.ai/?gpu=a10) [Interruptible $0.084 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=a10&type=spot) [Reserved 3 mo $0.109 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Ampere architecture

The A10 is Ampere's mainstream datacenter card: 24 GB GDDR6, 9,216 CUDA cores, 150 W, passively cooled. Rent it for 7B–13B inference, graphics-heavy virtual workstations and batch vision jobs at a rate well below the A100.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA A10 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA A10 · **Architecture**: Ampere (2021)
- **VRAM**: 24 GB GDDR6 · **Memory bandwidth**: 600 GB/s
- **FP16 tensor perf.**: 125 TFLOPS · **FP32 perf.**: 31.2 TFLOPS
- **CUDA cores**: 9,216 · **TDP**: 150 W
- **PowerScore ((RTX 3090 = 100))**: 88 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## A10 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the A10 ($0.24/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.168, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.168 | $4.03 | $123 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.084 | $2.02 | $61 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.109 | $2.62 | $80 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a A10 (24 GB VRAM)

With 24 GB of GDDR6, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.168.

- One-click template: [vLLM on a A10](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a A10](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a A10](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## A10 availability by region

8 × A10 across 3 machines, live from inventory:

- Ashburn, VA
- Osaka
- Chicago, IL

## A10 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **A10** (this card) | 24 GB | 125 | $0.168 | $1.34‰ |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | 125 | $0.130 | $1.04‰ |
| [L4](https://powergpu.ai/gpu/l4) | 24 GB | 121 | $0.225 | $1.86‰ |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | 65 | $0.103 | $1.58‰ |
| [L40](https://powergpu.ai/gpu/l40) | 48 GB | 181 | $0.235 | $1.30‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [L4 vs A10](https://powergpu.ai/compare/l4-vs-a10) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a A10: frequently asked questions

**How much does it cost to rent an NVIDIA A10 per hour?**

$0.168 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.24. Interruptible capacity costs $0.084/hr and a 3-month reservation $0.109/hr. Around $123/month if you keep one running non-stop, billed per second.

**What can a A10 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the A10 available to rent right now?**

Yes — 8 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a A10?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by A10, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu a10 --template pytorch.

**Why is the A10 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/a10 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4080 -->

---
title: "Rent RTX 4080 — $0.168/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4080 (16 GB GDDR6X) from $0.084/hr interruptible or $0.168/hr on-demand — fixed, ≥30% below market. 14 GPUs in 5 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4080
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2022 · prices checked 2026-09-14

# Rent NVIDIA RTX 4080 — 16 GB, $0.168/hr on-demand

- VRAM 16 GB GDDR6X
- FP16 tensor 195 TFLOPS
- PowerScore 137 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 14 5 regions

[On-demand (guaranteed) $0.168 /GPU-hr ≈ $123/mo · market ~~$0.24~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4080) [Interruptible $0.084 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4080&type=spot) [Reserved 3 mo $0.109 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

The RTX 4080 (16 GB GDDR6X, 9,728 CUDA cores) sits between the 4070 Ti and the 4090: SDXL, Flux FP8 and 8B FP16 inference run comfortably, at a noticeably lower rate than the 4090.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4080 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4080 · **Architecture**: Ada Lovelace (2022)
- **VRAM**: 16 GB GDDR6X · **Memory bandwidth**: 717 GB/s
- **FP16 tensor perf.**: 195 TFLOPS · **FP32 perf.**: 48.7 TFLOPS
- **CUDA cores**: 9,728 · **TDP**: 320 W
- **PowerScore ((RTX 3090 = 100))**: 137 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 1,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4080 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4080 ($0.24/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.168, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.168 | $4.03 | $123 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.084 | $2.02 | $61 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.109 | $2.62 | $80 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4080 (16 GB VRAM)

With 16 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.168.

- One-click template: [Ollama on a RTX 4080](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4080](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4080](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4080 availability by region

14 × RTX 4080 across 5 machines, live from inventory:

- Paris
- Johannesburg
- Warsaw
- Chicago, IL
- São Paulo

## RTX 4080 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4080** (this card) | 16 GB | 195 | $0.168 | $0.86‰ |
| [RTX 4080S](https://powergpu.ai/gpu/rtx-4080s) | 16 GB | 208 | $0.127 | $0.61‰ |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | 176 | $0.103 | $0.59‰ |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | 160 | $0.075 | $0.47‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4080: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4080 per hour?**

$0.168 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.24. Interruptible capacity costs $0.084/hr and a 3-month reservation $0.109/hr. Around $123/month if you keep one running non-stop, billed per second.

**What can a RTX 4080 with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4080 available to rent right now?**

Yes — 14 GPUs across 5 machines in 5 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4080?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4080, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4080 --template pytorch.

**Why is the RTX 4080 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4080 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-a5000 -->

---
title: "Rent RTX A5000 — $0.161/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX A5000 (24 GB GDDR6) from $0.080/hr interruptible or $0.161/hr on-demand — fixed, ≥30% below market. 52 GPUs in 4 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-a5000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA RTX A5000 — 24 GB, $0.161/hr on-demand

- VRAM 24 GB GDDR6
- FP16 tensor 111 TFLOPS
- PowerScore 78 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 52 4 regions

[On-demand (guaranteed) $0.161 /GPU-hr ≈ $118/mo · market ~~$0.23~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-a5000) [Interruptible $0.080 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-a5000&type=spot) [Reserved 3 mo $0.104 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ampere architecture

The RTX A5000 combines 24 GB of ECC GDDR6 with a 230 W blower design that survives weeks of sustained load in a chassis. It is the reliability pick for long computer-vision training queues and multi-day render jobs at 4090-class VRAM.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX A5000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX A5000 · **Architecture**: Ampere (2021)
- **VRAM**: 24 GB GDDR6 · **Memory bandwidth**: 768 GB/s
- **FP16 tensor perf.**: 111 TFLOPS · **FP32 perf.**: 27.8 TFLOPS
- **CUDA cores**: 8,192 · **TDP**: 230 W
- **PowerScore ((RTX 3090 = 100))**: 78 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX A5000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX A5000 ($0.23/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.161, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.161 | $3.86 | $118 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.080 | $1.92 | $58 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.104 | $2.50 | $76 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX A5000 (24 GB VRAM)

With 24 GB of GDDR6, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.161.

- Recommended for [computer vision](https://powergpu.ai/use-cases/computer-vision) — YOLO11 epochs from $0.09
- One-click template: [Linux Desktop on a RTX A5000](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX A5000](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX A5000](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX A5000 availability by region

52 × RTX A5000 across 6 machines, live from inventory:

- Singapore
- Chicago, IL
- Montréal
- Stockholm

## RTX A5000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX A5000** (this card) | 24 GB | 111 | $0.161 | $1.45‰ |
| [Q RTX 8000](https://powergpu.ai/gpu/q-rtx-8000) | 48 GB | 65 | $0.178 | $2.74‰ |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) | 20 GB | 150 | $0.183 | $1.22‰ |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB | 107 | $0.128 | $1.20‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX A5000: frequently asked questions

**How much does it cost to rent an NVIDIA RTX A5000 per hour?**

$0.161 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.23. Interruptible capacity costs $0.080/hr and a 3-month reservation $0.104/hr. Around $118/month if you keep one running non-stop, billed per second.

**What can a RTX A5000 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX A5000 available to rent right now?**

Yes — 52 GPUs across 6 machines in 4 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX A5000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX A5000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-a5000 --template pytorch.

**Why is the RTX A5000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-a5000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3090-ti -->

---
title: "Rent RTX 3090 Ti — $0.147/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3090 Ti (24 GB GDDR6X) from $0.073/hr interruptible or $0.147/hr on-demand — fixed, ≥30% below market. 8 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3090-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer flagship · Ampere · launched 2022 · prices checked 2026-09-14

# Rent NVIDIA RTX 3090 Ti — 24 GB, $0.147/hr on-demand

- VRAM 24 GB GDDR6X
- FP16 tensor 160 TFLOPS
- PowerScore 113 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 8 3 regions

[On-demand (guaranteed) $0.147 /GPU-hr ≈ $107/mo · market ~~$0.21~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3090-ti) [Interruptible $0.073 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3090-ti&type=spot) [Reserved 3 mo $0.095 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer flagship · Ampere architecture

The community favourite: consumer pricing with serious tensor throughput. Ideal for diffusion models, quantized LLMs and fine-tuning runs that fit in 24 GB. Supply is deep, so interruptible capacity is almost always available at half price.

## NVIDIA RTX 3090 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3090 Ti · **Architecture**: Ampere (2022)
- **VRAM**: 24 GB GDDR6X · **Memory bandwidth**: 1,008 GB/s
- **FP16 tensor perf.**: 160 TFLOPS · **FP32 perf.**: 40.0 TFLOPS
- **CUDA cores**: 10,752 · **TDP**: 450 W
- **PowerScore ((RTX 3090 = 100))**: 113 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 1,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3090 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3090 Ti ($0.21/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.147, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.147 | $3.53 | $107 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.073 | $1.75 | $53 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.095 | $2.28 | $69 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3090 Ti (24 GB VRAM)

With 24 GB of GDDR6X, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.147.

- One-click template: [ComfyUI on a RTX 3090 Ti](https://powergpu.ai/templates/comfyui)
- One-click template: [Ollama on a RTX 3090 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [Kohya's GUI on a RTX 3090 Ti](https://powergpu.ai/templates/kohya-s-gui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3090 Ti availability by region

8 × RTX 3090 Ti across 3 machines, live from inventory:

- Singapore
- London
- Helsinki

## RTX 3090 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3090 Ti** (this card) | 24 GB | 160 | $0.147 | $0.92‰ |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | 142 | $0.108 | $0.76‰ |
| [RTX 3080 Ti](https://powergpu.ai/gpu/rtx-3080-ti) | 12 GB | 136 | $0.094 | $0.69‰ |
| [RTX 3080](https://powergpu.ai/gpu/rtx-3080) | 10 GB | 119 | $0.075 | $0.63‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 3090 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3090 Ti per hour?**

$0.147 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.21. Interruptible capacity costs $0.073/hr and a 3-month reservation $0.095/hr. Around $107/month if you keep one running non-stop, billed per second.

**What can a RTX 3090 Ti with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3090 Ti available to rent right now?**

Yes — 8 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3090 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3090 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3090-ti --template pytorch.

**Why is the RTX 3090 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3090-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5070-ti -->

---
title: "Rent RTX 5070 Ti — $0.131/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5070 Ti (16 GB GDDR7) from $0.065/hr interruptible or $0.131/hr on-demand — fixed, ≥30% below market. 54 GPUs in 10 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5070-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX 5070 Ti — 16 GB, $0.131/hr on-demand

- VRAM 16 GB GDDR7
- FP16 tensor 176 TFLOPS
- PowerScore 124 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 54 10 regions

[On-demand (guaranteed) $0.131 /GPU-hr ≈ $96/mo · market ~~$0.19~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5070-ti) [Interruptible $0.065 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5070-ti&type=spot) [Reserved 3 mo $0.085 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Blackwell architecture

The RTX 5070 Ti carries 16 GB GDDR7 at 896 GB/s and 8,960 Blackwell CUDA cores at 300 W — a fast, efficient card for SDXL and Flux with FP8 weights and for 8B-class FP16 inference.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 5070 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5070 Ti · **Architecture**: Blackwell (2025)
- **VRAM**: 16 GB GDDR7 · **Memory bandwidth**: 896 GB/s
- **FP16 tensor perf.**: 176 TFLOPS · **FP32 perf.**: 43.9 TFLOPS
- **CUDA cores**: 8,960 · **TDP**: 300 W
- **PowerScore ((RTX 3090 = 100))**: 124 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5070 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5070 Ti ($0.19/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.131, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.131 | $3.14 | $96 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.065 | $1.56 | $47 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.085 | $2.04 | $62 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5070 Ti (16 GB VRAM)

With 16 GB of GDDR7, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.131.

- One-click template: [Ollama on a RTX 5070 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 5070 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 5070 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5070 Ti availability by region

54 × RTX 5070 Ti across 13 machines, live from inventory:

- Warsaw
- New York, NY
- Dubai
- São Paulo
- Miami, FL
- Stockholm
- Seoul
- Bucharest
- Jakarta
- Milan

## RTX 5070 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5070 Ti** (this card) | 16 GB | 176 | $0.131 | $0.74‰ |
| [RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB | 92 | $0.112 | $1.22‰ |
| [RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB | 123 | $0.112 | $0.91‰ |
| [RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB | 225 | $0.186 | $0.83‰ |
| [RTX 5060](https://powergpu.ai/gpu/rtx-5060) | 8 GB | 74 | $0.063 | $0.85‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 5070 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5070 Ti per hour?**

$0.131 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.19. Interruptible capacity costs $0.065/hr and a 3-month reservation $0.085/hr. Around $96/month if you keep one running non-stop, billed per second.

**What can a RTX 5070 Ti with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5070 Ti available to rent right now?**

Yes — 54 GPUs across 13 machines in 10 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5070 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5070 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5070-ti --template pytorch.

**Why is the RTX 5070 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5070-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/tesla-v100 -->

---
title: "Rent Tesla V100 — $0.130/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Tesla V100 (16 GB HBM2) from $0.065/hr interruptible or $0.130/hr on-demand — fixed, ≥30% below market. 96 GPUs in 14 regions, per-second billing."
url: https://powergpu.ai/gpu/tesla-v100
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Volta · launched 2017 · prices checked 2026-09-14

# Rent NVIDIA Tesla V100 — 16 GB, $0.130/hr on-demand

- VRAM 16 GB HBM2
- FP16 tensor 125 TFLOPS
- PowerScore 88 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 96 14 regions

[On-demand (guaranteed) $0.130 /GPU-hr ≈ $95/mo · market ~~$0.19~~ (−30%)](https://cloud.powergpu.ai/?gpu=tesla-v100) [Interruptible $0.065 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=tesla-v100&type=spot) [Reserved 3 mo $0.084 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Volta architecture

The Tesla V100 was the first tensor-core datacenter GPU: 16 GB HBM2 at 900 GB/s, 5,120 CUDA cores, honest FP64. In 2026 it is a bargain for coursework, classic deep-learning models, double-precision simulation and CI — not for modern LLMs.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA Tesla V100 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Tesla V100 · **Architecture**: Volta (2017)
- **VRAM**: 16 GB HBM2 · **Memory bandwidth**: 900 GB/s
- **FP16 tensor perf.**: 125 TFLOPS · **FP32 perf.**: 14.1 TFLOPS
- **CUDA cores**: 5,120 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 88 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Tesla V100 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Tesla V100 ($0.19/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.130, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.130 | $3.12 | $95 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.065 | $1.56 | $47 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.084 | $2.02 | $61 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Tesla V100 (16 GB VRAM)

With 16 GB of HBM2, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.130.

- One-click template: [vLLM on a Tesla V100](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a Tesla V100](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a Tesla V100](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Tesla V100 availability by region

96 × Tesla V100 across 22 machines, live from inventory:

- Mumbai
- Jakarta
- Seattle, WA
- São Paulo
- Dallas, TX
- Amsterdam
- Montréal
- Stockholm
- Milan
- Vancouver
- London
- Frankfurt
- +2 more

## Tesla V100 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Tesla V100** (this card) | 16 GB | 125 | $0.130 | $1.04‰ |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | 65 | $0.103 | $1.58‰ |
| [A10](https://powergpu.ai/gpu/a10) | 24 GB | 125 | $0.168 | $1.34‰ |
| [Tesla P100](https://powergpu.ai/gpu/tesla-p100) | 16 GB | 19 | $0.066 | $3.47‰ |
| [Tesla P40](https://powergpu.ai/gpu/tesla-p40) | 24 GB | 12 | $0.047 | $3.92‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [Tesla V100 vs RTX 3090](https://powergpu.ai/compare/tesla-v100-vs-rtx-3090) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a Tesla V100: frequently asked questions

**How much does it cost to rent an NVIDIA Tesla V100 per hour?**

$0.130 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.19. Interruptible capacity costs $0.065/hr and a 3-month reservation $0.084/hr. Around $95/month if you keep one running non-stop, billed per second.

**What can a Tesla V100 with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Tesla V100 available to rent right now?**

Yes — 96 GPUs across 22 machines in 14 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Tesla V100?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Tesla V100, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu tesla-v100 --template pytorch.

**Why is the Tesla V100 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/tesla-v100 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4000ada -->

---
title: "Rent RTX 4000Ada — $0.128/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4000Ada (20 GB GDDR6) from $0.064/hr interruptible or $0.128/hr on-demand — fixed, ≥30% below market. 6 GPUs in 1 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4000ada
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA RTX 4000Ada — 20 GB, $0.128/hr on-demand

- VRAM 20 GB GDDR6
- FP16 tensor 107 TFLOPS
- PowerScore 75 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 6 1 regions

[On-demand (guaranteed) $0.128 /GPU-hr ≈ $93/mo · market ~~$0.18~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4000ada) [Interruptible $0.064 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4000ada&type=spot) [Reserved 3 mo $0.083 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ada Lovelace architecture

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX 4000Ada specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4000Ada · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 20 GB GDDR6 · **Memory bandwidth**: 360 GB/s
- **FP16 tensor perf.**: 107 TFLOPS · **FP32 perf.**: 26.7 TFLOPS
- **CUDA cores**: 6,144 · **TDP**: 130 W
- **PowerScore ((RTX 3090 = 100))**: 75 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 5,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4000Ada price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4000Ada ($0.18/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.128, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.128 | $3.07 | $93 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.064 | $1.54 | $47 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.083 | $1.99 | $61 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4000Ada (20 GB VRAM)

With 20 GB of GDDR6, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.128.

- One-click template: [Linux Desktop on a RTX 4000Ada](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX 4000Ada](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX 4000Ada](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4000Ada availability by region

6 × RTX 4000Ada across 2 machines, live from inventory:

- Frankfurt

## RTX 4000Ada vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4000Ada** (this card) | 20 GB | 107 | $0.128 | $1.20‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |
| [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | 111 | $0.161 | $1.45‰ |
| [Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB | 65 | $0.094 | $1.45‰ |
| [Q RTX 8000](https://powergpu.ai/gpu/q-rtx-8000) | 48 GB | 65 | $0.178 | $2.74‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4000Ada: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4000Ada per hour?**

$0.128 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.18. Interruptible capacity costs $0.064/hr and a 3-month reservation $0.083/hr. Around $93/month if you keep one running non-stop, billed per second.

**What can a RTX 4000Ada with 20 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4000Ada available to rent right now?**

Yes — 6 GPUs across 2 machines in 1 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4000Ada?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4000Ada, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4000ada --template pytorch.

**Why is the RTX 4000Ada cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4000ada · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4080s -->

---
title: "Rent RTX 4080S — $0.127/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4080S (16 GB GDDR6X) from $0.063/hr interruptible or $0.127/hr on-demand — fixed, ≥30% below market. 33 GPUs in 14 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4080s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2024 · prices checked 2026-09-14

# Rent NVIDIA RTX 4080S — 16 GB, $0.127/hr on-demand

- VRAM 16 GB GDDR6X
- FP16 tensor 208 TFLOPS
- PowerScore 146 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 33 14 regions

[On-demand (guaranteed) $0.127 /GPU-hr ≈ $93/mo · market ~~$0.18~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4080s) [Interruptible $0.063 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4080s&type=spot) [Reserved 3 mo $0.082 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

The RTX 4080 SUPER offers 16 GB GDDR6X and 10,240 Ada CUDA cores at 320 W — most of a 4090's per-clock throughput with two-thirds of its VRAM. It is a strong choice for SDXL, Flux with FP8 weights and 7B–8B FP16 inference.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4080S specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4080S · **Architecture**: Ada Lovelace (2024)
- **VRAM**: 16 GB GDDR6X · **Memory bandwidth**: 736 GB/s
- **FP16 tensor perf.**: 208 TFLOPS · **FP32 perf.**: 52.2 TFLOPS
- **CUDA cores**: 10,240 · **TDP**: 320 W
- **PowerScore ((RTX 3090 = 100))**: 146 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4080S price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4080S ($0.18/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.127, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.127 | $3.05 | $93 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.063 | $1.51 | $46 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.082 | $1.97 | $60 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4080S (16 GB VRAM)

With 16 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.127.

- One-click template: [Ollama on a RTX 4080S](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4080S](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4080S](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4080S availability by region

33 × RTX 4080S across 17 machines, live from inventory:

- Osaka
- Frankfurt
- Querétaro
- Seattle, WA
- Paris
- Sydney
- Tokyo
- London
- Ashburn, VA
- New York, NY
- Santiago
- São Paulo
- +2 more

## RTX 4080S vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4080S** (this card) | 16 GB | 208 | $0.127 | $0.61‰ |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | 176 | $0.103 | $0.59‰ |
| [RTX 4080](https://powergpu.ai/gpu/rtx-4080) | 16 GB | 195 | $0.168 | $0.86‰ |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | 160 | $0.075 | $0.47‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4080S: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4080S per hour?**

$0.127 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.18. Interruptible capacity costs $0.063/hr and a 3-month reservation $0.082/hr. Around $93/month if you keep one running non-stop, billed per second.

**What can a RTX 4080S with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4080S available to rent right now?**

Yes — 33 GPUs across 17 machines in 14 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4080S?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4080S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4080s --template pytorch.

**Why is the RTX 4080S cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4080s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5060-ti -->

---
title: "Rent RTX 5060 Ti — $0.112/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5060 Ti (16 GB GDDR7) from $0.056/hr interruptible or $0.112/hr on-demand — fixed, ≥30% below market. 168 GPUs in 17 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5060-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX 5060 Ti — 16 GB, $0.112/hr on-demand

- VRAM 16 GB GDDR7
- FP16 tensor 92 TFLOPS
- PowerScore 65 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 168 17 regions

[On-demand (guaranteed) $0.112 /GPU-hr ≈ $82/mo · market ~~$0.16~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5060-ti) [Interruptible $0.056 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5060-ti&type=spot) [Reserved 3 mo $0.072 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Blackwell architecture

The RTX 5060 Ti 16 GB is the cheapest Blackwell card with enough VRAM for 8B FP16 or 14B 4-bit models and Flux at FP8 — deep supply keeps it among the most available cards in the catalogue.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 5060 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5060 Ti · **Architecture**: Blackwell (2025)
- **VRAM**: 16 GB GDDR7 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 92 TFLOPS · **FP32 perf.**: 23.7 TFLOPS
- **CUDA cores**: 4,608 · **TDP**: 180 W
- **PowerScore ((RTX 3090 = 100))**: 65 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5060 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5060 Ti ($0.16/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.112, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.112 | $2.69 | $82 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.056 | $1.34 | $41 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.072 | $1.73 | $53 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5060 Ti (16 GB VRAM)

With 16 GB of GDDR7, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.112.

- One-click template: [Ollama on a RTX 5060 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 5060 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 5060 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5060 Ti availability by region

168 × RTX 5060 Ti across 30 machines, live from inventory:

- Montréal
- Los Angeles, CA
- Milan
- Querétaro
- Dubai
- Seoul
- Singapore
- Paris
- Miami, FL
- Warsaw
- Ashburn, VA
- Sydney
- +5 more

## RTX 5060 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5060 Ti** (this card) | 16 GB | 92 | $0.112 | $1.22‰ |
| [RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB | 123 | $0.112 | $0.91‰ |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB | 176 | $0.131 | $0.74‰ |
| [RTX 5060](https://powergpu.ai/gpu/rtx-5060) | 8 GB | 74 | $0.063 | $0.85‰ |
| [RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB | 225 | $0.186 | $0.83‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 5060 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5060 Ti per hour?**

$0.112 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.16. Interruptible capacity costs $0.056/hr and a 3-month reservation $0.072/hr. Around $82/month if you keep one running non-stop, billed per second.

**What can a RTX 5060 Ti with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5060 Ti available to rent right now?**

Yes — 168 GPUs across 30 machines in 17 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5060 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5060 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5060-ti --template pytorch.

**Why is the RTX 5060 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5060-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5070 -->

---
title: "Rent RTX 5070 — $0.112/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5070 (12 GB GDDR7) from $0.056/hr interruptible or $0.112/hr on-demand — fixed, ≥30% below market. 32 GPUs in 4 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5070
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX 5070 — 12 GB, $0.112/hr on-demand

- VRAM 12 GB GDDR7
- FP16 tensor 123 TFLOPS
- PowerScore 87 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 32 4 regions

[On-demand (guaranteed) $0.112 /GPU-hr ≈ $82/mo · market ~~$0.16~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5070) [Interruptible $0.056 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5070&type=spot) [Reserved 3 mo $0.072 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Blackwell architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 5070 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5070 · **Architecture**: Blackwell (2025)
- **VRAM**: 12 GB GDDR7 · **Memory bandwidth**: 672 GB/s
- **FP16 tensor perf.**: 123 TFLOPS · **FP32 perf.**: 30.9 TFLOPS
- **CUDA cores**: 6,144 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 87 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5070 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5070 ($0.16/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.112, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.112 | $2.69 | $82 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.056 | $1.34 | $41 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.072 | $1.73 | $53 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5070 (12 GB VRAM)

With 12 GB of GDDR7, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.112.

- One-click template: [Ollama on a RTX 5070](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 5070](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 5070](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5070 availability by region

32 × RTX 5070 across 4 machines, live from inventory:

- Los Angeles, CA
- Querétaro
- Osaka
- Dubai

## RTX 5070 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5070** (this card) | 12 GB | 123 | $0.112 | $0.91‰ |
| [RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB | 92 | $0.112 | $1.22‰ |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB | 176 | $0.131 | $0.74‰ |
| [RTX 5060](https://powergpu.ai/gpu/rtx-5060) | 8 GB | 74 | $0.063 | $0.85‰ |
| [RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB | 225 | $0.186 | $0.83‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 5070: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5070 per hour?**

$0.112 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.16. Interruptible capacity costs $0.056/hr and a 3-month reservation $0.072/hr. Around $82/month if you keep one running non-stop, billed per second.

**What can a RTX 5070 with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5070 available to rent right now?**

Yes — 32 GPUs across 4 machines in 4 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5070?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5070, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5070 --template pytorch.

**Why is the RTX 5070 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5070 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3090 -->

---
title: "Rent RTX 3090 — $0.108/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3090 (24 GB GDDR6X) from $0.054/hr interruptible or $0.108/hr on-demand — fixed, ≥30% below market. 345 GPUs in 19 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer flagship · Ampere · launched 2020 · prices checked 2026-09-14

# Rent NVIDIA RTX 3090 — 24 GB, $0.108/hr on-demand

- VRAM 24 GB GDDR6X
- FP16 tensor 142 TFLOPS
- PowerScore 100 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 345 19 regions

[On-demand (guaranteed) $0.108 /GPU-hr ≈ $79/mo · market ~~$0.15~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3090) [Interruptible $0.054 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3090&type=spot) [Reserved 3 mo $0.070 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer flagship · Ampere architecture

The RTX 3090 is still the cheapest 24 GB card you can rent: Ampere, 10,496 CUDA cores, 936 GB/s GDDR6X. Older tensor cores mean lower FP16 throughput than a 4090, but for batch image generation, LoRA training and 13B-class inference it delivers the most VRAM per dollar on the sheet.

The community favourite: consumer pricing with serious tensor throughput. Ideal for diffusion models, quantized LLMs and fine-tuning runs that fit in 24 GB. Supply is deep, so interruptible capacity is almost always available at half price.

## NVIDIA RTX 3090 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3090 · **Architecture**: Ampere (2020)
- **VRAM**: 24 GB GDDR6X · **Memory bandwidth**: 936 GB/s
- **FP16 tensor perf.**: 142 TFLOPS · **FP32 perf.**: 35.6 TFLOPS
- **CUDA cores**: 10,496 · **TDP**: 350 W
- **PowerScore ((RTX 3090 = 100))**: 100 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3090 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3090 ($0.15/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.108, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.108 | $2.59 | $79 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.054 | $1.30 | $39 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.070 | $1.68 | $51 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3090 (24 GB VRAM)

With 24 GB of GDDR6X, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.108.

- Recommended for [computer vision](https://powergpu.ai/use-cases/computer-vision) — YOLO11 epochs from $0.09
- One-click template: [ComfyUI on a RTX 3090](https://powergpu.ai/templates/comfyui)
- One-click template: [Ollama on a RTX 3090](https://powergpu.ai/templates/ollama)
- One-click template: [Kohya's GUI on a RTX 3090](https://powergpu.ai/templates/kohya-s-gui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3090 availability by region

345 × RTX 3090 across 30 machines, live from inventory:

- São Paulo
- Sydney
- Chicago, IL
- Los Angeles, CA
- Montréal
- Amsterdam
- Paris
- Stockholm
- Warsaw
- Querétaro
- Singapore
- Jakarta
- +7 more

## RTX 3090 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3090** (this card) | 24 GB | 142 | $0.108 | $0.76‰ |
| [RTX 3080 Ti](https://powergpu.ai/gpu/rtx-3080-ti) | 12 GB | 136 | $0.094 | $0.69‰ |
| [RTX 3080](https://powergpu.ai/gpu/rtx-3080) | 10 GB | 119 | $0.075 | $0.63‰ |
| [RTX 3090 Ti](https://powergpu.ai/gpu/rtx-3090-ti) | 24 GB | 160 | $0.147 | $0.92‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX 4090 vs RTX 3090](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090) — specs, price per hour, which to rent
- [Tesla V100 vs RTX 3090](https://powergpu.ai/compare/tesla-v100-vs-rtx-3090) — specs, price per hour, which to rent
- [RTX 3090 vs RTX 3060](https://powergpu.ai/compare/rtx-3090-vs-rtx-3060) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX 3090: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3090 per hour?**

$0.108 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.15. Interruptible capacity costs $0.054/hr and a 3-month reservation $0.070/hr. Around $79/month if you keep one running non-stop, billed per second.

**What can a RTX 3090 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3090 available to rent right now?**

Yes — 345 GPUs across 30 machines in 19 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3090?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3090, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3090 --template pytorch.

**Why is the RTX 3090 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4070s-ti -->

---
title: "Rent RTX 4070S Ti — $0.103/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4070S Ti (16 GB GDDR6X) from $0.051/hr interruptible or $0.103/hr on-demand — fixed, ≥30% below market. 54 GPUs in 9 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4070s-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2024 · prices checked 2026-09-14

# Rent NVIDIA RTX 4070S Ti — 16 GB, $0.103/hr on-demand

- VRAM 16 GB GDDR6X
- FP16 tensor 176 TFLOPS
- PowerScore 124 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 54 9 regions

[On-demand (guaranteed) $0.103 /GPU-hr ≈ $75/mo · market ~~$0.15~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4070s-ti) [Interruptible $0.051 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4070s-ti&type=spot) [Reserved 3 mo $0.066 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4070S Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4070S Ti · **Architecture**: Ada Lovelace (2024)
- **VRAM**: 16 GB GDDR6X · **Memory bandwidth**: 672 GB/s
- **FP16 tensor perf.**: 176 TFLOPS · **FP32 perf.**: 44.1 TFLOPS
- **CUDA cores**: 8,448 · **TDP**: 285 W
- **PowerScore ((RTX 3090 = 100))**: 124 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4070S Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4070S Ti ($0.15/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.103, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.103 | $2.47 | $75 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.051 | $1.22 | $37 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.066 | $1.58 | $48 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4070S Ti (16 GB VRAM)

With 16 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.103.

- One-click template: [Ollama on a RTX 4070S Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4070S Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4070S Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4070S Ti availability by region

54 × RTX 4070S Ti across 9 machines, live from inventory:

- London
- Warsaw
- Los Angeles, CA
- Ashburn, VA
- Paris
- Montréal
- Bucharest
- Madrid
- Jakarta

## RTX 4070S Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4070S Ti** (this card) | 16 GB | 176 | $0.103 | $0.59‰ |
| [RTX 4080S](https://powergpu.ai/gpu/rtx-4080s) | 16 GB | 208 | $0.127 | $0.61‰ |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | 160 | $0.075 | $0.47‰ |
| [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | 88 | $0.074 | $0.84‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4070S Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4070S Ti per hour?**

$0.103 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.15. Interruptible capacity costs $0.051/hr and a 3-month reservation $0.066/hr. Around $75/month if you keep one running non-stop, billed per second.

**What can a RTX 4070S Ti with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4070S Ti available to rent right now?**

Yes — 54 GPUs across 9 machines in 9 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4070S Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4070S Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4070s-ti --template pytorch.

**Why is the RTX 4070S Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4070s-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/tesla-t4 -->

---
title: "Rent Tesla T4 — $0.103/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Tesla T4 (16 GB GDDR6) from $0.051/hr interruptible or $0.103/hr on-demand — fixed, ≥30% below market. 34 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/tesla-t4
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Turing · launched 2018 · prices checked 2026-09-14

# Rent NVIDIA Tesla T4 — 16 GB, $0.103/hr on-demand

- VRAM 16 GB GDDR6
- FP16 tensor 65 TFLOPS
- PowerScore 46 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 34 3 regions

[On-demand (guaranteed) $0.103 /GPU-hr ≈ $75/mo · market ~~$0.15~~ (−30%)](https://cloud.powergpu.ai/?gpu=tesla-t4) [Interruptible $0.051 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=tesla-t4&type=spot) [Reserved 3 mo $0.066 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Turing architecture

The Tesla T4 is the 70 W inference card found in every cloud: 16 GB GDDR6, Turing tensor cores, single-slot. Rent it for TensorRT-exported detectors, small-model APIs and cheap batch jobs — hundreds of frames per second for cents per hour.

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA Tesla T4 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Tesla T4 · **Architecture**: Turing (2018)
- **VRAM**: 16 GB GDDR6 · **Memory bandwidth**: 320 GB/s
- **FP16 tensor perf.**: 65 TFLOPS · **FP32 perf.**: 8.1 TFLOPS
- **CUDA cores**: 2,560 · **TDP**: 70 W
- **PowerScore ((RTX 3090 = 100))**: 46 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Tesla T4 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Tesla T4 ($0.15/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.103, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.103 | $2.47 | $75 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.051 | $1.22 | $37 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.066 | $1.58 | $48 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Tesla T4 (16 GB VRAM)

With 16 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.103.

- One-click template: [vLLM on a Tesla T4](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a Tesla T4](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a Tesla T4](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Tesla T4 availability by region

34 × Tesla T4 across 3 machines, live from inventory:

- Los Angeles, CA
- Frankfurt
- Mumbai

## Tesla T4 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Tesla T4** (this card) | 16 GB | 65 | $0.103 | $1.58‰ |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | 125 | $0.130 | $1.04‰ |
| [Tesla P100](https://powergpu.ai/gpu/tesla-p100) | 16 GB | 19 | $0.066 | $3.47‰ |
| [Tesla P40](https://powergpu.ai/gpu/tesla-p40) | 24 GB | 12 | $0.047 | $3.92‰ |
| [A10](https://powergpu.ai/gpu/a10) | 24 GB | 125 | $0.168 | $1.34‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [L4 vs Tesla T4](https://powergpu.ai/compare/l4-vs-tesla-t4) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a Tesla T4: frequently asked questions

**How much does it cost to rent an NVIDIA Tesla T4 per hour?**

$0.103 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.15. Interruptible capacity costs $0.051/hr and a 3-month reservation $0.066/hr. Around $75/month if you keep one running non-stop, billed per second.

**What can a Tesla T4 with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Tesla T4 available to rent right now?**

Yes — 34 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Tesla T4?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Tesla T4, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu tesla-t4 --template pytorch.

**Why is the Tesla T4 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/tesla-t4 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/titan-rtx -->

---
title: "Rent Titan RTX — $0.103/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Titan RTX (24 GB GDDR6) from $0.051/hr interruptible or $0.103/hr on-demand — fixed, ≥30% below market. 24 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/titan-rtx
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2018 · prices checked 2026-09-14

# Rent NVIDIA Titan RTX — 24 GB, $0.103/hr on-demand

- VRAM 24 GB GDDR6
- FP16 tensor 65 TFLOPS
- PowerScore 46 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 24 3 regions

[On-demand (guaranteed) $0.103 /GPU-hr ≈ $75/mo · market ~~$0.15~~ (−30%)](https://cloud.powergpu.ai/?gpu=titan-rtx) [Interruptible $0.051 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=titan-rtx&type=spot) [Reserved 3 mo $0.066 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

The Titan RTX is a Turing card with 24 GB of GDDR6 — the memory of an RTX 3090 with older tensor cores. It suits VRAM-bound jobs that are not throughput-critical: large-batch inference of older architectures, rendering, and 13B-class 4-bit models.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA Titan RTX specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Titan RTX · **Architecture**: Turing (2018)
- **VRAM**: 24 GB GDDR6 · **Memory bandwidth**: 672 GB/s
- **FP16 tensor perf.**: 65 TFLOPS · **FP32 perf.**: 16.3 TFLOPS
- **CUDA cores**: 4,608 · **TDP**: 280 W
- **PowerScore ((RTX 3090 = 100))**: 46 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Titan RTX price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Titan RTX ($0.15/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.103, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.103 | $2.47 | $75 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.051 | $1.22 | $37 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.066 | $1.58 | $48 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Titan RTX (24 GB VRAM)

With 24 GB of GDDR6, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.103.

- One-click template: [Ollama on a Titan RTX](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a Titan RTX](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a Titan RTX](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Titan RTX availability by region

24 × Titan RTX across 3 machines, live from inventory:

- Osaka
- Singapore
- Warsaw

## Titan RTX vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Titan RTX** (this card) | 24 GB | 65 | $0.103 | $1.58‰ |
| [Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB | 65 | $0.094 | $1.45‰ |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB | 107 | $0.128 | $1.20‰ |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB | 76 | $0.071 | $0.93‰ |
| [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | 111 | $0.161 | $1.45‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Titan RTX: frequently asked questions

**How much does it cost to rent an NVIDIA Titan RTX per hour?**

$0.103 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.15. Interruptible capacity costs $0.051/hr and a 3-month reservation $0.066/hr. Around $75/month if you keep one running non-stop, billed per second.

**What can a Titan RTX with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Titan RTX available to rent right now?**

Yes — 24 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Titan RTX?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Titan RTX, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu titan-rtx --template pytorch.

**Why is the Titan RTX cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/titan-rtx · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/q-rtx-6000 -->

---
title: "Rent Q RTX 6000 — $0.094/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Q RTX 6000 (24 GB GDDR6) from $0.047/hr interruptible or $0.094/hr on-demand — fixed, ≥30% below market. 5 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/q-rtx-6000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2018 · prices checked 2026-09-14

# Rent NVIDIA Q RTX 6000 — 24 GB, $0.094/hr on-demand

- VRAM 24 GB GDDR6
- FP16 tensor 65 TFLOPS
- PowerScore 46 RTX 3090 = 100
- Configs 1–4× PCIe 3.0
- Online now 5 3 regions

[On-demand (guaranteed) $0.094 /GPU-hr ≈ $69/mo · market ~~$0.13~~ (−30%)](https://cloud.powergpu.ai/?gpu=q-rtx-6000) [Interruptible $0.047 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=q-rtx-6000&type=spot) [Reserved 3 mo $0.061 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA Q RTX 6000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Q RTX 6000 · **Architecture**: Turing (2018)
- **VRAM**: 24 GB GDDR6 · **Memory bandwidth**: 672 GB/s
- **FP16 tensor perf.**: 65 TFLOPS · **FP32 perf.**: 16.3 TFLOPS
- **CUDA cores**: 4,608 · **TDP**: 260 W
- **PowerScore ((RTX 3090 = 100))**: 46 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 5,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Q RTX 6000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Q RTX 6000 ($0.13/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.094, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.094 | $2.26 | $69 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.047 | $1.13 | $34 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.061 | $1.46 | $45 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Q RTX 6000 (24 GB VRAM)

With 24 GB of GDDR6, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.094.

- One-click template: [Ollama on a Q RTX 6000](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a Q RTX 6000](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a Q RTX 6000](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Q RTX 6000 availability by region

5 × Q RTX 6000 across 3 machines, live from inventory:

- Singapore
- Frankfurt
- Seattle, WA

## Q RTX 6000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Q RTX 6000** (this card) | 24 GB | 65 | $0.094 | $1.45‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB | 76 | $0.071 | $0.93‰ |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) | 20 GB | 107 | $0.128 | $1.20‰ |
| [Quadro P4000](https://powergpu.ai/gpu/quadro-p4000) | 8 GB | 8 | $0.042 | $5.25‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Q RTX 6000: frequently asked questions

**How much does it cost to rent an NVIDIA Q RTX 6000 per hour?**

$0.094 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.13. Interruptible capacity costs $0.047/hr and a 3-month reservation $0.061/hr. Around $69/month if you keep one running non-stop, billed per second.

**What can a Q RTX 6000 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Q RTX 6000 available to rent right now?**

Yes — 5 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Q RTX 6000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Q RTX 6000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu q-rtx-6000 --template pytorch.

**Why is the Q RTX 6000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/q-rtx-6000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3080-ti -->

---
title: "Rent RTX 3080 Ti — $0.094/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3080 Ti (12 GB GDDR6) from $0.047/hr interruptible or $0.094/hr on-demand — fixed, ≥30% below market. 35 GPUs in 9 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3080-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA RTX 3080 Ti — 12 GB, $0.094/hr on-demand

- VRAM 12 GB GDDR6
- FP16 tensor 136 TFLOPS
- PowerScore 96 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 35 9 regions

[On-demand (guaranteed) $0.094 /GPU-hr ≈ $69/mo · market ~~$0.13~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3080-ti) [Interruptible $0.047 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3080-ti&type=spot) [Reserved 3 mo $0.061 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ampere architecture

The RTX 3080 Ti is close to a 3090 in compute (10,240 CUDA cores, 912 GB/s) with 12 GB of GDDR6X — a fast card for diffusion and 7B–13B quantized inference when 24 GB is not required.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 3080 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3080 Ti · **Architecture**: Ampere (2021)
- **VRAM**: 12 GB GDDR6 · **Memory bandwidth**: 912 GB/s
- **FP16 tensor perf.**: 136 TFLOPS · **FP32 perf.**: 34.1 TFLOPS
- **CUDA cores**: 10,240 · **TDP**: 350 W
- **PowerScore ((RTX 3090 = 100))**: 96 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3080 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3080 Ti ($0.13/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.094, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.094 | $2.26 | $69 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.047 | $1.13 | $34 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.061 | $1.46 | $45 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3080 Ti (12 GB VRAM)

With 12 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.094.

- One-click template: [Ollama on a RTX 3080 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 3080 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 3080 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3080 Ti availability by region

35 × RTX 3080 Ti across 10 machines, live from inventory:

- São Paulo
- New York, NY
- Amsterdam
- Montréal
- Miami, FL
- Helsinki
- Mumbai
- London
- Seoul

## RTX 3080 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3080 Ti** (this card) | 12 GB | 136 | $0.094 | $0.69‰ |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | 142 | $0.108 | $0.76‰ |
| [RTX 3080](https://powergpu.ai/gpu/rtx-3080) | 10 GB | 119 | $0.075 | $0.63‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | 81 | $0.056 | $0.69‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 3080 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3080 Ti per hour?**

$0.094 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.13. Interruptible capacity costs $0.047/hr and a 3-month reservation $0.061/hr. Around $69/month if you keep one running non-stop, billed per second.

**What can a RTX 3080 Ti with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3080 Ti available to rent right now?**

Yes — 35 GPUs across 10 machines in 9 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3080 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3080 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3080-ti --template pytorch.

**Why is the RTX 3080 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3080-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3080 -->

---
title: "Rent RTX 3080 — $0.075/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3080 (10 GB GDDR6) from $0.037/hr interruptible or $0.075/hr on-demand — fixed, ≥30% below market. 70 GPUs in 16 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3080
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ampere · launched 2020 · prices checked 2026-09-14

# Rent NVIDIA RTX 3080 — 10 GB, $0.075/hr on-demand

- VRAM 10 GB GDDR6
- FP16 tensor 119 TFLOPS
- PowerScore 84 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 70 16 regions

[On-demand (guaranteed) $0.075 /GPU-hr ≈ $55/mo · market ~~$0.11~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3080) [Interruptible $0.037 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3080&type=spot) [Reserved 3 mo $0.048 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ampere architecture

The RTX 3080 (10 GB) offers Ampere throughput close to a 3090 with less VRAM: fine for SD 1.5 and SDXL, 7B 4-bit inference and training smaller vision models at a low hourly rate.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 3080 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3080 · **Architecture**: Ampere (2020)
- **VRAM**: 10 GB GDDR6 · **Memory bandwidth**: 760 GB/s
- **FP16 tensor perf.**: 119 TFLOPS · **FP32 perf.**: 29.8 TFLOPS
- **CUDA cores**: 8,704 · **TDP**: 320 W
- **PowerScore ((RTX 3090 = 100))**: 84 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3080 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3080 ($0.11/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.075, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.075 | $1.80 | $55 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.037 | $0.89 | $27 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.048 | $1.15 | $35 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3080 (10 GB VRAM)

With 10 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.075.

- One-click template: [Ollama on a RTX 3080](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 3080](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 3080](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3080 availability by region

70 × RTX 3080 across 17 machines, live from inventory:

- Amsterdam
- Miami, FL
- London
- Sydney
- Querétaro
- Mumbai
- Osaka
- Bucharest
- Ashburn, VA
- Seattle, WA
- Santiago
- Seoul
- +4 more

## RTX 3080 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3080** (this card) | 10 GB | 119 | $0.075 | $0.63‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | 81 | $0.056 | $0.69‰ |
| [RTX 3080 Ti](https://powergpu.ai/gpu/rtx-3080-ti) | 12 GB | 136 | $0.094 | $0.69‰ |
| [RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB | 65 | $0.047 | $0.72‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 3080: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3080 per hour?**

$0.075 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.11. Interruptible capacity costs $0.037/hr and a 3-month reservation $0.048/hr. Around $55/month if you keep one running non-stop, billed per second.

**What can a RTX 3080 with 10 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3080 available to rent right now?**

Yes — 70 GPUs across 17 machines in 16 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3080?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3080, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3080 --template pytorch.

**Why is the RTX 3080 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3080 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4070 -->

---
title: "Rent RTX 4070 — $0.075/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4070 (12 GB GDDR6X) from $0.037/hr interruptible or $0.075/hr on-demand — fixed, ≥30% below market. 35 GPUs in 5 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4070
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA RTX 4070 — 12 GB, $0.075/hr on-demand

- VRAM 12 GB GDDR6X
- FP16 tensor 117 TFLOPS
- PowerScore 82 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 35 5 regions

[On-demand (guaranteed) $0.075 /GPU-hr ≈ $55/mo · market ~~$0.11~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4070) [Interruptible $0.037 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4070&type=spot) [Reserved 3 mo $0.048 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

The RTX 4070 brings Ada tensor cores and 12 GB GDDR6X at just 200 W. It runs 7B–8B quantized models and SDXL comfortably and is a cost-efficient card for inference endpoints that do not need 24 GB.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4070 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4070 · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 12 GB GDDR6X · **Memory bandwidth**: 504 GB/s
- **FP16 tensor perf.**: 117 TFLOPS · **FP32 perf.**: 29.1 TFLOPS
- **CUDA cores**: 5,888 · **TDP**: 200 W
- **PowerScore ((RTX 3090 = 100))**: 82 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4070 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4070 ($0.11/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.075, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.075 | $1.80 | $55 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.037 | $0.89 | $27 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.048 | $1.15 | $35 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4070 (12 GB VRAM)

With 12 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.075.

- One-click template: [Ollama on a RTX 4070](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4070](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4070](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4070 availability by region

35 × RTX 4070 across 5 machines, live from inventory:

- Miami, FL
- Dallas, TX
- Seoul
- Mumbai
- Stockholm

## RTX 4070 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4070** (this card) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | 160 | $0.075 | $0.47‰ |
| [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | 88 | $0.074 | $0.84‰ |
| [RTX 4070S](https://powergpu.ai/gpu/rtx-4070s) | 12 GB | 142 | $0.065 | $0.46‰ |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | 176 | $0.103 | $0.59‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4070: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4070 per hour?**

$0.075 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.11. Interruptible capacity costs $0.037/hr and a 3-month reservation $0.048/hr. Around $55/month if you keep one running non-stop, billed per second.

**What can a RTX 4070 with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4070 available to rent right now?**

Yes — 35 GPUs across 5 machines in 5 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4070?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4070, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4070 --template pytorch.

**Why is the RTX 4070 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4070 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4070-ti -->

---
title: "Rent RTX 4070 Ti — $0.075/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4070 Ti (12 GB GDDR6X) from $0.037/hr interruptible or $0.075/hr on-demand — fixed, ≥30% below market. 17 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4070-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA RTX 4070 Ti — 12 GB, $0.075/hr on-demand

- VRAM 12 GB GDDR6X
- FP16 tensor 160 TFLOPS
- PowerScore 113 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 17 3 regions

[On-demand (guaranteed) $0.075 /GPU-hr ≈ $55/mo · market ~~$0.11~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4070-ti) [Interruptible $0.037 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4070-ti&type=spot) [Reserved 3 mo $0.048 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4070 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4070 Ti · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 12 GB GDDR6X · **Memory bandwidth**: 504 GB/s
- **FP16 tensor perf.**: 160 TFLOPS · **FP32 perf.**: 40.1 TFLOPS
- **CUDA cores**: 7,680 · **TDP**: 285 W
- **PowerScore ((RTX 3090 = 100))**: 113 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4070 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4070 Ti ($0.11/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.075, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.075 | $1.80 | $55 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.037 | $0.89 | $27 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.048 | $1.15 | $35 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4070 Ti (12 GB VRAM)

With 12 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.075.

- One-click template: [Ollama on a RTX 4070 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4070 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4070 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4070 Ti availability by region

17 × RTX 4070 Ti across 3 machines, live from inventory:

- Seoul
- Dallas, TX
- New York, NY

## RTX 4070 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4070 Ti** (this card) | 12 GB | 160 | $0.075 | $0.47‰ |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | 88 | $0.074 | $0.84‰ |
| [RTX 4070S](https://powergpu.ai/gpu/rtx-4070s) | 12 GB | 142 | $0.065 | $0.46‰ |
| [RTX 4070S Ti](https://powergpu.ai/gpu/rtx-4070s-ti) | 16 GB | 176 | $0.103 | $0.59‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4070 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4070 Ti per hour?**

$0.075 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.11. Interruptible capacity costs $0.037/hr and a 3-month reservation $0.048/hr. Around $55/month if you keep one running non-stop, billed per second.

**What can a RTX 4070 Ti with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4070 Ti available to rent right now?**

Yes — 17 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4070 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4070 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4070-ti --template pytorch.

**Why is the RTX 4070 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4070-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4060-ti -->

---
title: "Rent RTX 4060 Ti — $0.074/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4060 Ti (8 GB GDDR6X) from $0.037/hr interruptible or $0.074/hr on-demand — fixed, ≥30% below market. 69 GPUs in 14 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4060-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA RTX 4060 Ti — 8 GB, $0.074/hr on-demand

- VRAM 8 GB GDDR6X
- FP16 tensor 88 TFLOPS
- PowerScore 62 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 69 14 regions

[On-demand (guaranteed) $0.074 /GPU-hr ≈ $54/mo · market ~~$0.11~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4060-ti) [Interruptible $0.037 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4060-ti&type=spot) [Reserved 3 mo $0.048 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4060 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4060 Ti · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 8 GB GDDR6X · **Memory bandwidth**: 288 GB/s
- **FP16 tensor perf.**: 88 TFLOPS · **FP32 perf.**: 22.1 TFLOPS
- **CUDA cores**: 4,352 · **TDP**: 160 W
- **PowerScore ((RTX 3090 = 100))**: 62 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4060 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4060 Ti ($0.11/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.074, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.074 | $1.78 | $54 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.037 | $0.89 | $27 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.048 | $1.15 | $35 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4060 Ti (8 GB VRAM)

With 8 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.074.

- One-click template: [Ollama on a RTX 4060 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4060 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4060 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4060 Ti availability by region

69 × RTX 4060 Ti across 15 machines, live from inventory:

- Los Angeles, CA
- Bucharest
- Montréal
- Mumbai
- Milan
- Jakarta
- Tokyo
- Santiago
- Ashburn, VA
- Johannesburg
- Dubai
- New York, NY
- +2 more

## RTX 4060 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4060 Ti** (this card) | 8 GB | 88 | $0.074 | $0.84‰ |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | 160 | $0.075 | $0.47‰ |
| [RTX 4070S](https://powergpu.ai/gpu/rtx-4070s) | 12 GB | 142 | $0.065 | $0.46‰ |
| [RTX 4060](https://powergpu.ai/gpu/rtx-4060) | 8 GB | 60 | $0.047 | $0.78‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4060 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4060 Ti per hour?**

$0.074 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.11. Interruptible capacity costs $0.037/hr and a 3-month reservation $0.048/hr. Around $54/month if you keep one running non-stop, billed per second.

**What can a RTX 4060 Ti with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4060 Ti available to rent right now?**

Yes — 69 GPUs across 15 machines in 14 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4060 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4060 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4060-ti --template pytorch.

**Why is the RTX 4060 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4060-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-a4000 -->

---
title: "Rent RTX A4000 — $0.071/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX A4000 (16 GB GDDR6) from $0.035/hr interruptible or $0.071/hr on-demand — fixed, ≥30% below market. 256 GPUs in 17 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-a4000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA RTX A4000 — 16 GB, $0.071/hr on-demand

- VRAM 16 GB GDDR6
- FP16 tensor 76 TFLOPS
- PowerScore 54 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 256 17 regions

[On-demand (guaranteed) $0.071 /GPU-hr ≈ $52/mo · market ~~$0.10~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-a4000) [Interruptible $0.035 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-a4000&type=spot) [Reserved 3 mo $0.046 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ampere architecture

The RTX A4000 is a single-slot, 140 W Ampere workstation card with 16 GB of ECC GDDR6. Hosts run many of them per machine, which keeps the price low — a good fit for 7B 4-bit inference, CI pipelines and classic ML at a fraction of a 3090's rate.

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX A4000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX A4000 · **Architecture**: Ampere (2021)
- **VRAM**: 16 GB GDDR6 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 76 TFLOPS · **FP32 perf.**: 19.2 TFLOPS
- **CUDA cores**: 6,144 · **TDP**: 140 W
- **PowerScore ((RTX 3090 = 100))**: 54 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX A4000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX A4000 ($0.10/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.071, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.071 | $1.70 | $52 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.035 | $0.84 | $26 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.046 | $1.10 | $34 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX A4000 (16 GB VRAM)

With 16 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.071.

- One-click template: [Linux Desktop on a RTX A4000](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX A4000](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX A4000](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX A4000 availability by region

256 × RTX A4000 across 30 machines, live from inventory:

- Frankfurt
- Chicago, IL
- Amsterdam
- Tokyo
- New York, NY
- Paris
- Seattle, WA
- London
- Sydney
- Los Angeles, CA
- Mumbai
- Ashburn, VA
- +5 more

## RTX A4000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX A4000** (this card) | 16 GB | 76 | $0.071 | $0.93‰ |
| [Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB | 65 | $0.094 | $1.45‰ |
| [Quadro P4000](https://powergpu.ai/gpu/quadro-p4000) | 8 GB | 8 | $0.042 | $5.25‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |
| [RTX A2000](https://powergpu.ai/gpu/rtx-a2000) | 6 GB | 32 | $0.024 | $0.75‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX A4000: frequently asked questions

**How much does it cost to rent an NVIDIA RTX A4000 per hour?**

$0.071 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.10. Interruptible capacity costs $0.035/hr and a 3-month reservation $0.046/hr. Around $52/month if you keep one running non-stop, billed per second.

**What can a RTX A4000 with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX A4000 available to rent right now?**

Yes — 256 GPUs across 30 machines in 17 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX A4000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX A4000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-a4000 --template pytorch.

**Why is the RTX A4000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-a4000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3070-ti -->

---
title: "Rent RTX 3070 Ti — $0.066/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3070 Ti (8 GB GDDR6) from $0.033/hr interruptible or $0.066/hr on-demand — fixed, ≥30% below market. 15 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3070-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA RTX 3070 Ti — 8 GB, $0.066/hr on-demand

- VRAM 8 GB GDDR6
- FP16 tensor 87 TFLOPS
- PowerScore 61 RTX 3090 = 100
- Configs 1–2× PCIe 4.0
- Online now 15 2 regions

[On-demand (guaranteed) $0.066 /GPU-hr ≈ $48/mo · market ~~$0.09~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3070-ti) [Interruptible $0.033 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3070-ti&type=spot) [Reserved 3 mo $0.042 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ampere architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 3070 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3070 Ti · **Architecture**: Ampere (2021)
- **VRAM**: 8 GB GDDR6 · **Memory bandwidth**: 608 GB/s
- **FP16 tensor perf.**: 87 TFLOPS · **FP32 perf.**: 21.7 TFLOPS
- **CUDA cores**: 6,144 · **TDP**: 290 W
- **PowerScore ((RTX 3090 = 100))**: 61 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 2× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3070 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3070 Ti ($0.09/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.066, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.066 | $1.58 | $48 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.033 | $0.79 | $24 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.042 | $1.01 | $31 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 2× machine costs exactly 2× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3070 Ti (8 GB VRAM)

With 8 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 2× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.066.

- One-click template: [Ollama on a RTX 3070 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 3070 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 3070 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3070 Ti availability by region

15 × RTX 3070 Ti across 2 machines, live from inventory:

- Ashburn, VA
- Seoul

## RTX 3070 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3070 Ti** (this card) | 8 GB | 87 | $0.066 | $0.76‰ |
| [RTX 3080](https://powergpu.ai/gpu/rtx-3080) | 10 GB | 119 | $0.075 | $0.63‰ |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | 81 | $0.056 | $0.69‰ |
| [RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB | 65 | $0.047 | $0.72‰ |
| [RTX 3060 laptop](https://powergpu.ai/gpu/rtx-3060-laptop) | 6 GB | 43 | $0.047 | $1.09‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 3070 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3070 Ti per hour?**

$0.066 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.09. Interruptible capacity costs $0.033/hr and a 3-month reservation $0.042/hr. Around $48/month if you keep one running non-stop, billed per second.

**What can a RTX 3070 Ti with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 2× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3070 Ti available to rent right now?**

Yes — 15 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 2×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3070 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3070 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3070-ti --template pytorch.

**Why is the RTX 3070 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3070-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/tesla-p100 -->

---
title: "Rent Tesla P100 — $0.066/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Tesla P100 (16 GB HBM2) from $0.033/hr interruptible or $0.066/hr on-demand — fixed, ≥30% below market. 12 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/tesla-p100
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Datacenter · Pascal · launched 2016 · prices checked 2026-09-14

# Rent NVIDIA Tesla P100 — 16 GB, $0.066/hr on-demand

- VRAM 16 GB HBM2
- FP16 tensor 19 TFLOPS
- PowerScore 13 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 12 3 regions

[On-demand (guaranteed) $0.066 /GPU-hr ≈ $48/mo · market ~~$0.09~~ (−30%)](https://cloud.powergpu.ai/?gpu=tesla-p100) [Interruptible $0.033 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=tesla-p100&type=spot) [Reserved 3 mo $0.042 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Datacenter · Pascal architecture

A server-class card built for sustained 24/7 load: passive cooling in proper chassis, ECC memory, and drivers validated for compute. The sweet spot for inference fleets and fine-tuning jobs that need stability more than headline FLOPS.

## NVIDIA Tesla P100 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Tesla P100 · **Architecture**: Pascal (2016)
- **VRAM**: 16 GB HBM2 · **Memory bandwidth**: 732 GB/s
- **FP16 tensor perf.**: 19 TFLOPS · **FP32 perf.**: 9.3 TFLOPS
- **CUDA cores**: 3,584 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 13 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 1,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Tesla P100 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Tesla P100 ($0.09/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.066, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.066 | $1.58 | $48 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.033 | $0.79 | $24 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.042 | $1.01 | $31 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Tesla P100 (16 GB VRAM)

With 16 GB of HBM2, a single card holds a **~3B-parameter LLM in FP16** or up to **~24B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.066.

- One-click template: [vLLM on a Tesla P100](https://powergpu.ai/templates/vllm)
- One-click template: [PyTorch on a Tesla P100](https://powergpu.ai/templates/pytorch)
- One-click template: [ComfyUI on a Tesla P100](https://powergpu.ai/templates/comfyui)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Tesla P100 availability by region

12 × Tesla P100 across 3 machines, live from inventory:

- Stockholm
- London
- Frankfurt

## Tesla P100 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Tesla P100** (this card) | 16 GB | 19 | $0.066 | $3.47‰ |
| [Tesla P40](https://powergpu.ai/gpu/tesla-p40) | 24 GB | 12 | $0.047 | $3.92‰ |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | 65 | $0.103 | $1.58‰ |
| [Tesla P4](https://powergpu.ai/gpu/tesla-p4) | 8 GB | 6 | $0.018 | $3.00‰ |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | 125 | $0.130 | $1.04‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Tesla P100: frequently asked questions

**How much does it cost to rent an NVIDIA Tesla P100 per hour?**

$0.066 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.09. Interruptible capacity costs $0.033/hr and a 3-month reservation $0.042/hr. Around $48/month if you keep one running non-stop, billed per second.

**What can a Tesla P100 with 16 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~24B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Tesla P100 available to rent right now?**

Yes — 12 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Tesla P100?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Tesla P100, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu tesla-p100 --template pytorch.

**Why is the Tesla P100 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/tesla-p100 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-2070 -->

---
title: "Rent RTX 2070 — $0.065/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 2070 (8 GB GDDR6) from $0.032/hr interruptible or $0.065/hr on-demand — fixed, ≥30% below market. 8 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-2070
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2018 · prices checked 2026-09-14

# Rent NVIDIA RTX 2070 — 8 GB, $0.065/hr on-demand

- VRAM 8 GB GDDR6
- FP16 tensor 42 TFLOPS
- PowerScore 30 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 8 2 regions

[On-demand (guaranteed) $0.065 /GPU-hr ≈ $47/mo · market ~~$0.09~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-2070) [Interruptible $0.032 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-2070&type=spot) [Reserved 3 mo $0.042 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 2070 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 2070 · **Architecture**: Turing (2018)
- **VRAM**: 8 GB GDDR6 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 42 TFLOPS · **FP32 perf.**: 7.5 TFLOPS
- **CUDA cores**: 2,304 · **TDP**: 175 W
- **PowerScore ((RTX 3090 = 100))**: 30 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 2070 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 2070 ($0.09/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.065, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.065 | $1.56 | $47 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.032 | $0.77 | $23 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.042 | $1.01 | $31 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 2070 (8 GB VRAM)

With 8 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.065.

- One-click template: [Ollama on a RTX 2070](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 2070](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 2070](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 2070 availability by region

8 × RTX 2070 across 2 machines, live from inventory:

- Mumbai
- Singapore

## RTX 2070 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 2070** (this card) | 8 GB | 42 | $0.065 | $1.55‰ |
| [RTX 2080 Ti](https://powergpu.ai/gpu/rtx-2080-ti) | 11 GB | 57 | $0.057 | $1.00‰ |
| [RTX 2070S](https://powergpu.ai/gpu/rtx-2070s) | 8 GB | 45 | $0.056 | $1.24‰ |
| [RTX 2060](https://powergpu.ai/gpu/rtx-2060) | 6 GB | 26 | $0.051 | $1.96‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 2070: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 2070 per hour?**

$0.065 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.09. Interruptible capacity costs $0.032/hr and a 3-month reservation $0.042/hr. Around $47/month if you keep one running non-stop, billed per second.

**What can a RTX 2070 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 2070 available to rent right now?**

Yes — 8 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 2070?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 2070, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-2070 --template pytorch.

**Why is the RTX 2070 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-2070 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-2070s -->

---
title: "Rent RTX 2070S — $0.056/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 2070S (8 GB GDDR6) from $0.028/hr interruptible or $0.056/hr on-demand — fixed, ≥30% below market. 2 GPUs in 5 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-2070s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2019 · prices checked 2026-09-14

# Rent NVIDIA RTX 2070S — 8 GB, $0.056/hr on-demand

- VRAM 8 GB GDDR6
- FP16 tensor 45 TFLOPS
- PowerScore 32 RTX 3090 = 100
- Configs 1–4× PCIe 3.0
- Online now 2 5 regions

[On-demand (guaranteed) $0.056 /GPU-hr ≈ $41/mo · market ~~$0.08~~ (−31%)](https://cloud.powergpu.ai/?gpu=rtx-2070s) [Interruptible $0.028 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-2070s&type=spot) [Reserved 3 mo $0.036 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 2070S specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 2070S · **Architecture**: Turing (2019)
- **VRAM**: 8 GB GDDR6 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 45 TFLOPS · **FP32 perf.**: 9.1 TFLOPS
- **CUDA cores**: 2,560 · **TDP**: 215 W
- **PowerScore ((RTX 3090 = 100))**: 32 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 2070S price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 2070S ($0.08/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.056, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.056 | $1.34 | $41 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.028 | $0.67 | $20 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.036 | $0.86 | $26 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 2070S (8 GB VRAM)

With 8 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.056.

- One-click template: [Ollama on a RTX 2070S](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 2070S](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 2070S](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 2070S availability by region

2 × RTX 2070S across 5 machines, live from inventory:

- Amsterdam
- Milan
- Montréal
- Ashburn, VA
- Stockholm

## RTX 2070S vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 2070S** (this card) | 8 GB | 45 | $0.056 | $1.24‰ |
| [RTX 2080 Ti](https://powergpu.ai/gpu/rtx-2080-ti) | 11 GB | 57 | $0.057 | $1.00‰ |
| [RTX 2060](https://powergpu.ai/gpu/rtx-2060) | 6 GB | 26 | $0.051 | $1.96‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |
| [GTX 1660 Ti](https://powergpu.ai/gpu/gtx-1660-ti) | 6 GB | 5 | $0.048 | $8.89‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 2070S: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 2070S per hour?**

$0.056 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.08. Interruptible capacity costs $0.028/hr and a 3-month reservation $0.036/hr. Around $41/month if you keep one running non-stop, billed per second.

**What can a RTX 2070S with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 2070S available to rent right now?**

Yes — 2 GPUs across 5 machines in 5 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 2070S?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 2070S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-2070s --template pytorch.

**Why is the RTX 2070S cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-2070s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4070s -->

---
title: "Rent RTX 4070S — $0.065/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4070S (12 GB GDDR6X) from $0.032/hr interruptible or $0.065/hr on-demand — fixed, ≥30% below market. 64 GPUs in 11 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4070s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2024 · prices checked 2026-09-14

# Rent NVIDIA RTX 4070S — 12 GB, $0.065/hr on-demand

- VRAM 12 GB GDDR6X
- FP16 tensor 142 TFLOPS
- PowerScore 100 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 64 11 regions

[On-demand (guaranteed) $0.065 /GPU-hr ≈ $47/mo · market ~~$0.09~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4070s) [Interruptible $0.032 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4070s&type=spot) [Reserved 3 mo $0.042 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4070S specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4070S · **Architecture**: Ada Lovelace (2024)
- **VRAM**: 12 GB GDDR6X · **Memory bandwidth**: 504 GB/s
- **FP16 tensor perf.**: 142 TFLOPS · **FP32 perf.**: 35.5 TFLOPS
- **CUDA cores**: 7,168 · **TDP**: 220 W
- **PowerScore ((RTX 3090 = 100))**: 100 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4070S price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4070S ($0.09/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.065, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.065 | $1.56 | $47 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.032 | $0.77 | $23 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.042 | $1.01 | $31 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4070S (12 GB VRAM)

With 12 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.065.

- One-click template: [Ollama on a RTX 4070S](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4070S](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4070S](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4070S availability by region

64 × RTX 4070S across 12 machines, live from inventory:

- Tokyo
- Sydney
- London
- Mumbai
- Johannesburg
- Osaka
- Dallas, TX
- Amsterdam
- Warsaw
- Montréal
- Vancouver

## RTX 4070S vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4070S** (this card) | 12 GB | 142 | $0.065 | $0.46‰ |
| [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | 88 | $0.074 | $0.84‰ |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | 160 | $0.075 | $0.47‰ |
| [RTX 4060](https://powergpu.ai/gpu/rtx-4060) | 8 GB | 60 | $0.047 | $0.78‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4070S: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4070S per hour?**

$0.065 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.09. Interruptible capacity costs $0.032/hr and a 3-month reservation $0.042/hr. Around $47/month if you keep one running non-stop, billed per second.

**What can a RTX 4070S with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4070S available to rent right now?**

Yes — 64 GPUs across 12 machines in 11 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4070S?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4070S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4070s --template pytorch.

**Why is the RTX 4070S cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4070s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-5060 -->

---
title: "Rent RTX 5060 — $0.063/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 5060 (8 GB GDDR7) from $0.031/hr interruptible or $0.063/hr on-demand — fixed, ≥30% below market. 14 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-5060
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Blackwell · launched 2025 · prices checked 2026-09-14

# Rent NVIDIA RTX 5060 — 8 GB, $0.063/hr on-demand

- VRAM 8 GB GDDR7
- FP16 tensor 74 TFLOPS
- PowerScore 52 RTX 3090 = 100
- Configs 1–8× PCIe 5.0
- Online now 14 3 regions

[On-demand (guaranteed) $0.063 /GPU-hr ≈ $46/mo · market ~~$0.09~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-5060) [Interruptible $0.031 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-5060&type=spot) [Reserved 3 mo $0.040 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Blackwell architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 5060 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 5060 · **Architecture**: Blackwell (2025)
- **VRAM**: 8 GB GDDR7 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 74 TFLOPS · **FP32 perf.**: 19.2 TFLOPS
- **CUDA cores**: 3,840 · **TDP**: 145 W
- **PowerScore ((RTX 3090 = 100))**: 52 · **PCIe generation**: Gen 5.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 5060 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 5060 ($0.09/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.063, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.063 | $1.51 | $46 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.031 | $0.74 | $23 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.040 | $0.96 | $29 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 5060 (8 GB VRAM)

With 8 GB of GDDR7, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.063.

- One-click template: [Ollama on a RTX 5060](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 5060](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 5060](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 5060 availability by region

14 × RTX 5060 across 3 machines, live from inventory:

- Sydney
- Montréal
- Tel Aviv

## RTX 5060 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 5060** (this card) | 8 GB | 74 | $0.063 | $0.85‰ |
| [RTX 5060 Ti](https://powergpu.ai/gpu/rtx-5060-ti) | 16 GB | 92 | $0.112 | $1.22‰ |
| [RTX 5070](https://powergpu.ai/gpu/rtx-5070) | 12 GB | 123 | $0.112 | $0.91‰ |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) | 16 GB | 176 | $0.131 | $0.74‰ |
| [RTX 5080](https://powergpu.ai/gpu/rtx-5080) | 16 GB | 225 | $0.186 | $0.83‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 5060: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 5060 per hour?**

$0.063 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.09. Interruptible capacity costs $0.031/hr and a 3-month reservation $0.040/hr. Around $46/month if you keep one running non-stop, billed per second.

**What can a RTX 5060 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 5060 available to rent right now?**

Yes — 14 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 5060?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 5060, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-5060 --template pytorch.

**Why is the RTX 5060 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-5060 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-2080-ti -->

---
title: "Rent RTX 2080 Ti — $0.057/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 2080 Ti (11 GB GDDR6) from $0.028/hr interruptible or $0.057/hr on-demand — fixed, ≥30% below market. 37 GPUs in 9 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-2080-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2018 · prices checked 2026-09-14

# Rent NVIDIA RTX 2080 Ti — 11 GB, $0.057/hr on-demand

- VRAM 11 GB GDDR6
- FP16 tensor 57 TFLOPS
- PowerScore 40 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 37 9 regions

[On-demand (guaranteed) $0.057 /GPU-hr ≈ $42/mo · market ~~$0.08~~ (−31%)](https://cloud.powergpu.ai/?gpu=rtx-2080-ti) [Interruptible $0.028 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-2080-ti&type=spot) [Reserved 3 mo $0.037 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

The RTX 2080 Ti was the first consumer card with tensor cores: 11 GB GDDR6, 4,352 CUDA cores. Today it is a budget option for older CUDA workloads, small-model inference and classic vision models.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 2080 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 2080 Ti · **Architecture**: Turing (2018)
- **VRAM**: 11 GB GDDR6 · **Memory bandwidth**: 616 GB/s
- **FP16 tensor perf.**: 57 TFLOPS · **FP32 perf.**: 13.4 TFLOPS
- **CUDA cores**: 4,352 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 40 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 2080 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 2080 Ti ($0.08/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.057, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.057 | $1.37 | $42 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.028 | $0.67 | $20 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.037 | $0.89 | $27 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 2080 Ti (11 GB VRAM)

With 11 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.057.

- One-click template: [Ollama on a RTX 2080 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 2080 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 2080 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 2080 Ti availability by region

37 × RTX 2080 Ti across 11 machines, live from inventory:

- Tokyo
- São Paulo
- Santiago
- Montréal
- Bucharest
- Osaka
- Frankfurt
- New York, NY
- Dubai

## RTX 2080 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 2080 Ti** (this card) | 11 GB | 57 | $0.057 | $1.00‰ |
| [RTX 2070S](https://powergpu.ai/gpu/rtx-2070s) | 8 GB | 45 | $0.056 | $1.24‰ |
| [RTX 2060](https://powergpu.ai/gpu/rtx-2060) | 6 GB | 26 | $0.051 | $1.96‰ |
| [RTX 2070](https://powergpu.ai/gpu/rtx-2070) | 8 GB | 42 | $0.065 | $1.55‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 2080 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 2080 Ti per hour?**

$0.057 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.08. Interruptible capacity costs $0.028/hr and a 3-month reservation $0.037/hr. Around $42/month if you keep one running non-stop, billed per second.

**What can a RTX 2080 Ti with 11 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 2080 Ti available to rent right now?**

Yes — 37 GPUs across 11 machines in 9 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 2080 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 2080 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-2080-ti --template pytorch.

**Why is the RTX 2080 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-2080-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1660 -->

---
title: "Rent GTX 1660 — $0.048/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1660 (6 GB GDDR6) from $0.024/hr interruptible or $0.048/hr on-demand — fixed, ≥30% below market. 4 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1660
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2019 · prices checked 2026-09-14

# Rent NVIDIA GTX 1660 — 6 GB, $0.048/hr on-demand

- VRAM 6 GB GDDR6
- FP16 tensor 5 TFLOPS
- PowerScore 4 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 4 2 regions

[On-demand (guaranteed) $0.048 /GPU-hr ≈ $35/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=gtx-1660) [Interruptible $0.024 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1660&type=spot) [Reserved 3 mo $0.031 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1660 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1660 · **Architecture**: Turing (2019)
- **VRAM**: 6 GB GDDR6 · **Memory bandwidth**: 192 GB/s
- **FP16 tensor perf.**: 5 TFLOPS · **FP32 perf.**: 5.0 TFLOPS
- **CUDA cores**: 1,408 · **TDP**: 120 W
- **PowerScore ((RTX 3090 = 100))**: 4 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1660 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1660 ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.048, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.048 | $1.15 | $35 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.024 | $0.58 | $18 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.031 | $0.74 | $23 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1660 (6 GB VRAM)

With 6 GB of GDDR6, a single card holds a **~1B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.048.

- One-click template: [Ollama on a GTX 1660](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1660](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1660](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1660 availability by region

4 × GTX 1660 across 2 machines, live from inventory:

- Stockholm
- Chicago, IL

## GTX 1660 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1660** (this card) | 6 GB | 5 | $0.048 | $9.60‰ |
| [GTX 1660 Ti](https://powergpu.ai/gpu/gtx-1660-ti) | 6 GB | 5 | $0.048 | $8.89‰ |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | 11 | $0.047 | $4.27‰ |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | 4 | $0.046 | $10.45‰ |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | 8 | $0.046 | $5.75‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1660: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1660 per hour?**

$0.048 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.024/hr and a 3-month reservation $0.031/hr. Around $35/month if you keep one running non-stop, billed per second.

**What can a GTX 1660 with 6 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1660 available to rent right now?**

Yes — 4 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1660?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1660, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1660 --template pytorch.

**Why is the GTX 1660 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1660 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3070 -->

---
title: "Rent RTX 3070 — $0.056/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3070 (8 GB GDDR6) from $0.028/hr interruptible or $0.056/hr on-demand — fixed, ≥30% below market. 93 GPUs in 12 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3070
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ampere · launched 2020 · prices checked 2026-09-14

# Rent NVIDIA RTX 3070 — 8 GB, $0.056/hr on-demand

- VRAM 8 GB GDDR6
- FP16 tensor 81 TFLOPS
- PowerScore 57 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 93 12 regions

[On-demand (guaranteed) $0.056 /GPU-hr ≈ $41/mo · market ~~$0.08~~ (−31%)](https://cloud.powergpu.ai/?gpu=rtx-3070) [Interruptible $0.028 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3070&type=spot) [Reserved 3 mo $0.036 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ampere architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 3070 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3070 · **Architecture**: Ampere (2020)
- **VRAM**: 8 GB GDDR6 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 81 TFLOPS · **FP32 perf.**: 20.3 TFLOPS
- **CUDA cores**: 5,888 · **TDP**: 220 W
- **PowerScore ((RTX 3090 = 100))**: 57 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3070 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3070 ($0.08/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.056, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.056 | $1.34 | $41 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.028 | $0.67 | $20 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.036 | $0.86 | $26 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3070 (8 GB VRAM)

With 8 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.056.

- One-click template: [Ollama on a RTX 3070](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 3070](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 3070](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3070 availability by region

93 × RTX 3070 across 14 machines, live from inventory:

- Chicago, IL
- Warsaw
- São Paulo
- Madrid
- Seattle, WA
- Sydney
- Bucharest
- Mumbai
- Vancouver
- Querétaro
- London
- Johannesburg

## RTX 3070 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3070** (this card) | 8 GB | 81 | $0.056 | $0.69‰ |
| [RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB | 65 | $0.047 | $0.72‰ |
| [RTX 3060 laptop](https://powergpu.ai/gpu/rtx-3060-laptop) | 6 GB | 43 | $0.047 | $1.09‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |
| [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | 51 | $0.042 | $0.82‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 3070: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3070 per hour?**

$0.056 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.08. Interruptible capacity costs $0.028/hr and a 3-month reservation $0.036/hr. Around $41/month if you keep one running non-stop, billed per second.

**What can a RTX 3070 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3070 available to rent right now?**

Yes — 93 GPUs across 14 machines in 12 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3070?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3070, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3070 --template pytorch.

**Why is the RTX 3070 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3070 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-2060 -->

---
title: "Rent RTX 2060 — $0.051/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 2060 (6 GB GDDR6) from $0.025/hr interruptible or $0.051/hr on-demand — fixed, ≥30% below market. 6 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-2060
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2019 · prices checked 2026-09-14

# Rent NVIDIA RTX 2060 — 6 GB, $0.051/hr on-demand

- VRAM 6 GB GDDR6
- FP16 tensor 26 TFLOPS
- PowerScore 18 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 6 3 regions

[On-demand (guaranteed) $0.051 /GPU-hr ≈ $37/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-2060) [Interruptible $0.025 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-2060&type=spot) [Reserved 3 mo $0.033 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 2060 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 2060 · **Architecture**: Turing (2019)
- **VRAM**: 6 GB GDDR6 · **Memory bandwidth**: 336 GB/s
- **FP16 tensor perf.**: 26 TFLOPS · **FP32 perf.**: 6.5 TFLOPS
- **CUDA cores**: 1,920 · **TDP**: 160 W
- **PowerScore ((RTX 3090 = 100))**: 18 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 750 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 2060 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 2060 ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.051, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.051 | $1.22 | $37 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.025 | $0.60 | $18 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.033 | $0.79 | $24 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 2060 (6 GB VRAM)

With 6 GB of GDDR6, a single card holds a **~1B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.051.

- One-click template: [Ollama on a RTX 2060](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 2060](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 2060](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 2060 availability by region

6 × RTX 2060 across 3 machines, live from inventory:

- Vancouver
- Singapore
- Seoul

## RTX 2060 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 2060** (this card) | 6 GB | 26 | $0.051 | $1.96‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |
| [GTX 1660 Ti](https://powergpu.ai/gpu/gtx-1660-ti) | 6 GB | 5 | $0.048 | $8.89‰ |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | 11 | $0.047 | $4.27‰ |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | 4 | $0.046 | $10.45‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 2060: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 2060 per hour?**

$0.051 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.025/hr and a 3-month reservation $0.033/hr. Around $37/month if you keep one running non-stop, billed per second.

**What can a RTX 2060 with 6 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 2060 available to rent right now?**

Yes — 6 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 2060?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 2060, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-2060 --template pytorch.

**Why is the RTX 2060 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-2060 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1660-ti -->

---
title: "Rent GTX 1660 Ti — $0.048/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1660 Ti (6 GB GDDR6) from $0.024/hr interruptible or $0.048/hr on-demand — fixed, ≥30% below market. 10 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1660-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2019 · prices checked 2026-09-14

# Rent NVIDIA GTX 1660 Ti — 6 GB, $0.048/hr on-demand

- VRAM 6 GB GDDR6
- FP16 tensor 5 TFLOPS
- PowerScore 4 RTX 3090 = 100
- Configs 1–4× PCIe 3.0
- Online now 10 2 regions

[On-demand (guaranteed) $0.048 /GPU-hr ≈ $35/mo · market ~~$0.07~~ (−31%)](https://cloud.powergpu.ai/?gpu=gtx-1660-ti) [Interruptible $0.024 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1660-ti&type=spot) [Reserved 3 mo $0.031 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1660 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1660 Ti · **Architecture**: Turing (2019)
- **VRAM**: 6 GB GDDR6 · **Memory bandwidth**: 288 GB/s
- **FP16 tensor perf.**: 5 TFLOPS · **FP32 perf.**: 5.4 TFLOPS
- **CUDA cores**: 1,536 · **TDP**: 120 W
- **PowerScore ((RTX 3090 = 100))**: 4 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 300 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1660 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1660 Ti ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.048, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.048 | $1.15 | $35 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.024 | $0.58 | $18 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.031 | $0.74 | $23 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1660 Ti (6 GB VRAM)

With 6 GB of GDDR6, a single card holds a **~1B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.048.

- One-click template: [Ollama on a GTX 1660 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1660 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1660 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1660 Ti availability by region

10 × GTX 1660 Ti across 2 machines, live from inventory:

- Seoul
- Jakarta

## GTX 1660 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1660 Ti** (this card) | 6 GB | 5 | $0.048 | $8.89‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | 11 | $0.047 | $4.27‰ |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | 4 | $0.046 | $10.45‰ |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | 8 | $0.046 | $5.75‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1660 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1660 Ti per hour?**

$0.048 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.024/hr and a 3-month reservation $0.031/hr. Around $35/month if you keep one running non-stop, billed per second.

**What can a GTX 1660 Ti with 6 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1660 Ti available to rent right now?**

Yes — 10 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1660 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1660 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1660-ti --template pytorch.

**Why is the GTX 1660 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1660-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1080-ti -->

---
title: "Rent GTX 1080 Ti — $0.047/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1080 Ti (11 GB GDDR5X) from $0.023/hr interruptible or $0.047/hr on-demand — fixed, ≥30% below market. 22 GPUs in 10 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1080-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2017 · prices checked 2026-09-14

# Rent NVIDIA GTX 1080 Ti — 11 GB, $0.047/hr on-demand

- VRAM 11 GB GDDR5X
- FP16 tensor 11 TFLOPS
- PowerScore 8 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 22 10 regions

[On-demand (guaranteed) $0.047 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=gtx-1080-ti) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1080-ti&type=spot) [Reserved 3 mo $0.030 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

The GTX 1080 Ti has no tensor cores, but 11 GB of GDDR5X and 3,584 Pascal CUDA cores still cover classic ML, CUDA coursework and FP32 batch jobs at one of the lowest prices on the sheet.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1080 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1080 Ti · **Architecture**: Pascal (2017)
- **VRAM**: 11 GB GDDR5X · **Memory bandwidth**: 484 GB/s
- **FP16 tensor perf.**: 11 TFLOPS · **FP32 perf.**: 11.3 TFLOPS
- **CUDA cores**: 3,584 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 8 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1080 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1080 Ti ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.047, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.047 | $1.13 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.030 | $0.72 | $22 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1080 Ti (11 GB VRAM)

With 11 GB of GDDR5X, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.047.

- One-click template: [Ollama on a GTX 1080 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1080 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1080 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1080 Ti availability by region

22 × GTX 1080 Ti across 10 machines, live from inventory:

- Chicago, IL
- Dubai
- Ashburn, VA
- Jakarta
- Warsaw
- Miami, FL
- São Paulo
- Madrid
- Milan
- Seattle, WA

## GTX 1080 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1080 Ti** (this card) | 11 GB | 11 | $0.047 | $4.27‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |
| [GTX 1660 Ti](https://powergpu.ai/gpu/gtx-1660-ti) | 6 GB | 5 | $0.048 | $8.89‰ |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | 4 | $0.046 | $10.45‰ |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | 8 | $0.046 | $5.75‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1080 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1080 Ti per hour?**

$0.047 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.030/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a GTX 1080 Ti with 11 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1080 Ti available to rent right now?**

Yes — 22 GPUs across 10 machines in 10 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1080 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1080 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1080-ti --template pytorch.

**Why is the GTX 1080 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1080-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3060-ti -->

---
title: "Rent RTX 3060 Ti — $0.047/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3060 Ti (8 GB GDDR6) from $0.023/hr interruptible or $0.047/hr on-demand — fixed, ≥30% below market. 52 GPUs in 7 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3060-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ampere · launched 2020 · prices checked 2026-09-14

# Rent NVIDIA RTX 3060 Ti — 8 GB, $0.047/hr on-demand

- VRAM 8 GB GDDR6
- FP16 tensor 65 TFLOPS
- PowerScore 46 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 52 7 regions

[On-demand (guaranteed) $0.047 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3060-ti) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3060-ti&type=spot) [Reserved 3 mo $0.030 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ampere architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 3060 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3060 Ti · **Architecture**: Ampere (2020)
- **VRAM**: 8 GB GDDR6 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 65 TFLOPS · **FP32 perf.**: 16.2 TFLOPS
- **CUDA cores**: 4,864 · **TDP**: 200 W
- **PowerScore ((RTX 3090 = 100))**: 46 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 1,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3060 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3060 Ti ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.047, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.047 | $1.13 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.030 | $0.72 | $22 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3060 Ti (8 GB VRAM)

With 8 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.047.

- One-click template: [Ollama on a RTX 3060 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 3060 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 3060 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3060 Ti availability by region

52 × RTX 3060 Ti across 7 machines, live from inventory:

- Mumbai
- Chicago, IL
- Querétaro
- Vancouver
- Montréal
- Bucharest
- Dallas, TX

## RTX 3060 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3060 Ti** (this card) | 8 GB | 65 | $0.047 | $0.72‰ |
| [RTX 3060 laptop](https://powergpu.ai/gpu/rtx-3060-laptop) | 6 GB | 43 | $0.047 | $1.09‰ |
| [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | 51 | $0.042 | $0.82‰ |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | 81 | $0.056 | $0.69‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 3060 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3060 Ti per hour?**

$0.047 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.030/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a RTX 3060 Ti with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3060 Ti available to rent right now?**

Yes — 52 GPUs across 7 machines in 7 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3060 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3060 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3060-ti --template pytorch.

**Why is the RTX 3060 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3060-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3060-laptop -->

---
title: "Rent RTX 3060 laptop — $0.047/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3060 laptop (6 GB GDDR6) from $0.023/hr interruptible or $0.047/hr on-demand — fixed, ≥30% below market. 36 GPUs in 6 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3060-laptop
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA RTX 3060 laptop — 6 GB, $0.047/hr on-demand

- VRAM 6 GB GDDR6
- FP16 tensor 43 TFLOPS
- PowerScore 30 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 36 6 regions

[On-demand (guaranteed) $0.047 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3060-laptop) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3060-laptop&type=spot) [Reserved 3 mo $0.030 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ampere architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 3060 laptop specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3060 laptop · **Architecture**: Ampere (2021)
- **VRAM**: 6 GB GDDR6 · **Memory bandwidth**: 336 GB/s
- **FP16 tensor perf.**: 43 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: 3,840 · **TDP**: 115 W
- **PowerScore ((RTX 3090 = 100))**: 30 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 1,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3060 laptop price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3060 laptop ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.047, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.047 | $1.13 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.030 | $0.72 | $22 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3060 laptop (6 GB VRAM)

With 6 GB of GDDR6, a single card holds a **~1B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.047.

- One-click template: [Ollama on a RTX 3060 laptop](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 3060 laptop](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 3060 laptop](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3060 laptop availability by region

36 × RTX 3060 laptop across 8 machines, live from inventory:

- Tel Aviv
- Madrid
- Stockholm
- Dallas, TX
- Dubai
- Sydney

## RTX 3060 laptop vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3060 laptop** (this card) | 6 GB | 43 | $0.047 | $1.09‰ |
| [RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB | 65 | $0.047 | $0.72‰ |
| [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | 51 | $0.042 | $0.82‰ |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | 81 | $0.056 | $0.69‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 3060 laptop: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3060 laptop per hour?**

$0.047 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.030/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a RTX 3060 laptop with 6 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3060 laptop available to rent right now?**

Yes — 36 GPUs across 8 machines in 6 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3060 laptop?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3060 laptop, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3060-laptop --template pytorch.

**Why is the RTX 3060 laptop cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3060-laptop · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-4060 -->

---
title: "Rent RTX 4060 — $0.047/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 4060 (8 GB GDDR6X) from $0.023/hr interruptible or $0.047/hr on-demand — fixed, ≥30% below market. 61 GPUs in 5 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-4060
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ada Lovelace · launched 2023 · prices checked 2026-09-14

# Rent NVIDIA RTX 4060 — 8 GB, $0.047/hr on-demand

- VRAM 8 GB GDDR6X
- FP16 tensor 60 TFLOPS
- PowerScore 42 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 61 5 regions

[On-demand (guaranteed) $0.047 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-4060) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-4060&type=spot) [Reserved 3 mo $0.030 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ada Lovelace architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 4060 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 4060 · **Architecture**: Ada Lovelace (2023)
- **VRAM**: 8 GB GDDR6X · **Memory bandwidth**: 272 GB/s
- **FP16 tensor perf.**: 60 TFLOPS · **FP32 perf.**: 15.1 TFLOPS
- **CUDA cores**: 3,072 · **TDP**: 115 W
- **PowerScore ((RTX 3090 = 100))**: 42 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 4060 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 4060 ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.047, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.047 | $1.13 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.030 | $0.72 | $22 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 4060 (8 GB VRAM)

With 8 GB of GDDR6X, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.047.

- One-click template: [Ollama on a RTX 4060](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 4060](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 4060](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 4060 availability by region

61 × RTX 4060 across 5 machines, live from inventory:

- Montréal
- Chicago, IL
- São Paulo
- Seoul
- Warsaw

## RTX 4060 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 4060** (this card) | 8 GB | 60 | $0.047 | $0.78‰ |
| [RTX 4070S](https://powergpu.ai/gpu/rtx-4070s) | 12 GB | 142 | $0.065 | $0.46‰ |
| [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | 88 | $0.074 | $0.84‰ |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | 117 | $0.075 | $0.64‰ |
| [RTX 4070 Ti](https://powergpu.ai/gpu/rtx-4070-ti) | 12 GB | 160 | $0.075 | $0.47‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 4060: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 4060 per hour?**

$0.047 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.030/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a RTX 4060 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 4060 available to rent right now?**

Yes — 61 GPUs across 5 machines in 5 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 4060?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 4060, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-4060 --template pytorch.

**Why is the RTX 4060 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-4060 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/tesla-p40 -->

---
title: "Rent Tesla P40 — $0.047/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Tesla P40 (24 GB GDDR5X) from $0.023/hr interruptible or $0.047/hr on-demand — fixed, ≥30% below market. 4 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/tesla-p40
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2016 · prices checked 2026-09-14

# Rent NVIDIA Tesla P40 — 24 GB, $0.047/hr on-demand

- VRAM 24 GB GDDR5X
- FP16 tensor 12 TFLOPS
- PowerScore 8 RTX 3090 = 100
- Configs 1–4× PCIe 3.0
- Online now 4 2 regions

[On-demand (guaranteed) $0.047 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=tesla-p40) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=tesla-p40&type=spot) [Reserved 3 mo $0.030 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA Tesla P40 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Tesla P40 · **Architecture**: Pascal (2016)
- **VRAM**: 24 GB GDDR5X · **Memory bandwidth**: 346 GB/s
- **FP16 tensor perf.**: 12 TFLOPS · **FP32 perf.**: 11.8 TFLOPS
- **CUDA cores**: 3,840 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 8 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 2,000 GB NVMe
- **Network up to**: 5,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Tesla P40 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Tesla P40 ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.047, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.047 | $1.13 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.030 | $0.72 | $22 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Tesla P40 (24 GB VRAM)

With 24 GB of GDDR5X, a single card holds a **~8B-parameter LLM in FP16** or up to **~32B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.047.

- One-click template: [Ollama on a Tesla P40](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a Tesla P40](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a Tesla P40](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Tesla P40 availability by region

4 × Tesla P40 across 2 machines, live from inventory:

- Ashburn, VA
- Mumbai

## Tesla P40 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Tesla P40** (this card) | 24 GB | 12 | $0.047 | $3.92‰ |
| [Tesla P100](https://powergpu.ai/gpu/tesla-p100) | 16 GB | 19 | $0.066 | $3.47‰ |
| [Tesla P4](https://powergpu.ai/gpu/tesla-p4) | 8 GB | 6 | $0.018 | $3.00‰ |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | 65 | $0.103 | $1.58‰ |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | 125 | $0.130 | $1.04‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Tesla P40: frequently asked questions

**How much does it cost to rent an NVIDIA Tesla P40 per hour?**

$0.047 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.030/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a Tesla P40 with 24 GB VRAM run?**

In LLM terms, roughly a 8B-parameter model in FP16 or up to ~32B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Tesla P40 available to rent right now?**

Yes — 4 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Tesla P40?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Tesla P40, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu tesla-p40 --template pytorch.

**Why is the Tesla P40 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/tesla-p40 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1060 -->

---
title: "Rent GTX 1060 — $0.046/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1060 (6 GB GDDR5X) from $0.023/hr interruptible or $0.046/hr on-demand — fixed, ≥30% below market. 3 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1060
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2016 · prices checked 2026-09-14

# Rent NVIDIA GTX 1060 — 6 GB, $0.046/hr on-demand

- VRAM 6 GB GDDR5X
- FP16 tensor 4 TFLOPS
- PowerScore 3 RTX 3090 = 100
- Configs 1–1× PCIe 3.0
- Online now 3 2 regions

[On-demand (guaranteed) $0.046 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−31%)](https://cloud.powergpu.ai/?gpu=gtx-1060) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1060&type=spot) [Reserved 3 mo $0.029 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1060 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1060 · **Architecture**: Pascal (2016)
- **VRAM**: 6 GB GDDR5X · **Memory bandwidth**: 192 GB/s
- **FP16 tensor perf.**: 4 TFLOPS · **FP32 perf.**: 4.4 TFLOPS
- **CUDA cores**: 1,280 · **TDP**: 120 W
- **PowerScore ((RTX 3090 = 100))**: 3 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 1× · **Max instance storage**: 1,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1060 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1060 ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.046, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.046 | $1.10 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.029 | $0.70 | $21 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 1× machine costs exactly 1× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1060 (6 GB VRAM)

With 6 GB of GDDR5X, a single card holds a **~1B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 1× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.046.

- One-click template: [Ollama on a GTX 1060](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1060](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1060](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1060 availability by region

3 × GTX 1060 across 2 machines, live from inventory:

- Dallas, TX
- New York, NY

## GTX 1060 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1060** (this card) | 6 GB | 4 | $0.046 | $10.45‰ |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | 8 | $0.046 | $5.75‰ |
| [RTX 2060S](https://powergpu.ai/gpu/rtx-2060s) | 8 GB | 40 | $0.046 | $1.15‰ |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | 11 | $0.047 | $4.27‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1060: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1060 per hour?**

$0.046 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.029/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a GTX 1060 with 6 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 1× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1060 available to rent right now?**

Yes — 3 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 1×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1060?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1060, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1060 --template pytorch.

**Why is the GTX 1060 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1060 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1070-ti -->

---
title: "Rent GTX 1070 Ti — $0.046/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1070 Ti (8 GB GDDR5X) from $0.023/hr interruptible or $0.046/hr on-demand — fixed, ≥30% below market. 25 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1070-ti
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2017 · prices checked 2026-09-14

# Rent NVIDIA GTX 1070 Ti — 8 GB, $0.046/hr on-demand

- VRAM 8 GB GDDR5X
- FP16 tensor 8 TFLOPS
- PowerScore 6 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 25 3 regions

[On-demand (guaranteed) $0.046 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−31%)](https://cloud.powergpu.ai/?gpu=gtx-1070-ti) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1070-ti&type=spot) [Reserved 3 mo $0.029 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1070 Ti specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1070 Ti · **Architecture**: Pascal (2017)
- **VRAM**: 8 GB GDDR5X · **Memory bandwidth**: 256 GB/s
- **FP16 tensor perf.**: 8 TFLOPS · **FP32 perf.**: 8.2 TFLOPS
- **CUDA cores**: 2,432 · **TDP**: 180 W
- **PowerScore ((RTX 3090 = 100))**: 6 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1070 Ti price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1070 Ti ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.046, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.046 | $1.10 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.029 | $0.70 | $21 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1070 Ti (8 GB VRAM)

With 8 GB of GDDR5X, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.046.

- One-click template: [Ollama on a GTX 1070 Ti](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1070 Ti](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1070 Ti](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1070 Ti availability by region

25 × GTX 1070 Ti across 5 machines, live from inventory:

- London
- Paris
- Johannesburg

## GTX 1070 Ti vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1070 Ti** (this card) | 8 GB | 8 | $0.046 | $5.75‰ |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | 4 | $0.046 | $10.45‰ |
| [RTX 2060S](https://powergpu.ai/gpu/rtx-2060s) | 8 GB | 40 | $0.046 | $1.15‰ |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | 11 | $0.047 | $4.27‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1070 Ti: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1070 Ti per hour?**

$0.046 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.029/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a GTX 1070 Ti with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1070 Ti available to rent right now?**

Yes — 25 GPUs across 5 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1070 Ti?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1070 Ti, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1070-ti --template pytorch.

**Why is the GTX 1070 Ti cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1070-ti · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-2060s -->

---
title: "Rent RTX 2060S — $0.046/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 2060S (8 GB GDDR6) from $0.023/hr interruptible or $0.046/hr on-demand — fixed, ≥30% below market. 6 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-2060s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2019 · prices checked 2026-09-14

# Rent NVIDIA RTX 2060S — 8 GB, $0.046/hr on-demand

- VRAM 8 GB GDDR6
- FP16 tensor 40 TFLOPS
- PowerScore 28 RTX 3090 = 100
- Configs 1–4× PCIe 3.0
- Online now 6 2 regions

[On-demand (guaranteed) $0.046 /GPU-hr ≈ $34/mo · market ~~$0.07~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-2060s) [Interruptible $0.023 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-2060s&type=spot) [Reserved 3 mo $0.029 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 2060S specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 2060S · **Architecture**: Turing (2019)
- **VRAM**: 8 GB GDDR6 · **Memory bandwidth**: 448 GB/s
- **FP16 tensor perf.**: 40 TFLOPS · **FP32 perf.**: 7.2 TFLOPS
- **CUDA cores**: 2,176 · **TDP**: 175 W
- **PowerScore ((RTX 3090 = 100))**: 28 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 2060S price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 2060S ($0.07/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.046, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.046 | $1.10 | $34 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.023 | $0.55 | $17 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.029 | $0.70 | $21 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 2060S (8 GB VRAM)

With 8 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.046.

- One-click template: [Ollama on a RTX 2060S](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 2060S](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 2060S](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 2060S availability by region

6 × RTX 2060S across 2 machines, live from inventory:

- Miami, FL
- São Paulo

## RTX 2060S vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 2060S** (this card) | 8 GB | 40 | $0.046 | $1.15‰ |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | 4 | $0.046 | $10.45‰ |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | 8 | $0.046 | $5.75‰ |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | 11 | $0.047 | $4.27‰ |
| [GTX 1660](https://powergpu.ai/gpu/gtx-1660) | 6 GB | 5 | $0.048 | $9.60‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX 2060S: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 2060S per hour?**

$0.046 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.07. Interruptible capacity costs $0.023/hr and a 3-month reservation $0.029/hr. Around $34/month if you keep one running non-stop, billed per second.

**What can a RTX 2060S with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 2060S available to rent right now?**

Yes — 6 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 2060S?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 2060S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-2060s --template pytorch.

**Why is the RTX 2060S cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-2060s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1660-s -->

---
title: "Rent GTX 1660 S — $0.043/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1660 S (6 GB GDDR6) from $0.021/hr interruptible or $0.043/hr on-demand — fixed, ≥30% below market. 15 GPUs in 6 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1660-s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · launched 2019 · prices checked 2026-09-14

# Rent NVIDIA GTX 1660 S — 6 GB, $0.043/hr on-demand

- VRAM 6 GB GDDR6
- FP16 tensor 5 TFLOPS
- PowerScore 4 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 15 6 regions

[On-demand (guaranteed) $0.043 /GPU-hr ≈ $31/mo · market ~~$0.06~~ (−31%)](https://cloud.powergpu.ai/?gpu=gtx-1660-s) [Interruptible $0.021 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1660-s&type=spot) [Reserved 3 mo $0.027 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1660 S specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1660 S · **Architecture**: Turing (2019)
- **VRAM**: 6 GB GDDR6 · **Memory bandwidth**: 336 GB/s
- **FP16 tensor perf.**: 5 TFLOPS · **FP32 perf.**: 5.0 TFLOPS
- **CUDA cores**: 1,408 · **TDP**: 125 W
- **PowerScore ((RTX 3090 = 100))**: 4 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1660 S price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1660 S ($0.06/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.043, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.043 | $1.03 | $31 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.021 | $0.50 | $15 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.027 | $0.65 | $20 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1660 S (6 GB VRAM)

With 6 GB of GDDR6, a single card holds a **~1B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.043.

- One-click template: [Ollama on a GTX 1660 S](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1660 S](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1660 S](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1660 S availability by region

15 × GTX 1660 S across 7 machines, live from inventory:

- Chicago, IL
- Helsinki
- Dallas, TX
- Osaka
- Querétaro
- Sydney

## GTX 1660 S vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1660 S** (this card) | 6 GB | 5 | $0.043 | $8.60‰ |
| [GTX 1060](https://powergpu.ai/gpu/gtx-1060) | 6 GB | 4 | $0.046 | $10.45‰ |
| [GTX 1070 Ti](https://powergpu.ai/gpu/gtx-1070-ti) | 8 GB | 8 | $0.046 | $5.75‰ |
| [RTX 2060S](https://powergpu.ai/gpu/rtx-2060s) | 8 GB | 40 | $0.046 | $1.15‰ |
| [GTX 1080 Ti](https://powergpu.ai/gpu/gtx-1080-ti) | 11 GB | 11 | $0.047 | $4.27‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1660 S: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1660 S per hour?**

$0.043 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.06. Interruptible capacity costs $0.021/hr and a 3-month reservation $0.027/hr. Around $31/month if you keep one running non-stop, billed per second.

**What can a GTX 1660 S with 6 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1660 S available to rent right now?**

Yes — 15 GPUs across 7 machines in 6 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1660 S?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1660 S, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1660-s --template pytorch.

**Why is the GTX 1660 S cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1660-s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/quadro-p4000 -->

---
title: "Rent Quadro P4000 — $0.042/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Quadro P4000 (8 GB GDDR5X) from $0.021/hr interruptible or $0.042/hr on-demand — fixed, ≥30% below market. 4 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/quadro-p4000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2017 · prices checked 2026-09-14

# Rent NVIDIA Quadro P4000 — 8 GB, $0.042/hr on-demand

- VRAM 8 GB GDDR5X
- FP16 tensor 8 TFLOPS
- PowerScore 6 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 4 2 regions

[On-demand (guaranteed) $0.042 /GPU-hr ≈ $31/mo · market ~~$0.06~~ (−31%)](https://cloud.powergpu.ai/?gpu=quadro-p4000) [Interruptible $0.021 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=quadro-p4000&type=spot) [Reserved 3 mo $0.027 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA Quadro P4000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Quadro P4000 · **Architecture**: Pascal (2017)
- **VRAM**: 8 GB GDDR5X · **Memory bandwidth**: 243 GB/s
- **FP16 tensor perf.**: 8 TFLOPS · **FP32 perf.**: 5.3 TFLOPS
- **CUDA cores**: 1,792 · **TDP**: 105 W
- **PowerScore ((RTX 3090 = 100))**: 6 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Quadro P4000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Quadro P4000 ($0.06/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.042, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.042 | $1.01 | $31 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.021 | $0.50 | $15 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.027 | $0.65 | $20 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Quadro P4000 (8 GB VRAM)

With 8 GB of GDDR5X, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.042.

- One-click template: [Ollama on a Quadro P4000](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a Quadro P4000](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a Quadro P4000](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Quadro P4000 availability by region

4 × Quadro P4000 across 2 machines, live from inventory:

- Dubai
- Stockholm

## Quadro P4000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Quadro P4000** (this card) | 8 GB | 8 | $0.042 | $5.25‰ |
| [RTX A2000](https://powergpu.ai/gpu/rtx-a2000) | 6 GB | 32 | $0.024 | $0.75‰ |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB | 76 | $0.071 | $0.93‰ |
| [Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB | 65 | $0.094 | $1.45‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Quadro P4000: frequently asked questions

**How much does it cost to rent an NVIDIA Quadro P4000 per hour?**

$0.042 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.06. Interruptible capacity costs $0.021/hr and a 3-month reservation $0.027/hr. Around $31/month if you keep one running non-stop, billed per second.

**What can a Quadro P4000 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Quadro P4000 available to rent right now?**

Yes — 4 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Quadro P4000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Quadro P4000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu quadro-p4000 --template pytorch.

**Why is the Quadro P4000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/quadro-p4000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-3060 -->

---
title: "Rent RTX 3060 — $0.042/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX 3060 (12 GB GDDR6) from $0.021/hr interruptible or $0.042/hr on-demand — fixed, ≥30% below market. 279 GPUs in 19 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-3060
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Consumer · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA RTX 3060 — 12 GB, $0.042/hr on-demand

- VRAM 12 GB GDDR6
- FP16 tensor 51 TFLOPS
- PowerScore 36 RTX 3090 = 100
- Configs 1–8× PCIe 4.0
- Online now 279 19 regions

[On-demand (guaranteed) $0.042 /GPU-hr ≈ $31/mo · market ~~$0.06~~ (−30%)](https://cloud.powergpu.ai/?gpu=rtx-3060) [Interruptible $0.021 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-3060&type=spot) [Reserved 3 mo $0.027 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Consumer · Ampere architecture

The RTX 3060 (12 GB) is the entry point of the catalogue: enough VRAM for 7B–8B models in 4-bit, Stable Diffusion 1.5 and SDXL at modest batch sizes, and student projects — for pennies an hour with deep supply.

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA RTX 3060 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX 3060 · **Architecture**: Ampere (2021)
- **VRAM**: 12 GB GDDR6 · **Memory bandwidth**: 360 GB/s
- **FP16 tensor perf.**: 51 TFLOPS · **FP32 perf.**: 12.7 TFLOPS
- **CUDA cores**: 3,584 · **TDP**: 170 W
- **PowerScore ((RTX 3090 = 100))**: 36 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX 3060 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX 3060 ($0.06/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.042, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.042 | $1.01 | $31 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.021 | $0.50 | $15 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.027 | $0.65 | $20 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX 3060 (12 GB VRAM)

With 12 GB of GDDR6, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.042.

- One-click template: [Ollama on a RTX 3060](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a RTX 3060](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a RTX 3060](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX 3060 availability by region

279 × RTX 3060 across 30 machines, live from inventory:

- Stockholm
- Warsaw
- Dallas, TX
- Osaka
- Singapore
- São Paulo
- Tokyo
- Amsterdam
- Seoul
- Santiago
- Jakarta
- Sydney
- +7 more

## RTX 3060 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX 3060** (this card) | 12 GB | 51 | $0.042 | $0.82‰ |
| [RTX 3060 Ti](https://powergpu.ai/gpu/rtx-3060-ti) | 8 GB | 65 | $0.047 | $0.72‰ |
| [RTX 3060 laptop](https://powergpu.ai/gpu/rtx-3060-laptop) | 6 GB | 43 | $0.047 | $1.09‰ |
| [RTX 3070](https://powergpu.ai/gpu/rtx-3070) | 8 GB | 81 | $0.056 | $0.69‰ |
| [RTX 3070 Ti](https://powergpu.ai/gpu/rtx-3070-ti) | 8 GB | 87 | $0.066 | $0.76‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

- [RTX 3090 vs RTX 3060](https://powergpu.ai/compare/rtx-3090-vs-rtx-3060) — specs, price per hour, which to rent
- [All GPU comparisons](https://powergpu.ai/compare)

## Renting a RTX 3060: frequently asked questions

**How much does it cost to rent an NVIDIA RTX 3060 per hour?**

$0.042 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.06. Interruptible capacity costs $0.021/hr and a 3-month reservation $0.027/hr. Around $31/month if you keep one running non-stop, billed per second.

**What can a RTX 3060 with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX 3060 available to rent right now?**

Yes — 279 GPUs across 30 machines in 19 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX 3060?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX 3060, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-3060 --template pytorch.

**Why is the RTX 3060 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-3060 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1080 -->

---
title: "Rent GTX 1080 — $0.038/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1080 (8 GB GDDR5X) from $0.019/hr interruptible or $0.038/hr on-demand — fixed, ≥30% below market. 21 GPUs in 5 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1080
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2016 · prices checked 2026-09-14

# Rent NVIDIA GTX 1080 — 8 GB, $0.038/hr on-demand

- VRAM 8 GB GDDR5X
- FP16 tensor 9 TFLOPS
- PowerScore 6 RTX 3090 = 100
- Configs 1–4× PCIe 3.0
- Online now 21 5 regions

[On-demand (guaranteed) $0.038 /GPU-hr ≈ $28/mo · market ~~$0.05~~ (−30%)](https://cloud.powergpu.ai/?gpu=gtx-1080) [Interruptible $0.019 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1080&type=spot) [Reserved 3 mo $0.024 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1080 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1080 · **Architecture**: Pascal (2016)
- **VRAM**: 8 GB GDDR5X · **Memory bandwidth**: 320 GB/s
- **FP16 tensor perf.**: 9 TFLOPS · **FP32 perf.**: 8.9 TFLOPS
- **CUDA cores**: 2,560 · **TDP**: 180 W
- **PowerScore ((RTX 3090 = 100))**: 6 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 1,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1080 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1080 ($0.05/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.038, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.038 | $0.91 | $28 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.019 | $0.46 | $14 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.024 | $0.58 | $18 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1080 (8 GB VRAM)

With 8 GB of GDDR5X, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.038.

- One-click template: [Ollama on a GTX 1080](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1080](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1080](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1080 availability by region

21 × GTX 1080 across 5 machines, live from inventory:

- Montréal
- Amsterdam
- Santiago
- Paris
- Stockholm

## GTX 1080 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1080** (this card) | 8 GB | 9 | $0.038 | $4.22‰ |
| [GTX 1660 S](https://powergpu.ai/gpu/gtx-1660-s) | 6 GB | 5 | $0.043 | $8.60‰ |
| [GTX 1070](https://powergpu.ai/gpu/gtx-1070) | 8 GB | 7 | $0.033 | $5.08‰ |
| [Titan Xp](https://powergpu.ai/gpu/titan-xp) | 12 GB | 12 | $0.033 | $2.75‰ |
| [GTX 1650](https://powergpu.ai/gpu/gtx-1650) | 4 GB | 12 | $0.032 | $2.67‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1080: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1080 per hour?**

$0.038 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.05. Interruptible capacity costs $0.019/hr and a 3-month reservation $0.024/hr. Around $28/month if you keep one running non-stop, billed per second.

**What can a GTX 1080 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1080 available to rent right now?**

Yes — 21 GPUs across 5 machines in 5 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1080?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1080, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1080 --template pytorch.

**Why is the GTX 1080 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1080 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1070 -->

---
title: "Rent GTX 1070 — $0.033/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1070 (8 GB GDDR5X) from $0.016/hr interruptible or $0.033/hr on-demand — fixed, ≥30% below market. 10 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1070
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2016 · prices checked 2026-09-14

# Rent NVIDIA GTX 1070 — 8 GB, $0.033/hr on-demand

- VRAM 8 GB GDDR5X
- FP16 tensor 7 TFLOPS
- PowerScore 5 RTX 3090 = 100
- Configs 1–2× PCIe 3.0
- Online now 10 2 regions

[On-demand (guaranteed) $0.033 /GPU-hr ≈ $24/mo · market ~~$0.05~~ (−31%)](https://cloud.powergpu.ai/?gpu=gtx-1070) [Interruptible $0.016 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1070&type=spot) [Reserved 3 mo $0.021 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1070 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1070 · **Architecture**: Pascal (2016)
- **VRAM**: 8 GB GDDR5X · **Memory bandwidth**: 256 GB/s
- **FP16 tensor perf.**: 7 TFLOPS · **FP32 perf.**: 6.5 TFLOPS
- **CUDA cores**: 1,920 · **TDP**: 150 W
- **PowerScore ((RTX 3090 = 100))**: 5 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 2× · **Max instance storage**: 1,000 GB NVMe
- **Network up to**: 750 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1070 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1070 ($0.05/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.033, **31% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.033 | $0.79 | $24 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.016 | $0.38 | $12 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.021 | $0.50 | $15 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 2× machine costs exactly 2× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1070 (8 GB VRAM)

With 8 GB of GDDR5X, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 2× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.033.

- One-click template: [Ollama on a GTX 1070](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1070](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1070](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1070 availability by region

10 × GTX 1070 across 2 machines, live from inventory:

- Tokyo
- Osaka

## GTX 1070 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1070** (this card) | 8 GB | 7 | $0.033 | $5.08‰ |
| [Titan Xp](https://powergpu.ai/gpu/titan-xp) | 12 GB | 12 | $0.033 | $2.75‰ |
| [GTX 1650](https://powergpu.ai/gpu/gtx-1650) | 4 GB | 12 | $0.032 | $2.67‰ |
| [GTX 1080](https://powergpu.ai/gpu/gtx-1080) | 8 GB | 9 | $0.038 | $4.22‰ |
| [GTX TITAN X](https://powergpu.ai/gpu/gtx-titan-x) | 12 GB | 7 | $0.024 | $3.43‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1070: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1070 per hour?**

$0.033 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.05. Interruptible capacity costs $0.016/hr and a 3-month reservation $0.021/hr. Around $24/month if you keep one running non-stop, billed per second.

**What can a GTX 1070 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 2× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1070 available to rent right now?**

Yes — 10 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 2×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1070?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1070, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1070 --template pytorch.

**Why is the GTX 1070 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1070 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/titan-xp -->

---
title: "Rent Titan Xp — $0.033/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Titan Xp (12 GB GDDR5X) from $0.016/hr interruptible or $0.033/hr on-demand — fixed, ≥30% below market. 55 GPUs in 7 regions, per-second billing."
url: https://powergpu.ai/gpu/titan-xp
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2017 · prices checked 2026-09-14

# Rent NVIDIA Titan Xp — 12 GB, $0.033/hr on-demand

- VRAM 12 GB GDDR5X
- FP16 tensor 12 TFLOPS
- PowerScore 8 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 55 7 regions

[On-demand (guaranteed) $0.033 /GPU-hr ≈ $24/mo · market ~~$0.05~~ (−30%)](https://cloud.powergpu.ai/?gpu=titan-xp) [Interruptible $0.016 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=titan-xp&type=spot) [Reserved 3 mo $0.021 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA Titan Xp specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Titan Xp · **Architecture**: Pascal (2017)
- **VRAM**: 12 GB GDDR5X · **Memory bandwidth**: 548 GB/s
- **FP16 tensor perf.**: 12 TFLOPS · **FP32 perf.**: 12.1 TFLOPS
- **CUDA cores**: 3,840 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 8 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 4,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Titan Xp price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Titan Xp ($0.05/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.033, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.033 | $0.79 | $24 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.016 | $0.38 | $12 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.021 | $0.50 | $15 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Titan Xp (12 GB VRAM)

With 12 GB of GDDR5X, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.033.

- One-click template: [Ollama on a Titan Xp](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a Titan Xp](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a Titan Xp](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Titan Xp availability by region

55 × Titan Xp across 7 machines, live from inventory:

- Dallas, TX
- Stockholm
- Paris
- Dubai
- Los Angeles, CA
- Madrid
- Seoul

## Titan Xp vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Titan Xp** (this card) | 12 GB | 12 | $0.033 | $2.75‰ |
| [GTX 1070](https://powergpu.ai/gpu/gtx-1070) | 8 GB | 7 | $0.033 | $5.08‰ |
| [GTX 1650](https://powergpu.ai/gpu/gtx-1650) | 4 GB | 12 | $0.032 | $2.67‰ |
| [GTX 1080](https://powergpu.ai/gpu/gtx-1080) | 8 GB | 9 | $0.038 | $4.22‰ |
| [GTX TITAN X](https://powergpu.ai/gpu/gtx-titan-x) | 12 GB | 7 | $0.024 | $3.43‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Titan Xp: frequently asked questions

**How much does it cost to rent an NVIDIA Titan Xp per hour?**

$0.033 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.05. Interruptible capacity costs $0.016/hr and a 3-month reservation $0.021/hr. Around $24/month if you keep one running non-stop, billed per second.

**What can a Titan Xp with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Titan Xp available to rent right now?**

Yes — 55 GPUs across 7 machines in 7 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Titan Xp?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Titan Xp, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu titan-xp --template pytorch.

**Why is the Titan Xp cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/titan-xp · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-1650 -->

---
title: "Rent GTX 1650 — $0.032/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX 1650 (4 GB GDDR6) from $0.016/hr interruptible or $0.032/hr on-demand — fixed, ≥30% below market. 2 GPUs in 0 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-1650
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Turing · prices checked 2026-09-14

# Rent NVIDIA GTX 1650 — 4 GB, $0.032/hr on-demand

- VRAM 4 GB GDDR6
- FP16 tensor 12 TFLOPS
- PowerScore 8 RTX 3090 = 100
- Configs 1–1× PCIe 3.0
- Online now 2 0 regions

[On-demand (guaranteed) $0.032 /GPU-hr ≈ $23/mo · market ~~$0.05~~ (−32%)](https://cloud.powergpu.ai/?gpu=gtx-1650) [Interruptible $0.016 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-1650&type=spot) [Reserved 3 mo $0.020 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Turing architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX 1650 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX 1650 · **Architecture**: Turing
- **VRAM**: 4 GB GDDR6 · **Memory bandwidth**: —
- **FP16 tensor perf.**: 12 TFLOPS · **FP32 perf.**: —
- **CUDA cores**: — · **TDP**: —
- **PowerScore ((RTX 3090 = 100))**: 8 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 1× · **Max instance storage**: 0 GB NVMe
- **Network up to**: 0 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX 1650 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX 1650 ($0.05/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.032, **32% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.032 | $0.77 | $23 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.016 | $0.38 | $12 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.020 | $0.48 | $15 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 1× machine costs exactly 1× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX 1650 (4 GB VRAM)

With 4 GB of GDDR6, a single card holds a **~1B-parameter LLM in FP16** or up to **~3B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 1× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.032.

- One-click template: [Ollama on a GTX 1650](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX 1650](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX 1650](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX 1650 availability by region

2 × GTX 1650 across 0 machines, live from inventory:

## GTX 1650 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX 1650** (this card) | 4 GB | 12 | $0.032 | $2.67‰ |
| [GTX 1070](https://powergpu.ai/gpu/gtx-1070) | 8 GB | 7 | $0.033 | $5.08‰ |
| [Titan Xp](https://powergpu.ai/gpu/titan-xp) | 12 GB | 12 | $0.033 | $2.75‰ |
| [GTX 1080](https://powergpu.ai/gpu/gtx-1080) | 8 GB | 9 | $0.038 | $4.22‰ |
| [GTX TITAN X](https://powergpu.ai/gpu/gtx-titan-x) | 12 GB | 7 | $0.024 | $3.43‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX 1650: frequently asked questions

**How much does it cost to rent an NVIDIA GTX 1650 per hour?**

$0.032 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.05. Interruptible capacity costs $0.016/hr and a 3-month reservation $0.020/hr. Around $23/month if you keep one running non-stop, billed per second.

**What can a GTX 1650 with 4 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~3B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 1× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX 1650 available to rent right now?**

Yes — 2 GPUs across 0 machines in 0 regions are listed as we render this page. Configurations go from 1× to 1×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX 1650?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX 1650, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-1650 --template pytorch.

**Why is the GTX 1650 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-1650 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/gtx-titan-x -->

---
title: "Rent GTX TITAN X — $0.024/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the GTX TITAN X (12 GB GDDR5) from $0.012/hr interruptible or $0.024/hr on-demand — fixed, ≥30% below market. 6 GPUs in 2 regions, per-second billing."
url: https://powergpu.ai/gpu/gtx-titan-x
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Maxwell · launched 2015 · prices checked 2026-09-14

# Rent NVIDIA GTX TITAN X — 12 GB, $0.024/hr on-demand

- VRAM 12 GB GDDR5
- FP16 tensor 7 TFLOPS
- PowerScore 5 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 6 2 regions

[On-demand (guaranteed) $0.024 /GPU-hr ≈ $18/mo · market ~~$0.03~~ (−30%)](https://cloud.powergpu.ai/?gpu=gtx-titan-x) [Interruptible $0.012 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=gtx-titan-x&type=spot) [Reserved 3 mo $0.015 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Maxwell architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA GTX TITAN X specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA GTX TITAN X · **Architecture**: Maxwell (2015)
- **VRAM**: 12 GB GDDR5 · **Memory bandwidth**: 337 GB/s
- **FP16 tensor perf.**: 7 TFLOPS · **FP32 perf.**: 6.7 TFLOPS
- **CUDA cores**: 3,072 · **TDP**: 250 W
- **PowerScore ((RTX 3090 = 100))**: 5 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 1,000 GB NVMe
- **Network up to**: 750 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## GTX TITAN X price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the GTX TITAN X ($0.03/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.024, **30% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.024 | $0.58 | $18 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.012 | $0.29 | $9 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.015 | $0.36 | $11 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a GTX TITAN X (12 GB VRAM)

With 12 GB of GDDR5, a single card holds a **~3B-parameter LLM in FP16** or up to **~14B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.024.

- One-click template: [Ollama on a GTX TITAN X](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a GTX TITAN X](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a GTX TITAN X](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## GTX TITAN X availability by region

6 × GTX TITAN X across 2 machines, live from inventory:

- Frankfurt
- Madrid

## GTX TITAN X vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **GTX TITAN X** (this card) | 12 GB | 7 | $0.024 | $3.43‰ |
| [GTX 1650](https://powergpu.ai/gpu/gtx-1650) | 4 GB | 12 | $0.032 | $2.67‰ |
| [GTX 1070](https://powergpu.ai/gpu/gtx-1070) | 8 GB | 7 | $0.033 | $5.08‰ |
| [Titan Xp](https://powergpu.ai/gpu/titan-xp) | 12 GB | 12 | $0.033 | $2.75‰ |
| [GTX 1080](https://powergpu.ai/gpu/gtx-1080) | 8 GB | 9 | $0.038 | $4.22‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a GTX TITAN X: frequently asked questions

**How much does it cost to rent an NVIDIA GTX TITAN X per hour?**

$0.024 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.03. Interruptible capacity costs $0.012/hr and a 3-month reservation $0.015/hr. Around $18/month if you keep one running non-stop, billed per second.

**What can a GTX TITAN X with 12 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~14B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the GTX TITAN X available to rent right now?**

Yes — 6 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a GTX TITAN X?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by GTX TITAN X, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu gtx-titan-x --template pytorch.

**Why is the GTX TITAN X cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/gtx-titan-x · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/tesla-p4 -->

---
title: "Rent Tesla P4 — $0.018/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the Tesla P4 (8 GB GDDR5X) from $0.009/hr interruptible or $0.018/hr on-demand — fixed, ≥30% below market. 18 GPUs in 3 regions, per-second billing."
url: https://powergpu.ai/gpu/tesla-p4
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Previous gen · Pascal · launched 2016 · prices checked 2026-09-14

# Rent NVIDIA Tesla P4 — 8 GB, $0.018/hr on-demand

- VRAM 8 GB GDDR5X
- FP16 tensor 6 TFLOPS
- PowerScore 4 RTX 3090 = 100
- Configs 1–8× PCIe 3.0
- Online now 18 3 regions

[On-demand (guaranteed) $0.018 /GPU-hr ≈ $13/mo · market ~~$0.03~~ (−33%)](https://cloud.powergpu.ai/?gpu=tesla-p4) [Interruptible $0.009 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=tesla-p4&type=spot) [Reserved 3 mo $0.011 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Previous gen · Pascal architecture

A cost-efficient card for right-sized jobs: batch inference, smaller models, CI pipelines and experiments where a flagship would idle. Per-second billing makes it perfect for short bursts.

## NVIDIA Tesla P4 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA Tesla P4 · **Architecture**: Pascal (2016)
- **VRAM**: 8 GB GDDR5X · **Memory bandwidth**: 192 GB/s
- **FP16 tensor perf.**: 6 TFLOPS · **FP32 perf.**: 5.5 TFLOPS
- **CUDA cores**: 2,560 · **TDP**: 75 W
- **PowerScore ((RTX 3090 = 100))**: 4 · **PCIe generation**: Gen 3.0
- **Multi-GPU**: 1× – 8× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 2,500 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## Tesla P4 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the Tesla P4 ($0.03/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.018, **33% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.018 | $0.43 | $13 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.009 | $0.22 | $7 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.011 | $0.26 | $8 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a Tesla P4 (8 GB VRAM)

With 8 GB of GDDR5X, a single card holds a **~3B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.018.

- One-click template: [Ollama on a Tesla P4](https://powergpu.ai/templates/ollama)
- One-click template: [SD WebUI Forge on a Tesla P4](https://powergpu.ai/templates/sd-webui-forge)
- One-click template: [Whisper WebUI & API on a Tesla P4](https://powergpu.ai/templates/whisper-webui-api)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## Tesla P4 availability by region

18 × Tesla P4 across 3 machines, live from inventory:

- Los Angeles, CA
- Miami, FL
- Seattle, WA

## Tesla P4 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **Tesla P4** (this card) | 8 GB | 6 | $0.018 | $3.00‰ |
| [Tesla P40](https://powergpu.ai/gpu/tesla-p40) | 24 GB | 12 | $0.047 | $3.92‰ |
| [Tesla P100](https://powergpu.ai/gpu/tesla-p100) | 16 GB | 19 | $0.066 | $3.47‰ |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) | 16 GB | 65 | $0.103 | $1.58‰ |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | 16 GB | 125 | $0.130 | $1.04‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a Tesla P4: frequently asked questions

**How much does it cost to rent an NVIDIA Tesla P4 per hour?**

$0.018 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.03. Interruptible capacity costs $0.009/hr and a 3-month reservation $0.011/hr. Around $13/month if you keep one running non-stop, billed per second.

**What can a Tesla P4 with 8 GB VRAM run?**

In LLM terms, roughly a 3B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the Tesla P4 available to rent right now?**

Yes — 18 GPUs across 3 machines in 3 regions are listed as we render this page. Configurations go from 1× to 8×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a Tesla P4?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by Tesla P4, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu tesla-p4 --template pytorch.

**Why is the Tesla P4 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/tesla-p4 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/gpu/rtx-a2000 -->

---
title: "Rent RTX A2000 — $0.024/hr Cloud GPU Pricing (2026) | PowerGPU"
description: "Rent the RTX A2000 (6 GB GDDR6) from $0.012/hr interruptible or $0.024/hr on-demand — fixed, ≥30% below market. 20 GPUs in 4 regions, per-second billing."
url: https://powergpu.ai/gpu/rtx-a2000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Workstation · Ampere · launched 2021 · prices checked 2026-09-14

# Rent NVIDIA RTX A2000 — 6 GB, $0.024/hr on-demand

- VRAM 6 GB GDDR6
- FP16 tensor 32 TFLOPS
- PowerScore 23 RTX 3090 = 100
- Configs 1–4× PCIe 4.0
- Online now 20 4 regions

[On-demand (guaranteed) $0.024 /GPU-hr ≈ $18/mo · market ~~$0.04~~ (−32%)](https://cloud.powergpu.ai/?gpu=rtx-a2000) [Interruptible $0.012 /GPU-hr flat −50% · pausable, disk kept](https://cloud.powergpu.ai/?gpu=rtx-a2000&type=spot) [Reserved 3 mo $0.015 /GPU-hr −35% · capacity held for you](https://powergpu.ai/products/reserved)

Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — [the whole fee schedule](https://powergpu.ai/pricing). Next weekly market re-check: 2026-09-21.

Workstation · Ampere architecture

Workstation silicon with ECC memory and studio-certified drivers. Renderers (Blender, Octane, Redshift), CAD and simulation get the large VRAM they want without paying datacenter-flagship rates — and per-second billing suits render queues perfectly.

## NVIDIA RTX A2000 specs: VRAM, TFLOPS, bandwidth

- **GPU model**: NVIDIA RTX A2000 · **Architecture**: Ampere (2021)
- **VRAM**: 6 GB GDDR6 · **Memory bandwidth**: 288 GB/s
- **FP16 tensor perf.**: 32 TFLOPS · **FP32 perf.**: 8.0 TFLOPS
- **CUDA cores**: 3,328 · **TDP**: 70 W
- **PowerScore ((RTX 3090 = 100))**: 23 · **PCIe generation**: Gen 4.0
- **Multi-GPU**: 1× – 4× · **Max instance storage**: 8,000 GB NVMe
- **Network up to**: 10,000 Mbps · **CUDA**: 12.4 – 13.0

Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.

## RTX A2000 price per hour: on-demand, interruptible, reserved

One public rule sets every price on this page: the marketplace median for the RTX A2000 ($0.04/hr, snapshot 2026-09-14) × 0.70, rounded down — so on-demand is $0.024, **32% below market**. Interruptible halves it; a 3-month reservation takes another 35% off.

| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
| --- | --- | --- | --- | --- |
| **On-demand** | $0.024 | $0.58 | $18 | Guaranteed capacity, price locked at deploy, stop anytime |
| **Interruptible** | $0.012 | $0.29 | $9 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| **Reserved (3 months)** | $0.015 | $0.36 | $11 | −35% on on-demand, rate locked for the term, capacity held |

Per GPU: an 4× machine costs exactly 4× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What you can run on a RTX A2000 (6 GB VRAM)

With 6 GB of GDDR6, a single card holds a **~1B-parameter LLM in FP16** or up to **~8B parameters quantized to 4-bit**, with room for KV-cache at practical context lengths. Scale to 4× GPUs on one machine for bigger models or bigger batches — the per-GPU price stays $0.024.

- One-click template: [Linux Desktop on a RTX A2000](https://powergpu.ai/templates/linux-desktop)
- One-click template: [ComfyUI on a RTX A2000](https://powergpu.ai/templates/comfyui)
- One-click template: [Ubuntu Desktop VM on a RTX A2000](https://powergpu.ai/templates/ubuntu-desktop-vm)
- Sizing help: [LLM VRAM requirements guide](https://powergpu.ai/guides/llm-vram-requirements)

## RTX A2000 availability by region

20 × RTX A2000 across 5 machines, live from inventory:

- Warsaw
- Singapore
- Seattle, WA
- Stockholm

## RTX A2000 vs alternatives: price per TFLOP

| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
| --- | --- | --- | --- | --- |
| **RTX A2000** (this card) | 6 GB | 32 | $0.024 | $0.75‰ |
| [Quadro P4000](https://powergpu.ai/gpu/quadro-p4000) | 8 GB | 8 | $0.042 | $5.25‰ |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | 16 GB | 76 | $0.071 | $0.93‰ |
| [Q RTX 6000](https://powergpu.ai/gpu/q-rtx-6000) | 24 GB | 65 | $0.094 | $1.45‰ |
| [Titan RTX](https://powergpu.ai/gpu/titan-rtx) | 24 GB | 65 | $0.103 | $1.58‰ |

‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.

## Renting a RTX A2000: frequently asked questions

**How much does it cost to rent an NVIDIA RTX A2000 per hour?**

$0.024 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $0.04. Interruptible capacity costs $0.012/hr and a 3-month reservation $0.015/hr. Around $18/month if you keep one running non-stop, billed per second.

**What can a RTX A2000 with 6 GB VRAM run?**

In LLM terms, roughly a 1B-parameter model in FP16 or up to ~8B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 4× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.

**Is the RTX A2000 available to rent right now?**

Yes — 20 GPUs across 5 machines in 4 regions are listed as we render this page. Configurations go from 1× to 4×. Deploy from the console and it is running in about 30 seconds.

**How do I deploy a RTX A2000?**

Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by RTX A2000, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu rtx-a2000 --template pytorch.

**Why is the RTX A2000 cheaper here than on GPU marketplaces?**

We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-21) and published — no auctions, no per-host roulette, no bidding.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/gpu/rtx-a2000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare -->

---
title: "Compare Cloud GPUs Side by Side — Specs & Price per Hour | PowerGPU"
description: "H100 vs A100, RTX 5090 vs RTX 4090, H200 vs B200 and 25 more GPU comparisons: VRAM, bandwidth, TFLOPS, live hourly prices and a clear rental verdict for each pair."
url: https://powergpu.ai/compare
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · 28 head-to-heads · live prices

# Compare cloud GPUs: specs, price per hour, which to rent

Every comparison puts two rentable cards side by side — VRAM, bandwidth, CUDA cores, FP16 throughput — with today's fixed prices from our sheet and a verdict written for real workloads, not benchmarks in a vacuum.

## Datacenter & HBM (14)

- [H100 SXM vs A100 SXM4 80 GB · $1.428/hr · 80 GB · $0.560/hr](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4)
- [H100 SXM vs H200 80 GB · $1.428/hr · 141 GB · $2.791/hr](https://powergpu.ai/compare/h100-sxm-vs-h200)
- [H200 vs B200 141 GB · $2.791/hr · 192 GB · $5.425/hr](https://powergpu.ai/compare/h200-vs-b200)
- [H100 SXM vs B200 80 GB · $1.428/hr · 192 GB · $5.425/hr](https://powergpu.ai/compare/h100-sxm-vs-b200)
- [H100 SXM vs H100 PCIE 80 GB · $1.428/hr · 80 GB · $1.867/hr](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)
- [H100 PCIE vs A100 PCIE 80 GB · $1.867/hr · 80 GB · $0.374/hr](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie)
- [A100 SXM4 vs A100 PCIE 80 GB · $0.560/hr · 80 GB · $0.374/hr](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie)
- [L40S vs A100 PCIE 48 GB · $0.514/hr · 80 GB · $0.374/hr](https://powergpu.ai/compare/l40s-vs-a100-pcie)
- [L40S vs H100 PCIE 48 GB · $0.514/hr · 80 GB · $1.867/hr](https://powergpu.ai/compare/l40s-vs-h100-pcie)
- [L4 vs Tesla T4 24 GB · $0.225/hr · 16 GB · $0.103/hr](https://powergpu.ai/compare/l4-vs-tesla-t4)
- [L4 vs A10 24 GB · $0.225/hr · 24 GB · $0.168/hr](https://powergpu.ai/compare/l4-vs-a10)
- [H100 NVL vs H100 SXM 80 GB · $1.811/hr · 80 GB · $1.428/hr](https://powergpu.ai/compare/h100-nvl-vs-h100-sxm)
- [H200 vs H200 NVL 141 GB · $2.791/hr · 141 GB · $2.650/hr](https://powergpu.ai/compare/h200-vs-h200-nvl)
- [B200 vs B300 192 GB · $5.425/hr · 288 GB · $6.737/hr](https://powergpu.ai/compare/b200-vs-b300)

## Consumer RTX (8)

- [RTX 5090 vs RTX 4090 32 GB · $0.439/hr · 24 GB · $0.327/hr](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 4090 vs RTX 3090 24 GB · $0.327/hr · 24 GB · $0.108/hr](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [RTX 4090 vs A100 PCIE 24 GB · $0.327/hr · 80 GB · $0.374/hr](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [RTX 4090 vs L40S 24 GB · $0.327/hr · 48 GB · $0.514/hr](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [RTX 5090 vs L40S 32 GB · $0.439/hr · 48 GB · $0.514/hr](https://powergpu.ai/compare/rtx-5090-vs-l40s)
- [RTX 5090 vs A100 SXM4 32 GB · $0.439/hr · 80 GB · $0.560/hr](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4)
- [RTX 3090 vs RTX 3060 24 GB · $0.108/hr · 12 GB · $0.042/hr](https://powergpu.ai/compare/rtx-3090-vs-rtx-3060)
- [RTX 5080 vs RTX 4090 16 GB · $0.186/hr · 24 GB · $0.327/hr](https://powergpu.ai/compare/rtx-5080-vs-rtx-4090)

## Workstation & inference (6)

- [RTX PRO 6000 WS vs RTX 6000Ada 96 GB · $1.040/hr · 48 GB · $0.467/hr](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada)
- [RTX PRO 6000 WS vs H100 PCIE 96 GB · $1.040/hr · 80 GB · $1.867/hr](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie)
- [RTX PRO 6000 WS vs RTX 5090 96 GB · $1.040/hr · 32 GB · $0.439/hr](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090)
- [RTX 6000Ada vs RTX A6000 48 GB · $0.467/hr · 48 GB · $0.281/hr](https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000)
- [RTX A6000 vs RTX 4090 48 GB · $0.281/hr · 24 GB · $0.327/hr](https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090)
- [Tesla V100 vs RTX 3090 16 GB · $0.130/hr · 24 GB · $0.108/hr](https://powergpu.ai/compare/tesla-v100-vs-rtx-3090)

## Comparing GPUs for rental: FAQ

Sizing questions live in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements); the [pricing guide](https://powergpu.ai/guides/cloud-gpu-pricing-explained) explains the billing modes.

**How do you compare two GPUs for rental?**

Four numbers decide most jobs: VRAM (does the model fit?), memory bandwidth (inference speed), FP16 tensor throughput (training speed) and price per hour. We show all four side by side, then price per TFLOP-hour and per GB of VRAM so the value gap is explicit.

**Are the prices on these pages live?**

Yes — every price is our current fixed rate, set at least 30% below the public marketplace median and re-checked weekly (snapshot 2026-09-14). Both cards in every comparison are rentable right now.

**What if the GPU I want to compare is missing?**

Every one of the 80 GPU pages carries a "vs alternatives" table with the four closest cards by family and price, and the pricing page lists all models on one sheet.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4 -->

---
title: "H100 SXM vs A100 SXM4: Specs & Price per Hour (2026) | PowerGPU"
description: "H100 SXM vs A100 SXM4: 80 vs 80 GB VRAM, 990 vs 312 FP16 TFLOPS, $1.428 vs $0.560/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H100 SXM vs A100 SXM4: specs, price per hour, which to rent

NVIDIA H100 SXM (80 GB, **$1.428** /hr) against NVIDIA A100 SXM4 (80 GB, **$0.560** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H100 SXM Hopper · 80 GB HBM3 · 990 FP16 TFLOPS · $1.428/hr on-demand · $0.714/hr interruptible · 54 online](https://powergpu.ai/gpu/h100-sxm)
- [NVIDIA A100 SXM4 Ampere · 80 GB HBM2e · 312 FP16 TFLOPS · $0.560/hr on-demand · $0.280/hr interruptible · 95 online](https://powergpu.ai/gpu/a100-sxm4)

## H100 SXM vs A100 SXM4 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H100 SXM relative to A100 SXM4.

| Spec | H100 SXM | A100 SXM4 | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2022) | Ampere (2020) | — |
| VRAM | 80 GB HBM3 | 80 GB HBM2e | same |
| Memory bandwidth | 3,350 GB/s | 2,039 GB/s | +64% |
| FP16 tensor (dense) | 990 TFLOPS | 312 TFLOPS | +217% |
| FP32 | 67.0 TFLOPS | 19.5 TFLOPS | +244% |
| CUDA cores | 16,896 | 6,912 | +144% |
| TDP | 700 W | 400 W | +75% |
| PCIe · NVLink | Gen 5.0 · NVLink | Gen 4.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 697 | 220 | +217% |
| Max GPUs per machine | 8× | 8× | — |

## H100 SXM vs A100 SXM4 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H100 SXM | A100 SXM4 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.428 | $0.560 | A100 SXM4 (−61%) |
| Interruptible, per GPU-hour | $0.714 | $0.280 | A100 SXM4 |
| Reserved (3 mo), per GPU-hour | $0.928 | $0.364 | A100 SXM4 |
| On-demand, per month | $1,042 | $409 | A100 SXM4 |
| Market median (reference) | $2.04 | $0.80 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.44 | $1.79 | H100 SXM (better value) |
| $ per GB of VRAM per hour | $0.0178 | $0.0070 | A100 SXM4 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 80 GB vs 80 GB

| Workload | H100 SXM | A100 SXM4 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~32B |
| Largest LLM at 4-bit, one card | ~123B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The H100 SXM trains transformers roughly 2–3× faster than the A100 SXM4 thanks to FP8 and 3.35 TB/s of HBM3; the A100 costs far less per hour and still wins dollars per step for ≤13B models, LoRA farms and FP64 HPC. Rent the H100 for deadlines and FP8 serving, the A100 for volume experiments.

- **Cheaper per hour:** A100 SXM4 ($0.560 vs $1.428, −61%).
- **More VRAM:** H100 SXM (80 GB vs 80 GB).
- **More FP16 throughput:** H100 SXM (about 3.2×).
- **Best value per TFLOP-hour:** H100 SXM.
- **Best value per GB of VRAM:** A100 SXM4.
- **Multi-GPU:** H100 SXM with NVLink · A100 SXM4 with NVLink.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs H200 $1.428 vs $2.791 per hour](https://powergpu.ai/compare/h100-sxm-vs-h200)
- [H100 SXM vs B200 $1.428 vs $5.425 per hour](https://powergpu.ai/compare/h100-sxm-vs-b200)
- [H100 SXM vs H100 PCIE $1.428 vs $1.867 per hour](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)
- [A100 SXM4 vs A100 PCIE $0.560 vs $0.374 per hour](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie)
- [RTX 5090 vs A100 SXM4 $0.439 vs $0.560 per hour](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4)
- [H100 NVL vs H100 SXM $1.811 vs $1.428 per hour](https://powergpu.ai/compare/h100-nvl-vs-h100-sxm)

## H100 SXM vs A100 SXM4: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H100 SXM faster than the A100 SXM4?**

On dense FP16 tensor throughput the H100 SXM leads by about 3.2× (990 vs 312 TFLOPS). Memory bandwidth matters as much for inference: H100 SXM 3,350 GB/s vs A100 SXM4 2,039 GB/s.

**Which is cheaper to rent, the H100 SXM or the A100 SXM4?**

The A100 SXM4: $0.560/hr on-demand versus $1.428/hr — 61% less. Interruptible rates are $0.714 (H100 SXM) and $0.280 (A100 SXM4). Per TFLOP-hour the better value is the H100 SXM.

**Which has more VRAM and what does that change?**

The H100 SXM has 80 GB versus 80 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 54 × H100 SXM and 95 × A100 SXM4 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h100-sxm-vs-h200 -->

---
title: "H100 SXM vs H200: Specs & Price per Hour (2026) | PowerGPU"
description: "H100 SXM vs H200: 80 vs 141 GB VRAM, 990 vs 990 FP16 TFLOPS, $1.428 vs $2.791/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h100-sxm-vs-h200
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H100 SXM vs H200: specs, price per hour, which to rent

NVIDIA H100 SXM (80 GB, **$1.428** /hr) against NVIDIA H200 (141 GB, **$2.791** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H100 SXM Hopper · 80 GB HBM3 · 990 FP16 TFLOPS · $1.428/hr on-demand · $0.714/hr interruptible · 54 online](https://powergpu.ai/gpu/h100-sxm)
- [NVIDIA H200 Hopper · 141 GB HBM3e · 990 FP16 TFLOPS · $2.791/hr on-demand · $1.395/hr interruptible · 50 online](https://powergpu.ai/gpu/h200)

## H100 SXM vs H200 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H100 SXM relative to H200.

| Spec | H100 SXM | H200 | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2022) | Hopper (2023) | — |
| VRAM | 80 GB HBM3 | 141 GB HBM3e | −43% |
| Memory bandwidth | 3,350 GB/s | 4,800 GB/s | −30% |
| FP16 tensor (dense) | 990 TFLOPS | 990 TFLOPS | same |
| FP32 | 67.0 TFLOPS | 67.0 TFLOPS | same |
| CUDA cores | 16,896 | 16,896 | same |
| TDP | 700 W | 700 W | same |
| PCIe · NVLink | Gen 5.0 · NVLink | Gen 5.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 697 | 697 | same |
| Max GPUs per machine | 8× | 8× | — |

## H100 SXM vs H200 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H100 SXM | H200 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.428 | $2.791 | H100 SXM (−49%) |
| Interruptible, per GPU-hour | $0.714 | $1.395 | H100 SXM |
| Reserved (3 mo), per GPU-hour | $0.928 | $1.814 | H100 SXM |
| On-demand, per month | $1,042 | $2,037 | H100 SXM |
| Market median (reference) | $2.04 | $3.99 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.44 | $2.82 | H100 SXM (better value) |
| $ per GB of VRAM per hour | $0.0178 | $0.0198 | H100 SXM (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 80 GB vs 141 GB

| Workload | H100 SXM | H200 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~49B |
| Largest LLM at 4-bit, one card | ~123B | ~141B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Same Hopper compute, 141 GB versus 80 GB and 4.8 versus 3.35 TB/s: the H200 is the memory-bound choice — long context, MoE serving, big micro-batches — while the H100 SXM wins compute per dollar for everything that fits in 80 GB.

- **Cheaper per hour:** H100 SXM ($1.428 vs $2.791, −49%).
- **More VRAM:** H200 (141 GB vs 80 GB).
- **More FP16 throughput:** H100 SXM (about 1.0×).
- **Best value per TFLOP-hour:** H100 SXM.
- **Best value per GB of VRAM:** H100 SXM.
- **Multi-GPU:** H100 SXM with NVLink · H200 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs A100 SXM4 $1.428 vs $0.560 per hour](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4)
- [H200 vs B200 $2.791 vs $5.425 per hour](https://powergpu.ai/compare/h200-vs-b200)
- [H100 SXM vs B200 $1.428 vs $5.425 per hour](https://powergpu.ai/compare/h100-sxm-vs-b200)
- [H100 SXM vs H100 PCIE $1.428 vs $1.867 per hour](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)
- [H100 NVL vs H100 SXM $1.811 vs $1.428 per hour](https://powergpu.ai/compare/h100-nvl-vs-h100-sxm)
- [H200 vs H200 NVL $2.791 vs $2.650 per hour](https://powergpu.ai/compare/h200-vs-h200-nvl)

## H100 SXM vs H200: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H100 SXM faster than the H200?**

On dense FP16 tensor throughput the H100 SXM leads by about 1.0× (990 vs 990 TFLOPS). Memory bandwidth matters as much for inference: H100 SXM 3,350 GB/s vs H200 4,800 GB/s.

**Which is cheaper to rent, the H100 SXM or the H200?**

The H100 SXM: $1.428/hr on-demand versus $2.791/hr — 49% less. Interruptible rates are $0.714 (H100 SXM) and $1.395 (H200). Per TFLOP-hour the better value is the H100 SXM.

**Which has more VRAM and what does that change?**

The H200 has 141 GB versus 80 GB. In LLM terms that is roughly a 49B FP16 model (or ~141B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 54 × H100 SXM and 50 × H200 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h100-sxm-vs-h200 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h200-vs-b200 -->

---
title: "H200 vs B200: Specs & Price per Hour (2026) | PowerGPU"
description: "H200 vs B200: 141 vs 192 GB VRAM, 990 vs 2,250 FP16 TFLOPS, $2.791 vs $5.425/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h200-vs-b200
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H200 vs B200: specs, price per hour, which to rent

NVIDIA H200 (141 GB, **$2.791** /hr) against NVIDIA B200 (192 GB, **$5.425** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H200 Hopper · 141 GB HBM3e · 990 FP16 TFLOPS · $2.791/hr on-demand · $1.395/hr interruptible · 50 online](https://powergpu.ai/gpu/h200)
- [NVIDIA B200 Blackwell · 192 GB HBM3e · 2,250 FP16 TFLOPS · $5.425/hr on-demand · $2.712/hr interruptible · 32 online](https://powergpu.ai/gpu/b200)

## H200 vs B200 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H200 relative to B200.

| Spec | H200 | B200 | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2023) | Blackwell (2024) | — |
| VRAM | 141 GB HBM3e | 192 GB HBM3e | −27% |
| Memory bandwidth | 4,800 GB/s | 8,000 GB/s | −40% |
| FP16 tensor (dense) | 990 TFLOPS | 2,250 TFLOPS | −56% |
| FP32 | 67.0 TFLOPS | — | — |
| CUDA cores | 16,896 | — | — |
| TDP | 700 W | 1000 W | −30% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 5.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 697 | 1585 | −56% |
| Max GPUs per machine | 8× | 8× | — |

## H200 vs B200 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H200 | B200 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $2.791 | $5.425 | H200 (−49%) |
| Interruptible, per GPU-hour | $1.395 | $2.712 | H200 |
| Reserved (3 mo), per GPU-hour | $1.814 | $3.526 | H200 |
| On-demand, per month | $2,037 | $3,960 | H200 |
| Market median (reference) | $3.99 | $7.75 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.82 | $2.41 | B200 (better value) |
| $ per GB of VRAM per hour | $0.0198 | $0.0283 | H200 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 141 GB vs 192 GB

| Workload | H200 | B200 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~49B | ~72B |
| Largest LLM at 4-bit, one card | ~141B | ~235B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The B200 roughly doubles FP16 throughput and bandwidth over the H200 and adds FP4; the H200 stays cheaper per hour with 141 GB. For throughput-critical training and FP4 inference the B200 usually costs less per token; for memory capacity at a lower rate, the H200.

- **Cheaper per hour:** H200 ($2.791 vs $5.425, −49%).
- **More VRAM:** B200 (192 GB vs 141 GB).
- **More FP16 throughput:** B200 (about 2.3×).
- **Best value per TFLOP-hour:** B200.
- **Best value per GB of VRAM:** H200.
- **Multi-GPU:** H200 over PCIe · B200 with NVLink.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs H200 $1.428 vs $2.791 per hour](https://powergpu.ai/compare/h100-sxm-vs-h200)
- [H100 SXM vs B200 $1.428 vs $5.425 per hour](https://powergpu.ai/compare/h100-sxm-vs-b200)
- [H200 vs H200 NVL $2.791 vs $2.650 per hour](https://powergpu.ai/compare/h200-vs-h200-nvl)
- [B200 vs B300 $5.425 vs $6.737 per hour](https://powergpu.ai/compare/b200-vs-b300)

## H200 vs B200: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the B200 faster than the H200?**

On dense FP16 tensor throughput the B200 leads by about 2.3× (2,250 vs 990 TFLOPS). Memory bandwidth matters as much for inference: H200 4,800 GB/s vs B200 8,000 GB/s.

**Which is cheaper to rent, the H200 or the B200?**

The H200: $2.791/hr on-demand versus $5.425/hr — 49% less. Interruptible rates are $1.395 (H200) and $2.712 (B200). Per TFLOP-hour the better value is the B200.

**Which has more VRAM and what does that change?**

The B200 has 192 GB versus 141 GB. In LLM terms that is roughly a 72B FP16 model (or ~235B in 4-bit) on one card against 49B FP16 (~141B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 50 × H200 and 32 × B200 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h200-vs-b200 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h100-sxm-vs-b200 -->

---
title: "H100 SXM vs B200: Specs & Price per Hour (2026) | PowerGPU"
description: "H100 SXM vs B200: 80 vs 192 GB VRAM, 990 vs 2,250 FP16 TFLOPS, $1.428 vs $5.425/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h100-sxm-vs-b200
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H100 SXM vs B200: specs, price per hour, which to rent

NVIDIA H100 SXM (80 GB, **$1.428** /hr) against NVIDIA B200 (192 GB, **$5.425** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H100 SXM Hopper · 80 GB HBM3 · 990 FP16 TFLOPS · $1.428/hr on-demand · $0.714/hr interruptible · 54 online](https://powergpu.ai/gpu/h100-sxm)
- [NVIDIA B200 Blackwell · 192 GB HBM3e · 2,250 FP16 TFLOPS · $5.425/hr on-demand · $2.712/hr interruptible · 32 online](https://powergpu.ai/gpu/b200)

## H100 SXM vs B200 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H100 SXM relative to B200.

| Spec | H100 SXM | B200 | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2022) | Blackwell (2024) | — |
| VRAM | 80 GB HBM3 | 192 GB HBM3e | −58% |
| Memory bandwidth | 3,350 GB/s | 8,000 GB/s | −58% |
| FP16 tensor (dense) | 990 TFLOPS | 2,250 TFLOPS | −56% |
| FP32 | 67.0 TFLOPS | — | — |
| CUDA cores | 16,896 | — | — |
| TDP | 700 W | 1000 W | −30% |
| PCIe · NVLink | Gen 5.0 · NVLink | Gen 5.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 697 | 1585 | −56% |
| Max GPUs per machine | 8× | 8× | — |

## H100 SXM vs B200 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H100 SXM | B200 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.428 | $5.425 | H100 SXM (−74%) |
| Interruptible, per GPU-hour | $0.714 | $2.712 | H100 SXM |
| Reserved (3 mo), per GPU-hour | $0.928 | $3.526 | H100 SXM |
| On-demand, per month | $1,042 | $3,960 | H100 SXM |
| Market median (reference) | $2.04 | $7.75 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.44 | $2.41 | H100 SXM (better value) |
| $ per GB of VRAM per hour | $0.0178 | $0.0283 | H100 SXM (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 80 GB vs 192 GB

| Workload | H100 SXM | B200 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~72B |
| Largest LLM at 4-bit, one card | ~123B | ~235B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Blackwell delivers about 2.3× the dense FP16 throughput of the H100 SXM with 2.4× the memory; the H100 is cheaper per hour and far deeper in supply. B200 for wall-clock time and FP4 serving, H100 SXM for price per FLOP and interruptible availability.

- **Cheaper per hour:** H100 SXM ($1.428 vs $5.425, −74%).
- **More VRAM:** B200 (192 GB vs 80 GB).
- **More FP16 throughput:** B200 (about 2.3×).
- **Best value per TFLOP-hour:** H100 SXM.
- **Best value per GB of VRAM:** H100 SXM.
- **Multi-GPU:** H100 SXM with NVLink · B200 with NVLink.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs A100 SXM4 $1.428 vs $0.560 per hour](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4)
- [H100 SXM vs H200 $1.428 vs $2.791 per hour](https://powergpu.ai/compare/h100-sxm-vs-h200)
- [H200 vs B200 $2.791 vs $5.425 per hour](https://powergpu.ai/compare/h200-vs-b200)
- [H100 SXM vs H100 PCIE $1.428 vs $1.867 per hour](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)
- [H100 NVL vs H100 SXM $1.811 vs $1.428 per hour](https://powergpu.ai/compare/h100-nvl-vs-h100-sxm)
- [B200 vs B300 $5.425 vs $6.737 per hour](https://powergpu.ai/compare/b200-vs-b300)

## H100 SXM vs B200: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the B200 faster than the H100 SXM?**

On dense FP16 tensor throughput the B200 leads by about 2.3× (2,250 vs 990 TFLOPS). Memory bandwidth matters as much for inference: H100 SXM 3,350 GB/s vs B200 8,000 GB/s.

**Which is cheaper to rent, the H100 SXM or the B200?**

The H100 SXM: $1.428/hr on-demand versus $5.425/hr — 74% less. Interruptible rates are $0.714 (H100 SXM) and $2.712 (B200). Per TFLOP-hour the better value is the H100 SXM.

**Which has more VRAM and what does that change?**

The B200 has 192 GB versus 80 GB. In LLM terms that is roughly a 72B FP16 model (or ~235B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 54 × H100 SXM and 32 × B200 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h100-sxm-vs-b200 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h100-sxm-vs-h100-pcie -->

---
title: "H100 SXM vs H100 PCIE: Specs & Price per Hour (2026) | PowerGPU"
description: "H100 SXM vs H100 PCIE: 80 vs 80 GB VRAM, 990 vs 756 FP16 TFLOPS, $1.428 vs $1.867/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h100-sxm-vs-h100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H100 SXM vs H100 PCIE: specs, price per hour, which to rent

NVIDIA H100 SXM (80 GB, **$1.428** /hr) against NVIDIA H100 PCIE (80 GB, **$1.867** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H100 SXM Hopper · 80 GB HBM3 · 990 FP16 TFLOPS · $1.428/hr on-demand · $0.714/hr interruptible · 54 online](https://powergpu.ai/gpu/h100-sxm)
- [NVIDIA H100 PCIE Hopper · 80 GB HBM2e · 756 FP16 TFLOPS · $1.867/hr on-demand · $0.933/hr interruptible · 73 online](https://powergpu.ai/gpu/h100-pcie)

## H100 SXM vs H100 PCIE specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H100 SXM relative to H100 PCIE.

| Spec | H100 SXM | H100 PCIE | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2022) | Hopper (2022) | — |
| VRAM | 80 GB HBM3 | 80 GB HBM2e | same |
| Memory bandwidth | 3,350 GB/s | 2,000 GB/s | +68% |
| FP16 tensor (dense) | 990 TFLOPS | 756 TFLOPS | +31% |
| FP32 | 67.0 TFLOPS | 51.0 TFLOPS | +31% |
| CUDA cores | 16,896 | 14,592 | +16% |
| TDP | 700 W | 350 W | +100% |
| PCIe · NVLink | Gen 5.0 · NVLink | Gen 5.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 697 | 532 | +31% |
| Max GPUs per machine | 8× | 8× | — |

## H100 SXM vs H100 PCIE price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H100 SXM | H100 PCIE | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.428 | $1.867 | H100 SXM (−24%) |
| Interruptible, per GPU-hour | $0.714 | $0.933 | H100 SXM |
| Reserved (3 mo), per GPU-hour | $0.928 | $1.213 | H100 SXM |
| On-demand, per month | $1,042 | $1,363 | H100 SXM |
| Market median (reference) | $2.04 | $2.67 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.44 | $2.47 | H100 SXM (better value) |
| $ per GB of VRAM per hour | $0.0178 | $0.0233 | H100 SXM (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 80 GB vs 80 GB

| Workload | H100 SXM | H100 PCIE |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~32B |
| Largest LLM at 4-bit, one card | ~123B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The same Hopper silicon in two packages: SXM brings 3.35 TB/s HBM3, 700 W and full NVLink; PCIe has 2 TB/s HBM2e and 350 W. Multi-GPU training wants SXM; single-GPU fine-tuning and inference get most of the value from the PCIe card at a lower rate.

- **Cheaper per hour:** H100 SXM ($1.428 vs $1.867, −24%).
- **More VRAM:** H100 SXM (80 GB vs 80 GB).
- **More FP16 throughput:** H100 SXM (about 1.3×).
- **Best value per TFLOP-hour:** H100 SXM.
- **Best value per GB of VRAM:** H100 SXM.
- **Multi-GPU:** H100 SXM with NVLink · H100 PCIE over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs A100 SXM4 $1.428 vs $0.560 per hour](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4)
- [H100 SXM vs H200 $1.428 vs $2.791 per hour](https://powergpu.ai/compare/h100-sxm-vs-h200)
- [H100 SXM vs B200 $1.428 vs $5.425 per hour](https://powergpu.ai/compare/h100-sxm-vs-b200)
- [H100 PCIE vs A100 PCIE $1.867 vs $0.374 per hour](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie)
- [L40S vs H100 PCIE $0.514 vs $1.867 per hour](https://powergpu.ai/compare/l40s-vs-h100-pcie)
- [RTX PRO 6000 WS vs H100 PCIE $1.040 vs $1.867 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie)

## H100 SXM vs H100 PCIE: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H100 SXM faster than the H100 PCIE?**

On dense FP16 tensor throughput the H100 SXM leads by about 1.3× (990 vs 756 TFLOPS). Memory bandwidth matters as much for inference: H100 SXM 3,350 GB/s vs H100 PCIE 2,000 GB/s.

**Which is cheaper to rent, the H100 SXM or the H100 PCIE?**

The H100 SXM: $1.428/hr on-demand versus $1.867/hr — 24% less. Interruptible rates are $0.714 (H100 SXM) and $0.933 (H100 PCIE). Per TFLOP-hour the better value is the H100 SXM.

**Which has more VRAM and what does that change?**

The H100 SXM has 80 GB versus 80 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 54 × H100 SXM and 73 × H100 PCIE are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h100-sxm-vs-h100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h100-pcie-vs-a100-pcie -->

---
title: "H100 PCIE vs A100 PCIE: Specs & Price per Hour (2026) | PowerGPU"
description: "H100 PCIE vs A100 PCIE: 80 vs 80 GB VRAM, 756 vs 312 FP16 TFLOPS, $1.867 vs $0.374/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h100-pcie-vs-a100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H100 PCIE vs A100 PCIE: specs, price per hour, which to rent

NVIDIA H100 PCIE (80 GB, **$1.867** /hr) against NVIDIA A100 PCIE (80 GB, **$0.374** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H100 PCIE Hopper · 80 GB HBM2e · 756 FP16 TFLOPS · $1.867/hr on-demand · $0.933/hr interruptible · 73 online](https://powergpu.ai/gpu/h100-pcie)
- [NVIDIA A100 PCIE Ampere · 80 GB HBM2e · 312 FP16 TFLOPS · $0.374/hr on-demand · $0.187/hr interruptible · 46 online](https://powergpu.ai/gpu/a100-pcie)

## H100 PCIE vs A100 PCIE specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H100 PCIE relative to A100 PCIE.

| Spec | H100 PCIE | A100 PCIE | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2022) | Ampere (2021) | — |
| VRAM | 80 GB HBM2e | 80 GB HBM2e | same |
| Memory bandwidth | 2,000 GB/s | 1,935 GB/s | +3% |
| FP16 tensor (dense) | 756 TFLOPS | 312 TFLOPS | +142% |
| FP32 | 51.0 TFLOPS | 19.5 TFLOPS | +162% |
| CUDA cores | 14,592 | 6,912 | +111% |
| TDP | 350 W | 300 W | +17% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 532 | 220 | +142% |
| Max GPUs per machine | 8× | 8× | — |

## H100 PCIE vs A100 PCIE price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H100 PCIE | A100 PCIE | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.867 | $0.374 | A100 PCIE (−80%) |
| Interruptible, per GPU-hour | $0.933 | $0.187 | A100 PCIE |
| Reserved (3 mo), per GPU-hour | $1.213 | $0.243 | A100 PCIE |
| On-demand, per month | $1,363 | $273 | A100 PCIE |
| Market median (reference) | $2.67 | $0.54 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.47 | $1.20 | A100 PCIE (better value) |
| $ per GB of VRAM per hour | $0.0233 | $0.0047 | A100 PCIE (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 80 GB vs 80 GB

| Workload | H100 PCIE | A100 PCIE |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~32B |
| Largest LLM at 4-bit, one card | ~123B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Both are 80 GB PCIe cards; the H100 PCIe adds FP8, roughly 2.4× the FP16 tensor throughput and Hopper's Transformer Engine. The A100 PCIe is the cheaper 80 GB and remains excellent for quantized 70B inference and mid-size training.

- **Cheaper per hour:** A100 PCIE ($0.374 vs $1.867, −80%).
- **More VRAM:** H100 PCIE (80 GB vs 80 GB).
- **More FP16 throughput:** H100 PCIE (about 2.4×).
- **Best value per TFLOP-hour:** A100 PCIE.
- **Best value per GB of VRAM:** A100 PCIE.
- **Multi-GPU:** H100 PCIE over PCIe · A100 PCIE over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs H100 PCIE $1.428 vs $1.867 per hour](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)
- [A100 SXM4 vs A100 PCIE $0.560 vs $0.374 per hour](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [L40S vs A100 PCIE $0.514 vs $0.374 per hour](https://powergpu.ai/compare/l40s-vs-a100-pcie)
- [L40S vs H100 PCIE $0.514 vs $1.867 per hour](https://powergpu.ai/compare/l40s-vs-h100-pcie)
- [RTX PRO 6000 WS vs H100 PCIE $1.040 vs $1.867 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie)

## H100 PCIE vs A100 PCIE: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H100 PCIE faster than the A100 PCIE?**

On dense FP16 tensor throughput the H100 PCIE leads by about 2.4× (756 vs 312 TFLOPS). Memory bandwidth matters as much for inference: H100 PCIE 2,000 GB/s vs A100 PCIE 1,935 GB/s.

**Which is cheaper to rent, the H100 PCIE or the A100 PCIE?**

The A100 PCIE: $0.374/hr on-demand versus $1.867/hr — 80% less. Interruptible rates are $0.933 (H100 PCIE) and $0.187 (A100 PCIE). Per TFLOP-hour the better value is the A100 PCIE.

**Which has more VRAM and what does that change?**

The H100 PCIE has 80 GB versus 80 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 73 × H100 PCIE and 46 × A100 PCIE are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h100-pcie-vs-a100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie -->

---
title: "A100 SXM4 vs A100 PCIE: Specs & Price per Hour (2026) | PowerGPU"
description: "A100 SXM4 vs A100 PCIE: 80 vs 80 GB VRAM, 312 vs 312 FP16 TFLOPS, $0.560 vs $0.374/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# A100 SXM4 vs A100 PCIE: specs, price per hour, which to rent

NVIDIA A100 SXM4 (80 GB, **$0.560** /hr) against NVIDIA A100 PCIE (80 GB, **$0.374** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA A100 SXM4 Ampere · 80 GB HBM2e · 312 FP16 TFLOPS · $0.560/hr on-demand · $0.280/hr interruptible · 95 online](https://powergpu.ai/gpu/a100-sxm4)
- [NVIDIA A100 PCIE Ampere · 80 GB HBM2e · 312 FP16 TFLOPS · $0.374/hr on-demand · $0.187/hr interruptible · 46 online](https://powergpu.ai/gpu/a100-pcie)

## A100 SXM4 vs A100 PCIE specifications

Public NVIDIA figures (dense, non-sparsity). The last column is A100 SXM4 relative to A100 PCIE.

| Spec | A100 SXM4 | A100 PCIE | Difference |
| --- | --- | --- | --- |
| Architecture | Ampere (2020) | Ampere (2021) | — |
| VRAM | 80 GB HBM2e | 80 GB HBM2e | same |
| Memory bandwidth | 2,039 GB/s | 1,935 GB/s | +5% |
| FP16 tensor (dense) | 312 TFLOPS | 312 TFLOPS | same |
| FP32 | 19.5 TFLOPS | 19.5 TFLOPS | same |
| CUDA cores | 6,912 | 6,912 | same |
| TDP | 400 W | 300 W | +33% |
| PCIe · NVLink | Gen 4.0 · NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 220 | 220 | same |
| Max GPUs per machine | 8× | 8× | — |

## A100 SXM4 vs A100 PCIE price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | A100 SXM4 | A100 PCIE | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.560 | $0.374 | A100 PCIE (−33%) |
| Interruptible, per GPU-hour | $0.280 | $0.187 | A100 PCIE |
| Reserved (3 mo), per GPU-hour | $0.364 | $0.243 | A100 PCIE |
| On-demand, per month | $409 | $273 | A100 PCIE |
| Market median (reference) | $0.80 | $0.54 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.79 | $1.20 | A100 PCIE (better value) |
| $ per GB of VRAM per hour | $0.0070 | $0.0047 | A100 PCIE (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 80 GB vs 80 GB

| Workload | A100 SXM4 | A100 PCIE |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~32B |
| Largest LLM at 4-bit, one card | ~123B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Identical GPU, different package: the SXM4 part has 600 GB/s NVLink, higher sustained clocks and a 400 W envelope; the PCIe card runs at 300 W without the fabric. Rent SXM4 for multi-GPU training, PCIe for single-card jobs.

- **Cheaper per hour:** A100 PCIE ($0.374 vs $0.560, −33%).
- **More VRAM:** A100 SXM4 (80 GB vs 80 GB).
- **More FP16 throughput:** A100 SXM4 (about 1.0×).
- **Best value per TFLOP-hour:** A100 PCIE.
- **Best value per GB of VRAM:** A100 PCIE.
- **Multi-GPU:** A100 SXM4 with NVLink · A100 PCIE over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs A100 SXM4 $1.428 vs $0.560 per hour](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4)
- [H100 PCIE vs A100 PCIE $1.867 vs $0.374 per hour](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [L40S vs A100 PCIE $0.514 vs $0.374 per hour](https://powergpu.ai/compare/l40s-vs-a100-pcie)
- [RTX 5090 vs A100 SXM4 $0.439 vs $0.560 per hour](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4)

## A100 SXM4 vs A100 PCIE: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the A100 SXM4 faster than the A100 PCIE?**

On dense FP16 tensor throughput the A100 SXM4 leads by about 1.0× (312 vs 312 TFLOPS). Memory bandwidth matters as much for inference: A100 SXM4 2,039 GB/s vs A100 PCIE 1,935 GB/s.

**Which is cheaper to rent, the A100 SXM4 or the A100 PCIE?**

The A100 PCIE: $0.374/hr on-demand versus $0.560/hr — 33% less. Interruptible rates are $0.280 (A100 SXM4) and $0.187 (A100 PCIE). Per TFLOP-hour the better value is the A100 PCIE.

**Which has more VRAM and what does that change?**

The A100 SXM4 has 80 GB versus 80 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 95 × A100 SXM4 and 46 × A100 PCIE are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-5090-vs-rtx-4090 -->

---
title: "RTX 5090 vs RTX 4090: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 5090 vs RTX 4090: 32 vs 24 GB VRAM, 419 vs 330 FP16 TFLOPS, $0.439 vs $0.327/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-5090-vs-rtx-4090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 5090 vs RTX 4090: specs, price per hour, which to rent

NVIDIA RTX 5090 (32 GB, **$0.439** /hr) against NVIDIA RTX 4090 (24 GB, **$0.327** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 5090 Blackwell · 32 GB GDDR7 · 419 FP16 TFLOPS · $0.439/hr on-demand · $0.219/hr interruptible · 922 online](https://powergpu.ai/gpu/rtx-5090)
- [NVIDIA RTX 4090 Ada Lovelace · 24 GB GDDR6X · 330 FP16 TFLOPS · $0.327/hr on-demand · $0.163/hr interruptible · 461 online](https://powergpu.ai/gpu/rtx-4090)

## RTX 5090 vs RTX 4090 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 5090 relative to RTX 4090.

| Spec | RTX 5090 | RTX 4090 | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2025) | Ada Lovelace (2022) | — |
| VRAM | 32 GB GDDR7 | 24 GB GDDR6X | +33% |
| Memory bandwidth | 1,792 GB/s | 1,008 GB/s | +78% |
| FP16 tensor (dense) | 419 TFLOPS | 330 TFLOPS | +27% |
| FP32 | 104.8 TFLOPS | 82.6 TFLOPS | +27% |
| CUDA cores | 21,760 | 16,384 | +33% |
| TDP | 575 W | 450 W | +28% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 295 | 232 | +27% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 5090 vs RTX 4090 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 5090 | RTX 4090 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.439 | $0.327 | RTX 4090 (−26%) |
| Interruptible, per GPU-hour | $0.219 | $0.163 | RTX 4090 |
| Reserved (3 mo), per GPU-hour | $0.285 | $0.212 | RTX 4090 |
| On-demand, per month | $320 | $239 | RTX 4090 |
| Market median (reference) | $0.63 | $0.47 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.05 | $0.99 | RTX 4090 (better value) |
| $ per GB of VRAM per hour | $0.0137 | $0.0136 | RTX 4090 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 32 GB vs 24 GB

| Workload | RTX 5090 | RTX 4090 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~13B | ~8B |
| Largest LLM at 4-bit, one card | ~49B | ~32B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The RTX 5090 adds 8 GB (32 vs 24 GB), 78% more memory bandwidth and FP4 tensor cores, landing 30–70% faster on LLM inference; the RTX 4090 is cheaper per hour with the deeper software ecosystem. Rent the 5090 for memory-bound serving and video models, the 4090 for image generation and budget LoRA runs.

- **Cheaper per hour:** RTX 4090 ($0.327 vs $0.439, −26%).
- **More VRAM:** RTX 5090 (32 GB vs 24 GB).
- **More FP16 throughput:** RTX 5090 (about 1.3×).
- **Best value per TFLOP-hour:** RTX 4090.
- **Best value per GB of VRAM:** RTX 4090.
- **Multi-GPU:** RTX 5090 over PCIe · RTX 4090 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 4090 vs RTX 3090 $0.327 vs $0.108 per hour](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [RTX 5090 vs L40S $0.439 vs $0.514 per hour](https://powergpu.ai/compare/rtx-5090-vs-l40s)
- [RTX 5090 vs A100 SXM4 $0.439 vs $0.560 per hour](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4)
- [RTX PRO 6000 WS vs RTX 5090 $1.040 vs $0.439 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090)

## RTX 5090 vs RTX 4090: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 5090 faster than the RTX 4090?**

On dense FP16 tensor throughput the RTX 5090 leads by about 1.3× (419 vs 330 TFLOPS). Memory bandwidth matters as much for inference: RTX 5090 1,792 GB/s vs RTX 4090 1,008 GB/s.

**Which is cheaper to rent, the RTX 5090 or the RTX 4090?**

The RTX 4090: $0.327/hr on-demand versus $0.439/hr — 26% less. Interruptible rates are $0.219 (RTX 5090) and $0.163 (RTX 4090). Per TFLOP-hour the better value is the RTX 4090.

**Which has more VRAM and what does that change?**

The RTX 5090 has 32 GB versus 24 GB. In LLM terms that is roughly a 13B FP16 model (or ~49B in 4-bit) on one card against 8B FP16 (~32B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 922 × RTX 5090 and 461 × RTX 4090 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-5090-vs-rtx-4090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-4090-vs-rtx-3090 -->

---
title: "RTX 4090 vs RTX 3090: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 4090 vs RTX 3090: 24 vs 24 GB VRAM, 330 vs 142 FP16 TFLOPS, $0.327 vs $0.108/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-4090-vs-rtx-3090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 4090 vs RTX 3090: specs, price per hour, which to rent

NVIDIA RTX 4090 (24 GB, **$0.327** /hr) against NVIDIA RTX 3090 (24 GB, **$0.108** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 4090 Ada Lovelace · 24 GB GDDR6X · 330 FP16 TFLOPS · $0.327/hr on-demand · $0.163/hr interruptible · 461 online](https://powergpu.ai/gpu/rtx-4090)
- [NVIDIA RTX 3090 Ampere · 24 GB GDDR6X · 142 FP16 TFLOPS · $0.108/hr on-demand · $0.054/hr interruptible · 345 online](https://powergpu.ai/gpu/rtx-3090)

## RTX 4090 vs RTX 3090 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 4090 relative to RTX 3090.

| Spec | RTX 4090 | RTX 3090 | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2022) | Ampere (2020) | — |
| VRAM | 24 GB GDDR6X | 24 GB GDDR6X | same |
| Memory bandwidth | 1,008 GB/s | 936 GB/s | +8% |
| FP16 tensor (dense) | 330 TFLOPS | 142 TFLOPS | +132% |
| FP32 | 82.6 TFLOPS | 35.6 TFLOPS | +132% |
| CUDA cores | 16,384 | 10,496 | +56% |
| TDP | 450 W | 350 W | +29% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 232 | 100 | +132% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 4090 vs RTX 3090 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 4090 | RTX 3090 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.327 | $0.108 | RTX 3090 (−67%) |
| Interruptible, per GPU-hour | $0.163 | $0.054 | RTX 3090 |
| Reserved (3 mo), per GPU-hour | $0.212 | $0.070 | RTX 3090 |
| On-demand, per month | $239 | $79 | RTX 3090 |
| Market median (reference) | $0.47 | $0.15 | — |
| $ per 1,000 FP16 TFLOP-hours | $0.99 | $0.76 | RTX 3090 (better value) |
| $ per GB of VRAM per hour | $0.0136 | $0.0045 | RTX 3090 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 24 GB vs 24 GB

| Workload | RTX 4090 | RTX 3090 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~8B | ~8B |
| Largest LLM at 4-bit, one card | ~32B | ~32B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Same 24 GB, but the 4090 has Ada tensor cores, about 2.3× the FP16 throughput and a newer NVENC; the 3090 costs far less per hour. If throughput drives your bill, the 4090 wins; if VRAM capacity is all you need, the 3090 is the cheapest 24 GB rental.

- **Cheaper per hour:** RTX 3090 ($0.108 vs $0.327, −67%).
- **More VRAM:** RTX 4090 (24 GB vs 24 GB).
- **More FP16 throughput:** RTX 4090 (about 2.3×).
- **Best value per TFLOP-hour:** RTX 3090.
- **Best value per GB of VRAM:** RTX 3090.
- **Multi-GPU:** RTX 4090 over PCIe · RTX 3090 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [RTX A6000 vs RTX 4090 $0.281 vs $0.327 per hour](https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090)
- [Tesla V100 vs RTX 3090 $0.130 vs $0.108 per hour](https://powergpu.ai/compare/tesla-v100-vs-rtx-3090)
- [RTX 3090 vs RTX 3060 $0.108 vs $0.042 per hour](https://powergpu.ai/compare/rtx-3090-vs-rtx-3060)

## RTX 4090 vs RTX 3090: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 4090 faster than the RTX 3090?**

On dense FP16 tensor throughput the RTX 4090 leads by about 2.3× (330 vs 142 TFLOPS). Memory bandwidth matters as much for inference: RTX 4090 1,008 GB/s vs RTX 3090 936 GB/s.

**Which is cheaper to rent, the RTX 4090 or the RTX 3090?**

The RTX 3090: $0.108/hr on-demand versus $0.327/hr — 67% less. Interruptible rates are $0.163 (RTX 4090) and $0.054 (RTX 3090). Per TFLOP-hour the better value is the RTX 3090.

**Which has more VRAM and what does that change?**

The RTX 4090 has 24 GB versus 24 GB. In LLM terms that is roughly a 8B FP16 model (or ~32B in 4-bit) on one card against 8B FP16 (~32B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 461 × RTX 4090 and 345 × RTX 3090 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-4090-vs-rtx-3090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-4090-vs-a100-pcie -->

---
title: "RTX 4090 vs A100 PCIE: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 4090 vs A100 PCIE: 24 vs 80 GB VRAM, 330 vs 312 FP16 TFLOPS, $0.327 vs $0.374/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-4090-vs-a100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 4090 vs A100 PCIE: specs, price per hour, which to rent

NVIDIA RTX 4090 (24 GB, **$0.327** /hr) against NVIDIA A100 PCIE (80 GB, **$0.374** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 4090 Ada Lovelace · 24 GB GDDR6X · 330 FP16 TFLOPS · $0.327/hr on-demand · $0.163/hr interruptible · 461 online](https://powergpu.ai/gpu/rtx-4090)
- [NVIDIA A100 PCIE Ampere · 80 GB HBM2e · 312 FP16 TFLOPS · $0.374/hr on-demand · $0.187/hr interruptible · 46 online](https://powergpu.ai/gpu/a100-pcie)

## RTX 4090 vs A100 PCIE specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 4090 relative to A100 PCIE.

| Spec | RTX 4090 | A100 PCIE | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2022) | Ampere (2021) | — |
| VRAM | 24 GB GDDR6X | 80 GB HBM2e | −70% |
| Memory bandwidth | 1,008 GB/s | 1,935 GB/s | −48% |
| FP16 tensor (dense) | 330 TFLOPS | 312 TFLOPS | +6% |
| FP32 | 82.6 TFLOPS | 19.5 TFLOPS | +324% |
| CUDA cores | 16,384 | 6,912 | +137% |
| TDP | 450 W | 300 W | +50% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 232 | 220 | +5% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 4090 vs A100 PCIE price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 4090 | A100 PCIE | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.327 | $0.374 | RTX 4090 (−13%) |
| Interruptible, per GPU-hour | $0.163 | $0.187 | RTX 4090 |
| Reserved (3 mo), per GPU-hour | $0.212 | $0.243 | RTX 4090 |
| On-demand, per month | $239 | $273 | RTX 4090 |
| Market median (reference) | $0.47 | $0.54 | — |
| $ per 1,000 FP16 TFLOP-hours | $0.99 | $1.20 | RTX 4090 (better value) |
| $ per GB of VRAM per hour | $0.0136 | $0.0047 | A100 PCIE (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 24 GB vs 80 GB

| Workload | RTX 4090 | A100 PCIE |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~8B | ~32B |
| Largest LLM at 4-bit, one card | ~32B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The consumer RTX 4090 is faster per clock on FP16 and much cheaper per hour; the A100 PCIe has 80 GB of HBM2e, ECC and NVLink pairing. Under 24 GB the 4090 wins almost every job on cost; above it — 70B models, big batches — only the A100 fits.

- **Cheaper per hour:** RTX 4090 ($0.327 vs $0.374, −13%).
- **More VRAM:** A100 PCIE (80 GB vs 24 GB).
- **More FP16 throughput:** RTX 4090 (about 1.1×).
- **Best value per TFLOP-hour:** RTX 4090.
- **Best value per GB of VRAM:** A100 PCIE.
- **Multi-GPU:** RTX 4090 over PCIe · A100 PCIE over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 PCIE vs A100 PCIE $1.867 vs $0.374 per hour](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie)
- [A100 SXM4 vs A100 PCIE $0.560 vs $0.374 per hour](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie)
- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 4090 vs RTX 3090 $0.327 vs $0.108 per hour](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [L40S vs A100 PCIE $0.514 vs $0.374 per hour](https://powergpu.ai/compare/l40s-vs-a100-pcie)

## RTX 4090 vs A100 PCIE: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 4090 faster than the A100 PCIE?**

On dense FP16 tensor throughput the RTX 4090 leads by about 1.1× (330 vs 312 TFLOPS). Memory bandwidth matters as much for inference: RTX 4090 1,008 GB/s vs A100 PCIE 1,935 GB/s.

**Which is cheaper to rent, the RTX 4090 or the A100 PCIE?**

The RTX 4090: $0.327/hr on-demand versus $0.374/hr — 13% less. Interruptible rates are $0.163 (RTX 4090) and $0.187 (A100 PCIE). Per TFLOP-hour the better value is the RTX 4090.

**Which has more VRAM and what does that change?**

The A100 PCIE has 80 GB versus 24 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 8B FP16 (~32B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 461 × RTX 4090 and 46 × A100 PCIE are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-4090-vs-a100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-4090-vs-l40s -->

---
title: "RTX 4090 vs L40S: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 4090 vs L40S: 24 vs 48 GB VRAM, 330 vs 362 FP16 TFLOPS, $0.327 vs $0.514/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-4090-vs-l40s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 4090 vs L40S: specs, price per hour, which to rent

NVIDIA RTX 4090 (24 GB, **$0.327** /hr) against NVIDIA L40S (48 GB, **$0.514** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 4090 Ada Lovelace · 24 GB GDDR6X · 330 FP16 TFLOPS · $0.327/hr on-demand · $0.163/hr interruptible · 461 online](https://powergpu.ai/gpu/rtx-4090)
- [NVIDIA L40S Ada Lovelace · 48 GB GDDR6 · 362 FP16 TFLOPS · $0.514/hr on-demand · $0.257/hr interruptible · 107 online](https://powergpu.ai/gpu/l40s)

## RTX 4090 vs L40S specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 4090 relative to L40S.

| Spec | RTX 4090 | L40S | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2022) | Ada Lovelace (2023) | — |
| VRAM | 24 GB GDDR6X | 48 GB GDDR6 | −50% |
| Memory bandwidth | 1,008 GB/s | 864 GB/s | +17% |
| FP16 tensor (dense) | 330 TFLOPS | 362 TFLOPS | −9% |
| FP32 | 82.6 TFLOPS | 91.6 TFLOPS | −10% |
| CUDA cores | 16,384 | 18,176 | −10% |
| TDP | 450 W | 350 W | +29% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 232 | 255 | −9% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 4090 vs L40S price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 4090 | L40S | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.327 | $0.514 | RTX 4090 (−36%) |
| Interruptible, per GPU-hour | $0.163 | $0.257 | RTX 4090 |
| Reserved (3 mo), per GPU-hour | $0.212 | $0.334 | RTX 4090 |
| On-demand, per month | $239 | $375 | RTX 4090 |
| Market median (reference) | $0.47 | $0.74 | — |
| $ per 1,000 FP16 TFLOP-hours | $0.99 | $1.42 | RTX 4090 (better value) |
| $ per GB of VRAM per hour | $0.0136 | $0.0107 | L40S (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 24 GB vs 48 GB

| Workload | RTX 4090 | L40S |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~8B | ~14B |
| Largest LLM at 4-bit, one card | ~32B | ~72B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The L40S is the datacenter Ada card: 48 GB, passive cooling, FP8, built for 24/7 duty; the RTX 4090 has similar per-clock throughput, 24 GB and a lower rate. Serve production endpoints on the L40S; run interactive and batch generation on the 4090.

- **Cheaper per hour:** RTX 4090 ($0.327 vs $0.514, −36%).
- **More VRAM:** L40S (48 GB vs 24 GB).
- **More FP16 throughput:** L40S (about 1.1×).
- **Best value per TFLOP-hour:** RTX 4090.
- **Best value per GB of VRAM:** L40S.
- **Multi-GPU:** RTX 4090 over PCIe · L40S over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 4090 vs RTX 3090 $0.327 vs $0.108 per hour](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [L40S vs A100 PCIE $0.514 vs $0.374 per hour](https://powergpu.ai/compare/l40s-vs-a100-pcie)
- [L40S vs H100 PCIE $0.514 vs $1.867 per hour](https://powergpu.ai/compare/l40s-vs-h100-pcie)
- [RTX 5090 vs L40S $0.439 vs $0.514 per hour](https://powergpu.ai/compare/rtx-5090-vs-l40s)

## RTX 4090 vs L40S: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the L40S faster than the RTX 4090?**

On dense FP16 tensor throughput the L40S leads by about 1.1× (362 vs 330 TFLOPS). Memory bandwidth matters as much for inference: RTX 4090 1,008 GB/s vs L40S 864 GB/s.

**Which is cheaper to rent, the RTX 4090 or the L40S?**

The RTX 4090: $0.327/hr on-demand versus $0.514/hr — 36% less. Interruptible rates are $0.163 (RTX 4090) and $0.257 (L40S). Per TFLOP-hour the better value is the RTX 4090.

**Which has more VRAM and what does that change?**

The L40S has 48 GB versus 24 GB. In LLM terms that is roughly a 14B FP16 model (or ~72B in 4-bit) on one card against 8B FP16 (~32B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 461 × RTX 4090 and 107 × L40S are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-4090-vs-l40s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/l40s-vs-a100-pcie -->

---
title: "L40S vs A100 PCIE: Specs & Price per Hour (2026) | PowerGPU"
description: "L40S vs A100 PCIE: 48 vs 80 GB VRAM, 362 vs 312 FP16 TFLOPS, $0.514 vs $0.374/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/l40s-vs-a100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# L40S vs A100 PCIE: specs, price per hour, which to rent

NVIDIA L40S (48 GB, **$0.514** /hr) against NVIDIA A100 PCIE (80 GB, **$0.374** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA L40S Ada Lovelace · 48 GB GDDR6 · 362 FP16 TFLOPS · $0.514/hr on-demand · $0.257/hr interruptible · 107 online](https://powergpu.ai/gpu/l40s)
- [NVIDIA A100 PCIE Ampere · 80 GB HBM2e · 312 FP16 TFLOPS · $0.374/hr on-demand · $0.187/hr interruptible · 46 online](https://powergpu.ai/gpu/a100-pcie)

## L40S vs A100 PCIE specifications

Public NVIDIA figures (dense, non-sparsity). The last column is L40S relative to A100 PCIE.

| Spec | L40S | A100 PCIE | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2023) | Ampere (2021) | — |
| VRAM | 48 GB GDDR6 | 80 GB HBM2e | −40% |
| Memory bandwidth | 864 GB/s | 1,935 GB/s | −55% |
| FP16 tensor (dense) | 362 TFLOPS | 312 TFLOPS | +16% |
| FP32 | 91.6 TFLOPS | 19.5 TFLOPS | +370% |
| CUDA cores | 18,176 | 6,912 | +163% |
| TDP | 350 W | 300 W | +17% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 255 | 220 | +16% |
| Max GPUs per machine | 8× | 8× | — |

## L40S vs A100 PCIE price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | L40S | A100 PCIE | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.514 | $0.374 | A100 PCIE (−27%) |
| Interruptible, per GPU-hour | $0.257 | $0.187 | A100 PCIE |
| Reserved (3 mo), per GPU-hour | $0.334 | $0.243 | A100 PCIE |
| On-demand, per month | $375 | $273 | A100 PCIE |
| Market median (reference) | $0.74 | $0.54 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.42 | $1.20 | A100 PCIE (better value) |
| $ per GB of VRAM per hour | $0.0107 | $0.0047 | A100 PCIE (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 48 GB vs 80 GB

| Workload | L40S | A100 PCIE |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~14B | ~32B |
| Largest LLM at 4-bit, one card | ~72B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The L40S (48 GB GDDR6, Ada, FP8) is faster for FP16/FP8 inference and image generation; the A100 PCIe (80 GB HBM2e) has more memory, more bandwidth and FP64. Pick the L40S for diffusion and ≤32B serving, the A100 for 70B models and training.

- **Cheaper per hour:** A100 PCIE ($0.374 vs $0.514, −27%).
- **More VRAM:** A100 PCIE (80 GB vs 48 GB).
- **More FP16 throughput:** L40S (about 1.2×).
- **Best value per TFLOP-hour:** A100 PCIE.
- **Best value per GB of VRAM:** A100 PCIE.
- **Multi-GPU:** L40S over PCIe · A100 PCIE over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 PCIE vs A100 PCIE $1.867 vs $0.374 per hour](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie)
- [A100 SXM4 vs A100 PCIE $0.560 vs $0.374 per hour](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [L40S vs H100 PCIE $0.514 vs $1.867 per hour](https://powergpu.ai/compare/l40s-vs-h100-pcie)
- [RTX 5090 vs L40S $0.439 vs $0.514 per hour](https://powergpu.ai/compare/rtx-5090-vs-l40s)

## L40S vs A100 PCIE: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the L40S faster than the A100 PCIE?**

On dense FP16 tensor throughput the L40S leads by about 1.2× (362 vs 312 TFLOPS). Memory bandwidth matters as much for inference: L40S 864 GB/s vs A100 PCIE 1,935 GB/s.

**Which is cheaper to rent, the L40S or the A100 PCIE?**

The A100 PCIE: $0.374/hr on-demand versus $0.514/hr — 27% less. Interruptible rates are $0.257 (L40S) and $0.187 (A100 PCIE). Per TFLOP-hour the better value is the A100 PCIE.

**Which has more VRAM and what does that change?**

The A100 PCIE has 80 GB versus 48 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 14B FP16 (~72B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 107 × L40S and 46 × A100 PCIE are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/l40s-vs-a100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/l40s-vs-h100-pcie -->

---
title: "L40S vs H100 PCIE: Specs & Price per Hour (2026) | PowerGPU"
description: "L40S vs H100 PCIE: 48 vs 80 GB VRAM, 362 vs 756 FP16 TFLOPS, $0.514 vs $1.867/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/l40s-vs-h100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# L40S vs H100 PCIE: specs, price per hour, which to rent

NVIDIA L40S (48 GB, **$0.514** /hr) against NVIDIA H100 PCIE (80 GB, **$1.867** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA L40S Ada Lovelace · 48 GB GDDR6 · 362 FP16 TFLOPS · $0.514/hr on-demand · $0.257/hr interruptible · 107 online](https://powergpu.ai/gpu/l40s)
- [NVIDIA H100 PCIE Hopper · 80 GB HBM2e · 756 FP16 TFLOPS · $1.867/hr on-demand · $0.933/hr interruptible · 73 online](https://powergpu.ai/gpu/h100-pcie)

## L40S vs H100 PCIE specifications

Public NVIDIA figures (dense, non-sparsity). The last column is L40S relative to H100 PCIE.

| Spec | L40S | H100 PCIE | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2023) | Hopper (2022) | — |
| VRAM | 48 GB GDDR6 | 80 GB HBM2e | −40% |
| Memory bandwidth | 864 GB/s | 2,000 GB/s | −57% |
| FP16 tensor (dense) | 362 TFLOPS | 756 TFLOPS | −52% |
| FP32 | 91.6 TFLOPS | 51.0 TFLOPS | +80% |
| CUDA cores | 18,176 | 14,592 | +25% |
| TDP | 350 W | 350 W | same |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 5.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 255 | 532 | −52% |
| Max GPUs per machine | 8× | 8× | — |

## L40S vs H100 PCIE price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | L40S | H100 PCIE | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.514 | $1.867 | L40S (−72%) |
| Interruptible, per GPU-hour | $0.257 | $0.933 | L40S |
| Reserved (3 mo), per GPU-hour | $0.334 | $1.213 | L40S |
| On-demand, per month | $375 | $1,363 | L40S |
| Market median (reference) | $0.74 | $2.67 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.42 | $2.47 | L40S (better value) |
| $ per GB of VRAM per hour | $0.0107 | $0.0233 | L40S (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 48 GB vs 80 GB

| Workload | L40S | H100 PCIE |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~14B | ~32B |
| Largest LLM at 4-bit, one card | ~72B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The H100 PCIe has 80 GB of HBM2e, 2 TB/s and roughly double the tensor throughput; the L40S has 48 GB of GDDR6 and costs a lot less. The L40S is the value choice for models under 32B; the H100 PCIe for 70B and throughput-bound serving.

- **Cheaper per hour:** L40S ($0.514 vs $1.867, −72%).
- **More VRAM:** H100 PCIE (80 GB vs 48 GB).
- **More FP16 throughput:** H100 PCIE (about 2.1×).
- **Best value per TFLOP-hour:** L40S.
- **Best value per GB of VRAM:** L40S.
- **Multi-GPU:** L40S over PCIe · H100 PCIE over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs H100 PCIE $1.428 vs $1.867 per hour](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)
- [H100 PCIE vs A100 PCIE $1.867 vs $0.374 per hour](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [L40S vs A100 PCIE $0.514 vs $0.374 per hour](https://powergpu.ai/compare/l40s-vs-a100-pcie)
- [RTX 5090 vs L40S $0.439 vs $0.514 per hour](https://powergpu.ai/compare/rtx-5090-vs-l40s)
- [RTX PRO 6000 WS vs H100 PCIE $1.040 vs $1.867 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie)

## L40S vs H100 PCIE: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H100 PCIE faster than the L40S?**

On dense FP16 tensor throughput the H100 PCIE leads by about 2.1× (756 vs 362 TFLOPS). Memory bandwidth matters as much for inference: L40S 864 GB/s vs H100 PCIE 2,000 GB/s.

**Which is cheaper to rent, the L40S or the H100 PCIE?**

The L40S: $0.514/hr on-demand versus $1.867/hr — 72% less. Interruptible rates are $0.257 (L40S) and $0.933 (H100 PCIE). Per TFLOP-hour the better value is the L40S.

**Which has more VRAM and what does that change?**

The H100 PCIE has 80 GB versus 48 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 14B FP16 (~72B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 107 × L40S and 73 × H100 PCIE are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/l40s-vs-h100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-5090-vs-l40s -->

---
title: "RTX 5090 vs L40S: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 5090 vs L40S: 32 vs 48 GB VRAM, 419 vs 362 FP16 TFLOPS, $0.439 vs $0.514/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-5090-vs-l40s
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 5090 vs L40S: specs, price per hour, which to rent

NVIDIA RTX 5090 (32 GB, **$0.439** /hr) against NVIDIA L40S (48 GB, **$0.514** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 5090 Blackwell · 32 GB GDDR7 · 419 FP16 TFLOPS · $0.439/hr on-demand · $0.219/hr interruptible · 922 online](https://powergpu.ai/gpu/rtx-5090)
- [NVIDIA L40S Ada Lovelace · 48 GB GDDR6 · 362 FP16 TFLOPS · $0.514/hr on-demand · $0.257/hr interruptible · 107 online](https://powergpu.ai/gpu/l40s)

## RTX 5090 vs L40S specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 5090 relative to L40S.

| Spec | RTX 5090 | L40S | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2025) | Ada Lovelace (2023) | — |
| VRAM | 32 GB GDDR7 | 48 GB GDDR6 | −33% |
| Memory bandwidth | 1,792 GB/s | 864 GB/s | +107% |
| FP16 tensor (dense) | 419 TFLOPS | 362 TFLOPS | +16% |
| FP32 | 104.8 TFLOPS | 91.6 TFLOPS | +14% |
| CUDA cores | 21,760 | 18,176 | +20% |
| TDP | 575 W | 350 W | +64% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 295 | 255 | +16% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 5090 vs L40S price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 5090 | L40S | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.439 | $0.514 | RTX 5090 (−15%) |
| Interruptible, per GPU-hour | $0.219 | $0.257 | RTX 5090 |
| Reserved (3 mo), per GPU-hour | $0.285 | $0.334 | RTX 5090 |
| On-demand, per month | $320 | $375 | RTX 5090 |
| Market median (reference) | $0.63 | $0.74 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.05 | $1.42 | RTX 5090 (better value) |
| $ per GB of VRAM per hour | $0.0137 | $0.0107 | L40S (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 32 GB vs 48 GB

| Workload | RTX 5090 | L40S |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~13B | ~14B |
| Largest LLM at 4-bit, one card | ~49B | ~72B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The RTX 5090 (32 GB GDDR7, 1.79 TB/s) is faster and cheaper; the L40S (48 GB, passive, ECC) fits bigger models and is built for continuous datacenter duty. Choose by VRAM and uptime needs rather than raw speed.

- **Cheaper per hour:** RTX 5090 ($0.439 vs $0.514, −15%).
- **More VRAM:** L40S (48 GB vs 32 GB).
- **More FP16 throughput:** RTX 5090 (about 1.2×).
- **Best value per TFLOP-hour:** RTX 5090.
- **Best value per GB of VRAM:** L40S.
- **Multi-GPU:** RTX 5090 over PCIe · L40S over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [L40S vs A100 PCIE $0.514 vs $0.374 per hour](https://powergpu.ai/compare/l40s-vs-a100-pcie)
- [L40S vs H100 PCIE $0.514 vs $1.867 per hour](https://powergpu.ai/compare/l40s-vs-h100-pcie)
- [RTX 5090 vs A100 SXM4 $0.439 vs $0.560 per hour](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4)
- [RTX PRO 6000 WS vs RTX 5090 $1.040 vs $0.439 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090)

## RTX 5090 vs L40S: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 5090 faster than the L40S?**

On dense FP16 tensor throughput the RTX 5090 leads by about 1.2× (419 vs 362 TFLOPS). Memory bandwidth matters as much for inference: RTX 5090 1,792 GB/s vs L40S 864 GB/s.

**Which is cheaper to rent, the RTX 5090 or the L40S?**

The RTX 5090: $0.439/hr on-demand versus $0.514/hr — 15% less. Interruptible rates are $0.219 (RTX 5090) and $0.257 (L40S). Per TFLOP-hour the better value is the RTX 5090.

**Which has more VRAM and what does that change?**

The L40S has 48 GB versus 32 GB. In LLM terms that is roughly a 14B FP16 model (or ~72B in 4-bit) on one card against 13B FP16 (~49B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 922 × RTX 5090 and 107 × L40S are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-5090-vs-l40s · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4 -->

---
title: "RTX 5090 vs A100 SXM4: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 5090 vs A100 SXM4: 32 vs 80 GB VRAM, 419 vs 312 FP16 TFLOPS, $0.439 vs $0.560/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 5090 vs A100 SXM4: specs, price per hour, which to rent

NVIDIA RTX 5090 (32 GB, **$0.439** /hr) against NVIDIA A100 SXM4 (80 GB, **$0.560** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 5090 Blackwell · 32 GB GDDR7 · 419 FP16 TFLOPS · $0.439/hr on-demand · $0.219/hr interruptible · 922 online](https://powergpu.ai/gpu/rtx-5090)
- [NVIDIA A100 SXM4 Ampere · 80 GB HBM2e · 312 FP16 TFLOPS · $0.560/hr on-demand · $0.280/hr interruptible · 95 online](https://powergpu.ai/gpu/a100-sxm4)

## RTX 5090 vs A100 SXM4 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 5090 relative to A100 SXM4.

| Spec | RTX 5090 | A100 SXM4 | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2025) | Ampere (2020) | — |
| VRAM | 32 GB GDDR7 | 80 GB HBM2e | −60% |
| Memory bandwidth | 1,792 GB/s | 2,039 GB/s | −12% |
| FP16 tensor (dense) | 419 TFLOPS | 312 TFLOPS | +34% |
| FP32 | 104.8 TFLOPS | 19.5 TFLOPS | +437% |
| CUDA cores | 21,760 | 6,912 | +215% |
| TDP | 575 W | 400 W | +44% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 4.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 295 | 220 | +34% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 5090 vs A100 SXM4 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 5090 | A100 SXM4 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.439 | $0.560 | RTX 5090 (−22%) |
| Interruptible, per GPU-hour | $0.219 | $0.280 | RTX 5090 |
| Reserved (3 mo), per GPU-hour | $0.285 | $0.364 | RTX 5090 |
| On-demand, per month | $320 | $409 | RTX 5090 |
| Market median (reference) | $0.63 | $0.80 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.05 | $1.79 | RTX 5090 (better value) |
| $ per GB of VRAM per hour | $0.0137 | $0.0070 | A100 SXM4 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 32 GB vs 80 GB

| Workload | RTX 5090 | A100 SXM4 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~13B | ~32B |
| Largest LLM at 4-bit, one card | ~49B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The RTX 5090 beats the A100 SXM4 on raw FP16/FP4 throughput and bandwidth at a lower hourly rate; the A100 has 80 GB of HBM2e, NVLink and FP64. Under 32 GB the 5090 is the better rental; for multi-GPU training and 70B models, the A100.

- **Cheaper per hour:** RTX 5090 ($0.439 vs $0.560, −22%).
- **More VRAM:** A100 SXM4 (80 GB vs 32 GB).
- **More FP16 throughput:** RTX 5090 (about 1.3×).
- **Best value per TFLOP-hour:** RTX 5090.
- **Best value per GB of VRAM:** A100 SXM4.
- **Multi-GPU:** RTX 5090 over PCIe · A100 SXM4 with NVLink.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs A100 SXM4 $1.428 vs $0.560 per hour](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4)
- [A100 SXM4 vs A100 PCIE $0.560 vs $0.374 per hour](https://powergpu.ai/compare/a100-sxm4-vs-a100-pcie)
- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 5090 vs L40S $0.439 vs $0.514 per hour](https://powergpu.ai/compare/rtx-5090-vs-l40s)
- [RTX PRO 6000 WS vs RTX 5090 $1.040 vs $0.439 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090)

## RTX 5090 vs A100 SXM4: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 5090 faster than the A100 SXM4?**

On dense FP16 tensor throughput the RTX 5090 leads by about 1.3× (419 vs 312 TFLOPS). Memory bandwidth matters as much for inference: RTX 5090 1,792 GB/s vs A100 SXM4 2,039 GB/s.

**Which is cheaper to rent, the RTX 5090 or the A100 SXM4?**

The RTX 5090: $0.439/hr on-demand versus $0.560/hr — 22% less. Interruptible rates are $0.219 (RTX 5090) and $0.280 (A100 SXM4). Per TFLOP-hour the better value is the RTX 5090.

**Which has more VRAM and what does that change?**

The A100 SXM4 has 80 GB versus 32 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 13B FP16 (~49B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 922 × RTX 5090 and 95 × A100 SXM4 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada -->

---
title: "RTX PRO 6000 WS vs RTX 6000Ada: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX PRO 6000 WS vs RTX 6000Ada: 96 vs 48 GB VRAM, 505 vs 364 FP16 TFLOPS, $1.040 vs $0.467/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX PRO 6000 WS vs RTX 6000Ada: specs, price per hour, which to rent

NVIDIA RTX PRO 6000 WS (96 GB, **$1.040** /hr) against NVIDIA RTX 6000Ada (48 GB, **$0.467** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX PRO 6000 WS Blackwell · 96 GB GDDR7 · 505 FP16 TFLOPS · $1.040/hr on-demand · $0.520/hr interruptible · 74 online](https://powergpu.ai/gpu/rtx-pro-6000-ws)
- [NVIDIA RTX 6000Ada Ada Lovelace · 48 GB GDDR6 · 364 FP16 TFLOPS · $0.467/hr on-demand · $0.233/hr interruptible · 34 online](https://powergpu.ai/gpu/rtx-6000ada)

## RTX PRO 6000 WS vs RTX 6000Ada specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX PRO 6000 WS relative to RTX 6000Ada.

| Spec | RTX PRO 6000 WS | RTX 6000Ada | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2025) | Ada Lovelace (2022) | — |
| VRAM | 96 GB GDDR7 | 48 GB GDDR6 | +100% |
| Memory bandwidth | 1,792 GB/s | 960 GB/s | +87% |
| FP16 tensor (dense) | 505 TFLOPS | 364 TFLOPS | +39% |
| FP32 | 125.0 TFLOPS | 91.1 TFLOPS | +37% |
| CUDA cores | 24,064 | 18,176 | +32% |
| TDP | 600 W | 300 W | +100% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 356 | 256 | +39% |
| Max GPUs per machine | 8× | 8× | — |

## RTX PRO 6000 WS vs RTX 6000Ada price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX PRO 6000 WS | RTX 6000Ada | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.040 | $0.467 | RTX 6000Ada (−55%) |
| Interruptible, per GPU-hour | $0.520 | $0.233 | RTX 6000Ada |
| Reserved (3 mo), per GPU-hour | $0.676 | $0.303 | RTX 6000Ada |
| On-demand, per month | $759 | $341 | RTX 6000Ada |
| Market median (reference) | $1.49 | $0.67 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.06 | $1.28 | RTX 6000Ada (better value) |
| $ per GB of VRAM per hour | $0.0108 | $0.0097 | RTX 6000Ada (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 96 GB vs 48 GB

| Workload | RTX PRO 6000 WS | RTX 6000Ada |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~14B |
| Largest LLM at 4-bit, one card | ~141B | ~72B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The RTX PRO 6000 Blackwell doubles VRAM (96 vs 48 GB), nearly doubles bandwidth with GDDR7 and adds FP4; the RTX 6000 Ada is cheaper and still a superb 48 GB workstation card. Go Blackwell for 70B single-card inference and huge scenes, Ada for everything under 48 GB.

- **Cheaper per hour:** RTX 6000Ada ($0.467 vs $1.040, −55%).
- **More VRAM:** RTX PRO 6000 WS (96 GB vs 48 GB).
- **More FP16 throughput:** RTX PRO 6000 WS (about 1.4×).
- **Best value per TFLOP-hour:** RTX 6000Ada.
- **Best value per GB of VRAM:** RTX 6000Ada.
- **Multi-GPU:** RTX PRO 6000 WS over PCIe · RTX 6000Ada over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX PRO 6000 WS vs H100 PCIE $1.040 vs $1.867 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie)
- [RTX PRO 6000 WS vs RTX 5090 $1.040 vs $0.439 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090)
- [RTX 6000Ada vs RTX A6000 $0.467 vs $0.281 per hour](https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000)

## RTX PRO 6000 WS vs RTX 6000Ada: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX PRO 6000 WS faster than the RTX 6000Ada?**

On dense FP16 tensor throughput the RTX PRO 6000 WS leads by about 1.4× (505 vs 364 TFLOPS). Memory bandwidth matters as much for inference: RTX PRO 6000 WS 1,792 GB/s vs RTX 6000Ada 960 GB/s.

**Which is cheaper to rent, the RTX PRO 6000 WS or the RTX 6000Ada?**

The RTX 6000Ada: $0.467/hr on-demand versus $1.040/hr — 55% less. Interruptible rates are $0.520 (RTX PRO 6000 WS) and $0.233 (RTX 6000Ada). Per TFLOP-hour the better value is the RTX 6000Ada.

**Which has more VRAM and what does that change?**

The RTX PRO 6000 WS has 96 GB versus 48 GB. In LLM terms that is roughly a 32B FP16 model (or ~141B in 4-bit) on one card against 14B FP16 (~72B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 74 × RTX PRO 6000 WS and 34 × RTX 6000Ada are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie -->

---
title: "RTX PRO 6000 WS vs H100 PCIE: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX PRO 6000 WS vs H100 PCIE: 96 vs 80 GB VRAM, 505 vs 756 FP16 TFLOPS, $1.040 vs $1.867/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX PRO 6000 WS vs H100 PCIE: specs, price per hour, which to rent

NVIDIA RTX PRO 6000 WS (96 GB, **$1.040** /hr) against NVIDIA H100 PCIE (80 GB, **$1.867** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX PRO 6000 WS Blackwell · 96 GB GDDR7 · 505 FP16 TFLOPS · $1.040/hr on-demand · $0.520/hr interruptible · 74 online](https://powergpu.ai/gpu/rtx-pro-6000-ws)
- [NVIDIA H100 PCIE Hopper · 80 GB HBM2e · 756 FP16 TFLOPS · $1.867/hr on-demand · $0.933/hr interruptible · 73 online](https://powergpu.ai/gpu/h100-pcie)

## RTX PRO 6000 WS vs H100 PCIE specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX PRO 6000 WS relative to H100 PCIE.

| Spec | RTX PRO 6000 WS | H100 PCIE | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2025) | Hopper (2022) | — |
| VRAM | 96 GB GDDR7 | 80 GB HBM2e | +20% |
| Memory bandwidth | 1,792 GB/s | 2,000 GB/s | −10% |
| FP16 tensor (dense) | 505 TFLOPS | 756 TFLOPS | −33% |
| FP32 | 125.0 TFLOPS | 51.0 TFLOPS | +145% |
| CUDA cores | 24,064 | 14,592 | +65% |
| TDP | 600 W | 350 W | +71% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 5.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 356 | 532 | −33% |
| Max GPUs per machine | 8× | 8× | — |

## RTX PRO 6000 WS vs H100 PCIE price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX PRO 6000 WS | H100 PCIE | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.040 | $1.867 | RTX PRO 6000 WS (−44%) |
| Interruptible, per GPU-hour | $0.520 | $0.933 | RTX PRO 6000 WS |
| Reserved (3 mo), per GPU-hour | $0.676 | $1.213 | RTX PRO 6000 WS |
| On-demand, per month | $759 | $1,363 | RTX PRO 6000 WS |
| Market median (reference) | $1.49 | $2.67 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.06 | $2.47 | RTX PRO 6000 WS (better value) |
| $ per GB of VRAM per hour | $0.0108 | $0.0233 | RTX PRO 6000 WS (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 96 GB vs 80 GB

| Workload | RTX PRO 6000 WS | H100 PCIE |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~32B |
| Largest LLM at 4-bit, one card | ~141B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

96 GB of GDDR7 and more CUDA cores on the RTX PRO 6000 versus 80 GB of HBM2e and datacenter interconnect on the H100 PCIe. The PRO 6000 wins rendering, FP4 inference and single-card capacity; the H100 wins bandwidth-bound training and FP8 throughput.

- **Cheaper per hour:** RTX PRO 6000 WS ($1.040 vs $1.867, −44%).
- **More VRAM:** RTX PRO 6000 WS (96 GB vs 80 GB).
- **More FP16 throughput:** H100 PCIE (about 1.5×).
- **Best value per TFLOP-hour:** RTX PRO 6000 WS.
- **Best value per GB of VRAM:** RTX PRO 6000 WS.
- **Multi-GPU:** RTX PRO 6000 WS over PCIe · H100 PCIE over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs H100 PCIE $1.428 vs $1.867 per hour](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)
- [H100 PCIE vs A100 PCIE $1.867 vs $0.374 per hour](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie)
- [L40S vs H100 PCIE $0.514 vs $1.867 per hour](https://powergpu.ai/compare/l40s-vs-h100-pcie)
- [RTX PRO 6000 WS vs RTX 6000Ada $1.040 vs $0.467 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada)
- [RTX PRO 6000 WS vs RTX 5090 $1.040 vs $0.439 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090)

## RTX PRO 6000 WS vs H100 PCIE: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H100 PCIE faster than the RTX PRO 6000 WS?**

On dense FP16 tensor throughput the H100 PCIE leads by about 1.5× (756 vs 505 TFLOPS). Memory bandwidth matters as much for inference: RTX PRO 6000 WS 1,792 GB/s vs H100 PCIE 2,000 GB/s.

**Which is cheaper to rent, the RTX PRO 6000 WS or the H100 PCIE?**

The RTX PRO 6000 WS: $1.040/hr on-demand versus $1.867/hr — 44% less. Interruptible rates are $0.520 (RTX PRO 6000 WS) and $0.933 (H100 PCIE). Per TFLOP-hour the better value is the RTX PRO 6000 WS.

**Which has more VRAM and what does that change?**

The RTX PRO 6000 WS has 96 GB versus 80 GB. In LLM terms that is roughly a 32B FP16 model (or ~141B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 74 × RTX PRO 6000 WS and 73 × H100 PCIE are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090 -->

---
title: "RTX PRO 6000 WS vs RTX 5090: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX PRO 6000 WS vs RTX 5090: 96 vs 32 GB VRAM, 505 vs 419 FP16 TFLOPS, $1.040 vs $0.439/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX PRO 6000 WS vs RTX 5090: specs, price per hour, which to rent

NVIDIA RTX PRO 6000 WS (96 GB, **$1.040** /hr) against NVIDIA RTX 5090 (32 GB, **$0.439** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX PRO 6000 WS Blackwell · 96 GB GDDR7 · 505 FP16 TFLOPS · $1.040/hr on-demand · $0.520/hr interruptible · 74 online](https://powergpu.ai/gpu/rtx-pro-6000-ws)
- [NVIDIA RTX 5090 Blackwell · 32 GB GDDR7 · 419 FP16 TFLOPS · $0.439/hr on-demand · $0.219/hr interruptible · 922 online](https://powergpu.ai/gpu/rtx-5090)

## RTX PRO 6000 WS vs RTX 5090 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX PRO 6000 WS relative to RTX 5090.

| Spec | RTX PRO 6000 WS | RTX 5090 | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2025) | Blackwell (2025) | — |
| VRAM | 96 GB GDDR7 | 32 GB GDDR7 | +200% |
| Memory bandwidth | 1,792 GB/s | 1,792 GB/s | same |
| FP16 tensor (dense) | 505 TFLOPS | 419 TFLOPS | +21% |
| FP32 | 125.0 TFLOPS | 104.8 TFLOPS | +19% |
| CUDA cores | 24,064 | 21,760 | +11% |
| TDP | 600 W | 575 W | +4% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 5.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 356 | 295 | +21% |
| Max GPUs per machine | 8× | 8× | — |

## RTX PRO 6000 WS vs RTX 5090 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX PRO 6000 WS | RTX 5090 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.040 | $0.439 | RTX 5090 (−58%) |
| Interruptible, per GPU-hour | $0.520 | $0.219 | RTX 5090 |
| Reserved (3 mo), per GPU-hour | $0.676 | $0.285 | RTX 5090 |
| On-demand, per month | $759 | $320 | RTX 5090 |
| Market median (reference) | $1.49 | $0.63 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.06 | $1.05 | RTX 5090 (better value) |
| $ per GB of VRAM per hour | $0.0108 | $0.0137 | RTX PRO 6000 WS (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 96 GB vs 32 GB

| Workload | RTX PRO 6000 WS | RTX 5090 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~13B |
| Largest LLM at 4-bit, one card | ~141B | ~49B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Same Blackwell generation: the PRO 6000 triples VRAM (96 vs 32 GB) and adds ECC and studio drivers; the RTX 5090 is much cheaper per hour with similar per-core throughput. Rent the 5090 unless the model or scene simply does not fit.

- **Cheaper per hour:** RTX 5090 ($0.439 vs $1.040, −58%).
- **More VRAM:** RTX PRO 6000 WS (96 GB vs 32 GB).
- **More FP16 throughput:** RTX PRO 6000 WS (about 1.2×).
- **Best value per TFLOP-hour:** RTX 5090.
- **Best value per GB of VRAM:** RTX PRO 6000 WS.
- **Multi-GPU:** RTX PRO 6000 WS over PCIe · RTX 5090 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 5090 vs L40S $0.439 vs $0.514 per hour](https://powergpu.ai/compare/rtx-5090-vs-l40s)
- [RTX 5090 vs A100 SXM4 $0.439 vs $0.560 per hour](https://powergpu.ai/compare/rtx-5090-vs-a100-sxm4)
- [RTX PRO 6000 WS vs RTX 6000Ada $1.040 vs $0.467 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada)
- [RTX PRO 6000 WS vs H100 PCIE $1.040 vs $1.867 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-h100-pcie)

## RTX PRO 6000 WS vs RTX 5090: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX PRO 6000 WS faster than the RTX 5090?**

On dense FP16 tensor throughput the RTX PRO 6000 WS leads by about 1.2× (505 vs 419 TFLOPS). Memory bandwidth matters as much for inference: RTX PRO 6000 WS 1,792 GB/s vs RTX 5090 1,792 GB/s.

**Which is cheaper to rent, the RTX PRO 6000 WS or the RTX 5090?**

The RTX 5090: $0.439/hr on-demand versus $1.040/hr — 58% less. Interruptible rates are $0.520 (RTX PRO 6000 WS) and $0.219 (RTX 5090). Per TFLOP-hour the better value is the RTX 5090.

**Which has more VRAM and what does that change?**

The RTX PRO 6000 WS has 96 GB versus 32 GB. In LLM terms that is roughly a 32B FP16 model (or ~141B in 4-bit) on one card against 13B FP16 (~49B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 74 × RTX PRO 6000 WS and 922 × RTX 5090 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-5090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000 -->

---
title: "RTX 6000Ada vs RTX A6000: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 6000Ada vs RTX A6000: 48 vs 48 GB VRAM, 364 vs 155 FP16 TFLOPS, $0.467 vs $0.281/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 6000Ada vs RTX A6000: specs, price per hour, which to rent

NVIDIA RTX 6000Ada (48 GB, **$0.467** /hr) against NVIDIA RTX A6000 (48 GB, **$0.281** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 6000Ada Ada Lovelace · 48 GB GDDR6 · 364 FP16 TFLOPS · $0.467/hr on-demand · $0.233/hr interruptible · 34 online](https://powergpu.ai/gpu/rtx-6000ada)
- [NVIDIA RTX A6000 Ampere · 48 GB GDDR6 · 155 FP16 TFLOPS · $0.281/hr on-demand · $0.140/hr interruptible · 12 online](https://powergpu.ai/gpu/rtx-a6000)

## RTX 6000Ada vs RTX A6000 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 6000Ada relative to RTX A6000.

| Spec | RTX 6000Ada | RTX A6000 | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2022) | Ampere (2020) | — |
| VRAM | 48 GB GDDR6 | 48 GB GDDR6 | same |
| Memory bandwidth | 960 GB/s | 768 GB/s | +25% |
| FP16 tensor (dense) | 364 TFLOPS | 155 TFLOPS | +135% |
| FP32 | 91.1 TFLOPS | 38.7 TFLOPS | +135% |
| CUDA cores | 18,176 | 10,752 | +69% |
| TDP | 300 W | 300 W | same |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 256 | 109 | +135% |
| Max GPUs per machine | 8× | 4× | — |

## RTX 6000Ada vs RTX A6000 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 6000Ada | RTX A6000 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.467 | $0.281 | RTX A6000 (−40%) |
| Interruptible, per GPU-hour | $0.233 | $0.140 | RTX A6000 |
| Reserved (3 mo), per GPU-hour | $0.303 | $0.182 | RTX A6000 |
| On-demand, per month | $341 | $205 | RTX A6000 |
| Market median (reference) | $0.67 | $0.40 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.28 | $1.81 | RTX 6000Ada (better value) |
| $ per GB of VRAM per hour | $0.0097 | $0.0059 | RTX A6000 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 48 GB vs 48 GB

| Workload | RTX 6000Ada | RTX A6000 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~14B | ~14B |
| Largest LLM at 4-bit, one card | ~72B | ~72B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Both are 48 GB workstation cards; the RTX 6000 Ada has about 2.3× the FP16 tensor throughput, 960 GB/s and newer drivers; the A6000 is the cheapest 48 GB rental on the sheet. Ada for speed, A6000 for capacity on a budget.

- **Cheaper per hour:** RTX A6000 ($0.281 vs $0.467, −40%).
- **More VRAM:** RTX 6000Ada (48 GB vs 48 GB).
- **More FP16 throughput:** RTX 6000Ada (about 2.3×).
- **Best value per TFLOP-hour:** RTX 6000Ada.
- **Best value per GB of VRAM:** RTX A6000.
- **Multi-GPU:** RTX 6000Ada over PCIe · RTX A6000 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX PRO 6000 WS vs RTX 6000Ada $1.040 vs $0.467 per hour](https://powergpu.ai/compare/rtx-pro-6000-ws-vs-rtx-6000ada)
- [RTX A6000 vs RTX 4090 $0.281 vs $0.327 per hour](https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090)

## RTX 6000Ada vs RTX A6000: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 6000Ada faster than the RTX A6000?**

On dense FP16 tensor throughput the RTX 6000Ada leads by about 2.3× (364 vs 155 TFLOPS). Memory bandwidth matters as much for inference: RTX 6000Ada 960 GB/s vs RTX A6000 768 GB/s.

**Which is cheaper to rent, the RTX 6000Ada or the RTX A6000?**

The RTX A6000: $0.281/hr on-demand versus $0.467/hr — 40% less. Interruptible rates are $0.233 (RTX 6000Ada) and $0.140 (RTX A6000). Per TFLOP-hour the better value is the RTX 6000Ada.

**Which has more VRAM and what does that change?**

The RTX 6000Ada has 48 GB versus 48 GB. In LLM terms that is roughly a 14B FP16 model (or ~72B in 4-bit) on one card against 14B FP16 (~72B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 34 × RTX 6000Ada and 12 × RTX A6000 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090 -->

---
title: "RTX A6000 vs RTX 4090: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX A6000 vs RTX 4090: 48 vs 24 GB VRAM, 155 vs 330 FP16 TFLOPS, $0.281 vs $0.327/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX A6000 vs RTX 4090: specs, price per hour, which to rent

NVIDIA RTX A6000 (48 GB, **$0.281** /hr) against NVIDIA RTX 4090 (24 GB, **$0.327** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX A6000 Ampere · 48 GB GDDR6 · 155 FP16 TFLOPS · $0.281/hr on-demand · $0.140/hr interruptible · 12 online](https://powergpu.ai/gpu/rtx-a6000)
- [NVIDIA RTX 4090 Ada Lovelace · 24 GB GDDR6X · 330 FP16 TFLOPS · $0.327/hr on-demand · $0.163/hr interruptible · 461 online](https://powergpu.ai/gpu/rtx-4090)

## RTX A6000 vs RTX 4090 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX A6000 relative to RTX 4090.

| Spec | RTX A6000 | RTX 4090 | Difference |
| --- | --- | --- | --- |
| Architecture | Ampere (2020) | Ada Lovelace (2022) | — |
| VRAM | 48 GB GDDR6 | 24 GB GDDR6X | +100% |
| Memory bandwidth | 768 GB/s | 1,008 GB/s | −24% |
| FP16 tensor (dense) | 155 TFLOPS | 330 TFLOPS | −53% |
| FP32 | 38.7 TFLOPS | 82.6 TFLOPS | −53% |
| CUDA cores | 10,752 | 16,384 | −34% |
| TDP | 300 W | 450 W | −33% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 109 | 232 | −53% |
| Max GPUs per machine | 4× | 8× | — |

## RTX A6000 vs RTX 4090 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX A6000 | RTX 4090 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.281 | $0.327 | RTX A6000 (−14%) |
| Interruptible, per GPU-hour | $0.140 | $0.163 | RTX A6000 |
| Reserved (3 mo), per GPU-hour | $0.182 | $0.212 | RTX A6000 |
| On-demand, per month | $205 | $239 | RTX A6000 |
| Market median (reference) | $0.40 | $0.47 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.81 | $0.99 | RTX 4090 (better value) |
| $ per GB of VRAM per hour | $0.0059 | $0.0136 | RTX A6000 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 48 GB vs 24 GB

| Workload | RTX A6000 | RTX 4090 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~14B | ~8B |
| Largest LLM at 4-bit, one card | ~72B | ~32B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The A6000 has 48 GB with ECC at a modest rate; the 4090 has 24 GB but about twice the tensor throughput and costs less. VRAM-bound jobs (32B 4-bit, heavy scenes) go A6000; everything that fits in 24 GB runs cheaper on the 4090.

- **Cheaper per hour:** RTX A6000 ($0.281 vs $0.327, −14%).
- **More VRAM:** RTX A6000 (48 GB vs 24 GB).
- **More FP16 throughput:** RTX 4090 (about 2.1×).
- **Best value per TFLOP-hour:** RTX 4090.
- **Best value per GB of VRAM:** RTX A6000.
- **Multi-GPU:** RTX A6000 over PCIe · RTX 4090 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 4090 vs RTX 3090 $0.327 vs $0.108 per hour](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [RTX 6000Ada vs RTX A6000 $0.467 vs $0.281 per hour](https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000)
- [RTX 5080 vs RTX 4090 $0.186 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5080-vs-rtx-4090)

## RTX A6000 vs RTX 4090: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 4090 faster than the RTX A6000?**

On dense FP16 tensor throughput the RTX 4090 leads by about 2.1× (330 vs 155 TFLOPS). Memory bandwidth matters as much for inference: RTX A6000 768 GB/s vs RTX 4090 1,008 GB/s.

**Which is cheaper to rent, the RTX A6000 or the RTX 4090?**

The RTX A6000: $0.281/hr on-demand versus $0.327/hr — 14% less. Interruptible rates are $0.140 (RTX A6000) and $0.163 (RTX 4090). Per TFLOP-hour the better value is the RTX 4090.

**Which has more VRAM and what does that change?**

The RTX A6000 has 48 GB versus 24 GB. In LLM terms that is roughly a 14B FP16 model (or ~72B in 4-bit) on one card against 8B FP16 (~32B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 12 × RTX A6000 and 461 × RTX 4090 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/l4-vs-tesla-t4 -->

---
title: "L4 vs Tesla T4: Specs & Price per Hour (2026) | PowerGPU"
description: "L4 vs Tesla T4: 24 vs 16 GB VRAM, 121 vs 65 FP16 TFLOPS, $0.225 vs $0.103/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/l4-vs-tesla-t4
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# L4 vs Tesla T4: specs, price per hour, which to rent

NVIDIA L4 (24 GB, **$0.225** /hr) against NVIDIA Tesla T4 (16 GB, **$0.103** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA L4 Ada Lovelace · 24 GB GDDR6 · 121 FP16 TFLOPS · $0.225/hr on-demand · $0.112/hr interruptible · 28 online](https://powergpu.ai/gpu/l4)
- [NVIDIA Tesla T4 Turing · 16 GB GDDR6 · 65 FP16 TFLOPS · $0.103/hr on-demand · $0.051/hr interruptible · 34 online](https://powergpu.ai/gpu/tesla-t4)

## L4 vs Tesla T4 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is L4 relative to Tesla T4.

| Spec | L4 | Tesla T4 | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2023) | Turing (2018) | — |
| VRAM | 24 GB GDDR6 | 16 GB GDDR6 | +50% |
| Memory bandwidth | 300 GB/s | 320 GB/s | −6% |
| FP16 tensor (dense) | 121 TFLOPS | 65 TFLOPS | +86% |
| FP32 | 30.3 TFLOPS | 8.1 TFLOPS | +274% |
| CUDA cores | 7,680 | 2,560 | +200% |
| TDP | 72 W | 70 W | +3% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 3.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 85 | 46 | +85% |
| Max GPUs per machine | 4× | 8× | — |

## L4 vs Tesla T4 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | L4 | Tesla T4 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.225 | $0.103 | Tesla T4 (−54%) |
| Interruptible, per GPU-hour | $0.112 | $0.051 | Tesla T4 |
| Reserved (3 mo), per GPU-hour | $0.146 | $0.066 | Tesla T4 |
| On-demand, per month | $164 | $75 | Tesla T4 |
| Market median (reference) | $0.32 | $0.15 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.86 | $1.58 | Tesla T4 (better value) |
| $ per GB of VRAM per hour | $0.0094 | $0.0064 | Tesla T4 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 24 GB vs 16 GB

| Workload | L4 | Tesla T4 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~8B | ~3B |
| Largest LLM at 4-bit, one card | ~32B | ~24B |
| Flux dev (FP8, ~17 GB) | fits | tight |
| Wan 2.x 14B video (offloaded) | yes | no |
| 70B 4-bit LLM on one card | no | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The L4 is the T4's Ada successor: 24 vs 16 GB, about twice the tensor throughput, AV1 encode, still 72 W and single-slot. The T4 is cheaper for tiny models; the L4 is the better rental for anything that touches video or exceeds 16 GB.

- **Cheaper per hour:** Tesla T4 ($0.103 vs $0.225, −54%).
- **More VRAM:** L4 (24 GB vs 16 GB).
- **More FP16 throughput:** L4 (about 1.9×).
- **Best value per TFLOP-hour:** Tesla T4.
- **Best value per GB of VRAM:** Tesla T4.
- **Multi-GPU:** L4 over PCIe · Tesla T4 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [L4 vs A10 $0.225 vs $0.168 per hour](https://powergpu.ai/compare/l4-vs-a10)

## L4 vs Tesla T4: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the L4 faster than the Tesla T4?**

On dense FP16 tensor throughput the L4 leads by about 1.9× (121 vs 65 TFLOPS). Memory bandwidth matters as much for inference: L4 300 GB/s vs Tesla T4 320 GB/s.

**Which is cheaper to rent, the L4 or the Tesla T4?**

The Tesla T4: $0.103/hr on-demand versus $0.225/hr — 54% less. Interruptible rates are $0.112 (L4) and $0.051 (Tesla T4). Per TFLOP-hour the better value is the Tesla T4.

**Which has more VRAM and what does that change?**

The L4 has 24 GB versus 16 GB. In LLM terms that is roughly a 8B FP16 model (or ~32B in 4-bit) on one card against 3B FP16 (~24B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 28 × L4 and 34 × Tesla T4 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/l4-vs-tesla-t4 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/l4-vs-a10 -->

---
title: "L4 vs A10: Specs & Price per Hour (2026) | PowerGPU"
description: "L4 vs A10: 24 vs 24 GB VRAM, 121 vs 125 FP16 TFLOPS, $0.225 vs $0.168/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/l4-vs-a10
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# L4 vs A10: specs, price per hour, which to rent

NVIDIA L4 (24 GB, **$0.225** /hr) against NVIDIA A10 (24 GB, **$0.168** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA L4 Ada Lovelace · 24 GB GDDR6 · 121 FP16 TFLOPS · $0.225/hr on-demand · $0.112/hr interruptible · 28 online](https://powergpu.ai/gpu/l4)
- [NVIDIA A10 Ampere · 24 GB GDDR6 · 125 FP16 TFLOPS · $0.168/hr on-demand · $0.084/hr interruptible · 8 online](https://powergpu.ai/gpu/a10)

## L4 vs A10 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is L4 relative to A10.

| Spec | L4 | A10 | Difference |
| --- | --- | --- | --- |
| Architecture | Ada Lovelace (2023) | Ampere (2021) | — |
| VRAM | 24 GB GDDR6 | 24 GB GDDR6 | same |
| Memory bandwidth | 300 GB/s | 600 GB/s | −50% |
| FP16 tensor (dense) | 121 TFLOPS | 125 TFLOPS | −3% |
| FP32 | 30.3 TFLOPS | 31.2 TFLOPS | −3% |
| CUDA cores | 7,680 | 9,216 | −17% |
| TDP | 72 W | 150 W | −52% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 85 | 88 | −3% |
| Max GPUs per machine | 4× | 8× | — |

## L4 vs A10 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | L4 | A10 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.225 | $0.168 | A10 (−25%) |
| Interruptible, per GPU-hour | $0.112 | $0.084 | A10 |
| Reserved (3 mo), per GPU-hour | $0.146 | $0.109 | A10 |
| On-demand, per month | $164 | $123 | A10 |
| Market median (reference) | $0.32 | $0.24 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.86 | $1.34 | A10 (better value) |
| $ per GB of VRAM per hour | $0.0094 | $0.0070 | A10 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 24 GB vs 24 GB

| Workload | L4 | A10 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~8B | ~8B |
| Largest LLM at 4-bit, one card | ~32B | ~32B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | no | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Similar 24 GB envelopes: the A10 (Ampere, 150 W) has more CUDA cores and bandwidth; the L4 (Ada, 72 W) has newer tensor cores, FP8 and better video engines. A10 for raw throughput, L4 for efficiency and transcoding.

- **Cheaper per hour:** A10 ($0.168 vs $0.225, −25%).
- **More VRAM:** L4 (24 GB vs 24 GB).
- **More FP16 throughput:** A10 (about 1.0×).
- **Best value per TFLOP-hour:** A10.
- **Best value per GB of VRAM:** A10.
- **Multi-GPU:** L4 over PCIe · A10 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [L4 vs Tesla T4 $0.225 vs $0.103 per hour](https://powergpu.ai/compare/l4-vs-tesla-t4)

## L4 vs A10: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the A10 faster than the L4?**

On dense FP16 tensor throughput the A10 leads by about 1.0× (125 vs 121 TFLOPS). Memory bandwidth matters as much for inference: L4 300 GB/s vs A10 600 GB/s.

**Which is cheaper to rent, the L4 or the A10?**

The A10: $0.168/hr on-demand versus $0.225/hr — 25% less. Interruptible rates are $0.112 (L4) and $0.084 (A10). Per TFLOP-hour the better value is the A10.

**Which has more VRAM and what does that change?**

The L4 has 24 GB versus 24 GB. In LLM terms that is roughly a 8B FP16 model (or ~32B in 4-bit) on one card against 8B FP16 (~32B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 28 × L4 and 8 × A10 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/l4-vs-a10 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/tesla-v100-vs-rtx-3090 -->

---
title: "Tesla V100 vs RTX 3090: Specs & Price per Hour (2026) | PowerGPU"
description: "Tesla V100 vs RTX 3090: 16 vs 24 GB VRAM, 125 vs 142 FP16 TFLOPS, $0.130 vs $0.108/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/tesla-v100-vs-rtx-3090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# Tesla V100 vs RTX 3090: specs, price per hour, which to rent

NVIDIA Tesla V100 (16 GB, **$0.130** /hr) against NVIDIA RTX 3090 (24 GB, **$0.108** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA Tesla V100 Volta · 16 GB HBM2 · 125 FP16 TFLOPS · $0.130/hr on-demand · $0.065/hr interruptible · 96 online](https://powergpu.ai/gpu/tesla-v100)
- [NVIDIA RTX 3090 Ampere · 24 GB GDDR6X · 142 FP16 TFLOPS · $0.108/hr on-demand · $0.054/hr interruptible · 345 online](https://powergpu.ai/gpu/rtx-3090)

## Tesla V100 vs RTX 3090 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is Tesla V100 relative to RTX 3090.

| Spec | Tesla V100 | RTX 3090 | Difference |
| --- | --- | --- | --- |
| Architecture | Volta (2017) | Ampere (2020) | — |
| VRAM | 16 GB HBM2 | 24 GB GDDR6X | −33% |
| Memory bandwidth | 900 GB/s | 936 GB/s | −4% |
| FP16 tensor (dense) | 125 TFLOPS | 142 TFLOPS | −12% |
| FP32 | 14.1 TFLOPS | 35.6 TFLOPS | −60% |
| CUDA cores | 5,120 | 10,496 | −51% |
| TDP | 250 W | 350 W | −29% |
| PCIe · NVLink | Gen 3.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 88 | 100 | −12% |
| Max GPUs per machine | 8× | 8× | — |

## Tesla V100 vs RTX 3090 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | Tesla V100 | RTX 3090 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.130 | $0.108 | RTX 3090 (−17%) |
| Interruptible, per GPU-hour | $0.065 | $0.054 | RTX 3090 |
| Reserved (3 mo), per GPU-hour | $0.084 | $0.070 | RTX 3090 |
| On-demand, per month | $95 | $79 | RTX 3090 |
| Market median (reference) | $0.19 | $0.15 | — |
| $ per 1,000 FP16 TFLOP-hours | $1.04 | $0.76 | RTX 3090 (better value) |
| $ per GB of VRAM per hour | $0.0081 | $0.0045 | RTX 3090 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 16 GB vs 24 GB

| Workload | Tesla V100 | RTX 3090 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~3B | ~8B |
| Largest LLM at 4-bit, one card | ~24B | ~32B |
| Flux dev (FP8, ~17 GB) | tight | fits |
| Wan 2.x 14B video (offloaded) | no | yes |
| 70B 4-bit LLM on one card | no | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The V100 offers 16 GB HBM2, real FP64 and datacenter design; the RTX 3090 has 24 GB and roughly the same FP16 tensor throughput at a lower rate. Rent the V100 for double precision, the 3090 for ML with more memory.

- **Cheaper per hour:** RTX 3090 ($0.108 vs $0.130, −17%).
- **More VRAM:** RTX 3090 (24 GB vs 16 GB).
- **More FP16 throughput:** RTX 3090 (about 1.1×).
- **Best value per TFLOP-hour:** RTX 3090.
- **Best value per GB of VRAM:** RTX 3090.
- **Multi-GPU:** Tesla V100 over PCIe · RTX 3090 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 4090 vs RTX 3090 $0.327 vs $0.108 per hour](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [RTX 3090 vs RTX 3060 $0.108 vs $0.042 per hour](https://powergpu.ai/compare/rtx-3090-vs-rtx-3060)

## Tesla V100 vs RTX 3090: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 3090 faster than the Tesla V100?**

On dense FP16 tensor throughput the RTX 3090 leads by about 1.1× (142 vs 125 TFLOPS). Memory bandwidth matters as much for inference: Tesla V100 900 GB/s vs RTX 3090 936 GB/s.

**Which is cheaper to rent, the Tesla V100 or the RTX 3090?**

The RTX 3090: $0.108/hr on-demand versus $0.130/hr — 17% less. Interruptible rates are $0.065 (Tesla V100) and $0.054 (RTX 3090). Per TFLOP-hour the better value is the RTX 3090.

**Which has more VRAM and what does that change?**

The RTX 3090 has 24 GB versus 16 GB. In LLM terms that is roughly a 8B FP16 model (or ~32B in 4-bit) on one card against 3B FP16 (~24B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 96 × Tesla V100 and 345 × RTX 3090 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/tesla-v100-vs-rtx-3090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-3090-vs-rtx-3060 -->

---
title: "RTX 3090 vs RTX 3060: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 3090 vs RTX 3060: 24 vs 12 GB VRAM, 142 vs 51 FP16 TFLOPS, $0.108 vs $0.042/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-3090-vs-rtx-3060
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 3090 vs RTX 3060: specs, price per hour, which to rent

NVIDIA RTX 3090 (24 GB, **$0.108** /hr) against NVIDIA RTX 3060 (12 GB, **$0.042** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 3090 Ampere · 24 GB GDDR6X · 142 FP16 TFLOPS · $0.108/hr on-demand · $0.054/hr interruptible · 345 online](https://powergpu.ai/gpu/rtx-3090)
- [NVIDIA RTX 3060 Ampere · 12 GB GDDR6 · 51 FP16 TFLOPS · $0.042/hr on-demand · $0.021/hr interruptible · 279 online](https://powergpu.ai/gpu/rtx-3060)

## RTX 3090 vs RTX 3060 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 3090 relative to RTX 3060.

| Spec | RTX 3090 | RTX 3060 | Difference |
| --- | --- | --- | --- |
| Architecture | Ampere (2020) | Ampere (2021) | — |
| VRAM | 24 GB GDDR6X | 12 GB GDDR6 | +100% |
| Memory bandwidth | 936 GB/s | 360 GB/s | +160% |
| FP16 tensor (dense) | 142 TFLOPS | 51 TFLOPS | +178% |
| FP32 | 35.6 TFLOPS | 12.7 TFLOPS | +180% |
| CUDA cores | 10,496 | 3,584 | +193% |
| TDP | 350 W | 170 W | +106% |
| PCIe · NVLink | Gen 4.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 100 | 36 | +178% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 3090 vs RTX 3060 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 3090 | RTX 3060 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.108 | $0.042 | RTX 3060 (−61%) |
| Interruptible, per GPU-hour | $0.054 | $0.021 | RTX 3060 |
| Reserved (3 mo), per GPU-hour | $0.070 | $0.027 | RTX 3060 |
| On-demand, per month | $79 | $31 | RTX 3060 |
| Market median (reference) | $0.15 | $0.06 | — |
| $ per 1,000 FP16 TFLOP-hours | $0.76 | $0.82 | RTX 3090 (better value) |
| $ per GB of VRAM per hour | $0.0045 | $0.0035 | RTX 3060 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 24 GB vs 12 GB

| Workload | RTX 3090 | RTX 3060 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~8B | ~3B |
| Largest LLM at 4-bit, one card | ~32B | ~14B |
| Flux dev (FP8, ~17 GB) | fits | no |
| Wan 2.x 14B video (offloaded) | yes | no |
| 70B 4-bit LLM on one card | no | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Both Ampere: the 3090 has 24 GB and about 2.8× the throughput of the 12 GB 3060. The 3060 is the cheapest way to run 7B–8B 4-bit models and SD 1.5; the 3090 opens 13B FP16, SDXL batches and LoRA training.

- **Cheaper per hour:** RTX 3060 ($0.042 vs $0.108, −61%).
- **More VRAM:** RTX 3090 (24 GB vs 12 GB).
- **More FP16 throughput:** RTX 3090 (about 2.8×).
- **Best value per TFLOP-hour:** RTX 3090.
- **Best value per GB of VRAM:** RTX 3060.
- **Multi-GPU:** RTX 3090 over PCIe · RTX 3060 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 4090 vs RTX 3090 $0.327 vs $0.108 per hour](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [Tesla V100 vs RTX 3090 $0.130 vs $0.108 per hour](https://powergpu.ai/compare/tesla-v100-vs-rtx-3090)

## RTX 3090 vs RTX 3060: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 3090 faster than the RTX 3060?**

On dense FP16 tensor throughput the RTX 3090 leads by about 2.8× (142 vs 51 TFLOPS). Memory bandwidth matters as much for inference: RTX 3090 936 GB/s vs RTX 3060 360 GB/s.

**Which is cheaper to rent, the RTX 3090 or the RTX 3060?**

The RTX 3060: $0.042/hr on-demand versus $0.108/hr — 61% less. Interruptible rates are $0.054 (RTX 3090) and $0.021 (RTX 3060). Per TFLOP-hour the better value is the RTX 3090.

**Which has more VRAM and what does that change?**

The RTX 3090 has 24 GB versus 12 GB. In LLM terms that is roughly a 8B FP16 model (or ~32B in 4-bit) on one card against 3B FP16 (~14B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 345 × RTX 3090 and 279 × RTX 3060 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-3090-vs-rtx-3060 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/rtx-5080-vs-rtx-4090 -->

---
title: "RTX 5080 vs RTX 4090: Specs & Price per Hour (2026) | PowerGPU"
description: "RTX 5080 vs RTX 4090: 16 vs 24 GB VRAM, 225 vs 330 FP16 TFLOPS, $0.186 vs $0.327/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/rtx-5080-vs-rtx-4090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# RTX 5080 vs RTX 4090: specs, price per hour, which to rent

NVIDIA RTX 5080 (16 GB, **$0.186** /hr) against NVIDIA RTX 4090 (24 GB, **$0.327** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA RTX 5080 Blackwell · 16 GB GDDR7 · 225 FP16 TFLOPS · $0.186/hr on-demand · $0.093/hr interruptible · 97 online](https://powergpu.ai/gpu/rtx-5080)
- [NVIDIA RTX 4090 Ada Lovelace · 24 GB GDDR6X · 330 FP16 TFLOPS · $0.327/hr on-demand · $0.163/hr interruptible · 461 online](https://powergpu.ai/gpu/rtx-4090)

## RTX 5080 vs RTX 4090 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is RTX 5080 relative to RTX 4090.

| Spec | RTX 5080 | RTX 4090 | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2025) | Ada Lovelace (2022) | — |
| VRAM | 16 GB GDDR7 | 24 GB GDDR6X | −33% |
| Memory bandwidth | 960 GB/s | 1,008 GB/s | −5% |
| FP16 tensor (dense) | 225 TFLOPS | 330 TFLOPS | −32% |
| FP32 | 56.3 TFLOPS | 82.6 TFLOPS | −32% |
| CUDA cores | 10,752 | 16,384 | −34% |
| TDP | 360 W | 450 W | −20% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 4.0 · no NVLink | — |
| PowerScore (RTX 3090 = 100) | 158 | 232 | −32% |
| Max GPUs per machine | 8× | 8× | — |

## RTX 5080 vs RTX 4090 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | RTX 5080 | RTX 4090 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $0.186 | $0.327 | RTX 5080 (−43%) |
| Interruptible, per GPU-hour | $0.093 | $0.163 | RTX 5080 |
| Reserved (3 mo), per GPU-hour | $0.120 | $0.212 | RTX 5080 |
| On-demand, per month | $136 | $239 | RTX 5080 |
| Market median (reference) | $0.27 | $0.47 | — |
| $ per 1,000 FP16 TFLOP-hours | $0.83 | $0.99 | RTX 5080 (better value) |
| $ per GB of VRAM per hour | $0.0116 | $0.0136 | RTX 5080 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 16 GB vs 24 GB

| Workload | RTX 5080 | RTX 4090 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~3B | ~8B |
| Largest LLM at 4-bit, one card | ~24B | ~32B |
| Flux dev (FP8, ~17 GB) | tight | fits |
| Wan 2.x 14B video (offloaded) | no | yes |
| 70B 4-bit LLM on one card | no | no |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The RTX 5080 has 16 GB GDDR7 and FP4 tensor cores; the RTX 4090 has 24 GB GDDR6X and more CUDA cores. The 4090's extra 8 GB decides most AI jobs; the 5080 is the cheaper pick for workloads that fit in 16 GB.

- **Cheaper per hour:** RTX 5080 ($0.186 vs $0.327, −43%).
- **More VRAM:** RTX 4090 (24 GB vs 16 GB).
- **More FP16 throughput:** RTX 4090 (about 1.5×).
- **Best value per TFLOP-hour:** RTX 5080.
- **Best value per GB of VRAM:** RTX 5080.
- **Multi-GPU:** RTX 5080 over PCIe · RTX 4090 over PCIe.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [RTX 5090 vs RTX 4090 $0.439 vs $0.327 per hour](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090)
- [RTX 4090 vs RTX 3090 $0.327 vs $0.108 per hour](https://powergpu.ai/compare/rtx-4090-vs-rtx-3090)
- [RTX 4090 vs A100 PCIE $0.327 vs $0.374 per hour](https://powergpu.ai/compare/rtx-4090-vs-a100-pcie)
- [RTX 4090 vs L40S $0.327 vs $0.514 per hour](https://powergpu.ai/compare/rtx-4090-vs-l40s)
- [RTX A6000 vs RTX 4090 $0.281 vs $0.327 per hour](https://powergpu.ai/compare/rtx-a6000-vs-rtx-4090)

## RTX 5080 vs RTX 4090: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the RTX 4090 faster than the RTX 5080?**

On dense FP16 tensor throughput the RTX 4090 leads by about 1.5× (330 vs 225 TFLOPS). Memory bandwidth matters as much for inference: RTX 5080 960 GB/s vs RTX 4090 1,008 GB/s.

**Which is cheaper to rent, the RTX 5080 or the RTX 4090?**

The RTX 5080: $0.186/hr on-demand versus $0.327/hr — 43% less. Interruptible rates are $0.093 (RTX 5080) and $0.163 (RTX 4090). Per TFLOP-hour the better value is the RTX 5080.

**Which has more VRAM and what does that change?**

The RTX 4090 has 24 GB versus 16 GB. In LLM terms that is roughly a 8B FP16 model (or ~32B in 4-bit) on one card against 3B FP16 (~24B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 97 × RTX 5080 and 461 × RTX 4090 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/rtx-5080-vs-rtx-4090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h100-nvl-vs-h100-sxm -->

---
title: "H100 NVL vs H100 SXM: Specs & Price per Hour (2026) | PowerGPU"
description: "H100 NVL vs H100 SXM: 80 vs 80 GB VRAM, 835 vs 990 FP16 TFLOPS, $1.811 vs $1.428/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h100-nvl-vs-h100-sxm
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H100 NVL vs H100 SXM: specs, price per hour, which to rent

NVIDIA H100 NVL (80 GB, **$1.811** /hr) against NVIDIA H100 SXM (80 GB, **$1.428** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H100 NVL Hopper · 80 GB HBM3 · 835 FP16 TFLOPS · $1.811/hr on-demand · $0.905/hr interruptible · 7 online](https://powergpu.ai/gpu/h100-nvl)
- [NVIDIA H100 SXM Hopper · 80 GB HBM3 · 990 FP16 TFLOPS · $1.428/hr on-demand · $0.714/hr interruptible · 54 online](https://powergpu.ai/gpu/h100-sxm)

## H100 NVL vs H100 SXM specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H100 NVL relative to H100 SXM.

| Spec | H100 NVL | H100 SXM | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2023) | Hopper (2022) | — |
| VRAM | 80 GB HBM3 | 80 GB HBM3 | same |
| Memory bandwidth | 3,900 GB/s | 3,350 GB/s | +16% |
| FP16 tensor (dense) | 835 TFLOPS | 990 TFLOPS | −16% |
| FP32 | 60.0 TFLOPS | 67.0 TFLOPS | −10% |
| CUDA cores | 14,592 | 16,896 | −14% |
| TDP | 400 W | 700 W | −43% |
| PCIe · NVLink | Gen 5.0 · NVLink | Gen 5.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 588 | 697 | −16% |
| Max GPUs per machine | 4× | 8× | — |

## H100 NVL vs H100 SXM price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H100 NVL | H100 SXM | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $1.811 | $1.428 | H100 SXM (−21%) |
| Interruptible, per GPU-hour | $0.905 | $0.714 | H100 SXM |
| Reserved (3 mo), per GPU-hour | $1.177 | $0.928 | H100 SXM |
| On-demand, per month | $1,322 | $1,042 | H100 SXM |
| Market median (reference) | $2.59 | $2.04 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.17 | $1.44 | H100 SXM (better value) |
| $ per GB of VRAM per hour | $0.0226 | $0.0178 | H100 SXM (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 80 GB vs 80 GB

| Workload | H100 NVL | H100 SXM |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~32B | ~32B |
| Largest LLM at 4-bit, one card | ~123B | ~123B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The H100 NVL is a PCIe card bridged in pairs with higher clocks than the H100 PCIe; the H100 SXM has full 900 GB/s NVLink across 8 GPUs and 3.35 TB/s. NVL for inference on PCIe hosts, SXM for multi-GPU training.

- **Cheaper per hour:** H100 SXM ($1.428 vs $1.811, −21%).
- **More VRAM:** H100 NVL (80 GB vs 80 GB).
- **More FP16 throughput:** H100 SXM (about 1.2×).
- **Best value per TFLOP-hour:** H100 SXM.
- **Best value per GB of VRAM:** H100 SXM.
- **Multi-GPU:** H100 NVL with NVLink · H100 SXM with NVLink.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs A100 SXM4 $1.428 vs $0.560 per hour](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4)
- [H100 SXM vs H200 $1.428 vs $2.791 per hour](https://powergpu.ai/compare/h100-sxm-vs-h200)
- [H100 SXM vs B200 $1.428 vs $5.425 per hour](https://powergpu.ai/compare/h100-sxm-vs-b200)
- [H100 SXM vs H100 PCIE $1.428 vs $1.867 per hour](https://powergpu.ai/compare/h100-sxm-vs-h100-pcie)

## H100 NVL vs H100 SXM: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H100 SXM faster than the H100 NVL?**

On dense FP16 tensor throughput the H100 SXM leads by about 1.2× (990 vs 835 TFLOPS). Memory bandwidth matters as much for inference: H100 NVL 3,900 GB/s vs H100 SXM 3,350 GB/s.

**Which is cheaper to rent, the H100 NVL or the H100 SXM?**

The H100 SXM: $1.428/hr on-demand versus $1.811/hr — 21% less. Interruptible rates are $0.905 (H100 NVL) and $0.714 (H100 SXM). Per TFLOP-hour the better value is the H100 SXM.

**Which has more VRAM and what does that change?**

The H100 NVL has 80 GB versus 80 GB. In LLM terms that is roughly a 32B FP16 model (or ~123B in 4-bit) on one card against 32B FP16 (~123B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 7 × H100 NVL and 54 × H100 SXM are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h100-nvl-vs-h100-sxm · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/h200-vs-h200-nvl -->

---
title: "H200 vs H200 NVL: Specs & Price per Hour (2026) | PowerGPU"
description: "H200 vs H200 NVL: 141 vs 141 GB VRAM, 990 vs 835 FP16 TFLOPS, $2.791 vs $2.650/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/h200-vs-h200-nvl
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# H200 vs H200 NVL: specs, price per hour, which to rent

NVIDIA H200 (141 GB, **$2.791** /hr) against NVIDIA H200 NVL (141 GB, **$2.650** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA H200 Hopper · 141 GB HBM3e · 990 FP16 TFLOPS · $2.791/hr on-demand · $1.395/hr interruptible · 50 online](https://powergpu.ai/gpu/h200)
- [NVIDIA H200 NVL Hopper · 141 GB HBM3e · 835 FP16 TFLOPS · $2.650/hr on-demand · $1.325/hr interruptible · 27 online](https://powergpu.ai/gpu/h200-nvl)

## H200 vs H200 NVL specifications

Public NVIDIA figures (dense, non-sparsity). The last column is H200 relative to H200 NVL.

| Spec | H200 | H200 NVL | Difference |
| --- | --- | --- | --- |
| Architecture | Hopper (2023) | Hopper (2024) | — |
| VRAM | 141 GB HBM3e | 141 GB HBM3e | same |
| Memory bandwidth | 4,800 GB/s | 4,800 GB/s | same |
| FP16 tensor (dense) | 990 TFLOPS | 835 TFLOPS | +19% |
| FP32 | 67.0 TFLOPS | 60.0 TFLOPS | +12% |
| CUDA cores | 16,896 | 16,896 | same |
| TDP | 700 W | 600 W | +17% |
| PCIe · NVLink | Gen 5.0 · no NVLink | Gen 5.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 697 | 588 | +19% |
| Max GPUs per machine | 8× | 8× | — |

## H200 vs H200 NVL price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | H200 | H200 NVL | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $2.791 | $2.650 | H200 NVL (−5%) |
| Interruptible, per GPU-hour | $1.395 | $1.325 | H200 NVL |
| Reserved (3 mo), per GPU-hour | $1.814 | $1.722 | H200 NVL |
| On-demand, per month | $2,037 | $1,935 | H200 NVL |
| Market median (reference) | $3.99 | $3.79 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.82 | $3.17 | H200 (better value) |
| $ per GB of VRAM per hour | $0.0198 | $0.0188 | H200 NVL (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 141 GB vs 141 GB

| Workload | H200 | H200 NVL |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~49B | ~49B |
| Largest LLM at 4-bit, one card | ~141B | ~141B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

Same 141 GB of HBM3e: the SXM H200 offers 700 W, 8-way NVLink and slightly higher clocks; the H200 NVL is a PCIe card in bridged sets of two or four. NVL is cheaper per hour for inference; SXM for large training jobs.

- **Cheaper per hour:** H200 NVL ($2.650 vs $2.791, −5%).
- **More VRAM:** H200 (141 GB vs 141 GB).
- **More FP16 throughput:** H200 (about 1.2×).
- **Best value per TFLOP-hour:** H200.
- **Best value per GB of VRAM:** H200 NVL.
- **Multi-GPU:** H200 over PCIe · H200 NVL with NVLink.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H100 SXM vs H200 $1.428 vs $2.791 per hour](https://powergpu.ai/compare/h100-sxm-vs-h200)
- [H200 vs B200 $2.791 vs $5.425 per hour](https://powergpu.ai/compare/h200-vs-b200)

## H200 vs H200 NVL: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the H200 faster than the H200 NVL?**

On dense FP16 tensor throughput the H200 leads by about 1.2× (990 vs 835 TFLOPS). Memory bandwidth matters as much for inference: H200 4,800 GB/s vs H200 NVL 4,800 GB/s.

**Which is cheaper to rent, the H200 or the H200 NVL?**

The H200 NVL: $2.650/hr on-demand versus $2.791/hr — 5% less. Interruptible rates are $1.395 (H200) and $1.325 (H200 NVL). Per TFLOP-hour the better value is the H200.

**Which has more VRAM and what does that change?**

The H200 has 141 GB versus 141 GB. In LLM terms that is roughly a 49B FP16 model (or ~141B in 4-bit) on one card against 49B FP16 (~141B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 50 × H200 and 27 × H200 NVL are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/h200-vs-h200-nvl · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/compare/b200-vs-b300 -->

---
title: "B200 vs B300: Specs & Price per Hour (2026) | PowerGPU"
description: "B200 vs B300: 192 vs 288 GB VRAM, 2,250 vs 2,800 FP16 TFLOPS, $5.425 vs $6.737/hr on PowerGPU. Specs side by side, live rental prices, which to rent."
url: https://powergpu.ai/compare/b200-vs-b300
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compare · prices checked 2026-09-14

# B200 vs B300: specs, price per hour, which to rent

NVIDIA B200 (192 GB, **$5.425** /hr) against NVIDIA B300 (288 GB, **$6.737** /hr): public specs side by side, live fixed prices from our sheet, what fits in each card's VRAM, and a verdict written for real workloads — both are rentable right now.

- [NVIDIA B200 Blackwell · 192 GB HBM3e · 2,250 FP16 TFLOPS · $5.425/hr on-demand · $2.712/hr interruptible · 32 online](https://powergpu.ai/gpu/b200)
- [NVIDIA B300 Blackwell · 288 GB HBM3e · 2,800 FP16 TFLOPS · $6.737/hr on-demand · $3.368/hr interruptible · 34 online](https://powergpu.ai/gpu/b300)

## B200 vs B300 specifications

Public NVIDIA figures (dense, non-sparsity). The last column is B200 relative to B300.

| Spec | B200 | B300 | Difference |
| --- | --- | --- | --- |
| Architecture | Blackwell (2024) | Blackwell (2025) | — |
| VRAM | 192 GB HBM3e | 288 GB HBM3e | −33% |
| Memory bandwidth | 8,000 GB/s | 8,000 GB/s | same |
| FP16 tensor (dense) | 2,250 TFLOPS | 2,800 TFLOPS | −20% |
| FP32 | — | — | — |
| CUDA cores | — | — | — |
| TDP | 1000 W | 1100 W | −9% |
| PCIe · NVLink | Gen 5.0 · NVLink | Gen 5.0 · NVLink | — |
| PowerScore (RTX 3090 = 100) | 1585 | 1972 | −20% |
| Max GPUs per machine | 8× | 2× | — |

## B200 vs B300 price per hour

Fixed rates from our sheet — every on-demand price is the marketplace median × 0.70, rounded down. Monthly = 730 hours.

| Rate | B200 | B300 | Cheaper |
| --- | --- | --- | --- |
| On-demand, per GPU-hour | $5.425 | $6.737 | B200 (−19%) |
| Interruptible, per GPU-hour | $2.712 | $3.368 | B200 |
| Reserved (3 mo), per GPU-hour | $3.526 | $4.379 | B200 |
| On-demand, per month | $3,960 | $4,918 | B200 |
| Market median (reference) | $7.75 | $9.63 | — |
| $ per 1,000 FP16 TFLOP-hours | $2.41 | $2.41 | B300 (better value) |
| $ per GB of VRAM per hour | $0.0283 | $0.0234 | B300 (better value) |

Try a full month with storage and bandwidth in the [GPU cost calculator](https://powergpu.ai/calculator).

## What fits in VRAM: 192 GB vs 288 GB

| Workload | B200 | B300 |
| --- | --- | --- |
| Largest LLM in FP16, one card | ~72B | ~105B |
| Largest LLM at 4-bit, one card | ~235B | ~405B |
| Flux dev (FP8, ~17 GB) | fits | fits |
| Wan 2.x 14B video (offloaded) | yes | yes |
| 70B 4-bit LLM on one card | yes | yes |

Rules of thumb: ~2.4 GB per billion parameters in FP16 all-in, ~0.62 GB in 4-bit. Full tables in the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements).

## Verdict: which should you rent?

The B300 (Blackwell Ultra) raises memory to 288 GB of HBM3e and boosts FP4 inference throughput; the B200 offers 192 GB at a lower hourly rate. Reserve the B300 for the largest inference deployments; the B200 for most Blackwell training.

- **Cheaper per hour:** B200 ($5.425 vs $6.737, −19%).
- **More VRAM:** B300 (288 GB vs 192 GB).
- **More FP16 throughput:** B300 (about 1.2×).
- **Best value per TFLOP-hour:** B300.
- **Best value per GB of VRAM:** B300.
- **Multi-GPU:** B200 with NVLink · B300 with NVLink.

## Related comparisons

[All comparisons](https://powergpu.ai/compare)

- [H200 vs B200 $2.791 vs $5.425 per hour](https://powergpu.ai/compare/h200-vs-b200)
- [H100 SXM vs B200 $1.428 vs $5.425 per hour](https://powergpu.ai/compare/h100-sxm-vs-b200)

## B200 vs B300: FAQ

Deeper reading: [H100 vs H200 vs B200](https://powergpu.ai/guides/h100-vs-h200-vs-b200), [RTX 4090 vs RTX 5090](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090), [how cloud GPU pricing works](https://powergpu.ai/guides/cloud-gpu-pricing-explained).

**Is the B300 faster than the B200?**

On dense FP16 tensor throughput the B300 leads by about 1.2× (2,800 vs 2,250 TFLOPS). Memory bandwidth matters as much for inference: B200 8,000 GB/s vs B300 8,000 GB/s.

**Which is cheaper to rent, the B200 or the B300?**

The B200: $5.425/hr on-demand versus $6.737/hr — 19% less. Interruptible rates are $2.712 (B200) and $3.368 (B300). Per TFLOP-hour the better value is the B300.

**Which has more VRAM and what does that change?**

The B300 has 288 GB versus 192 GB. In LLM terms that is roughly a 105B FP16 model (or ~405B in 4-bit) on one card against 72B FP16 (~235B 4-bit). If the model does not fit, speed is irrelevant.

**Can I rent both on PowerGPU right now?**

Yes — 32 × B200 and 34 × B300 are online as this page renders, deployable in about 30 seconds, billed per second, paid in crypto with no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/compare/b200-vs-b300 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates -->

---
title: "GPU Templates: ComfyUI, vLLM, PyTorch, Ollama & 33 More | PowerGPU"
description: "37 one-click GPU templates with official images: PyTorch, vLLM, Ollama, ComfyUI, Axolotl, full VMs — or any Docker image. Running in ~30 s on fixed-price GPUs."
url: https://powergpu.ai/templates
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Templates · 37 maintained stacks

# 37 GPU templates: from "deploy" to working stack in 30 seconds

Every template is a pinned, GPU-ready image with ports mapped and storage mounted — official builds of the tools you already use. Launch PyTorch, serve with vLLM, generate with ComfyUI, fine-tune with Axolotl — or point us at your own Docker image.

## Base & frameworks (6)

- **[NVIDIA CUDA](https://powergpu.ai/templates/nvidia-cuda)** The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top. (CUDA 13) (ARM) (SSH) (Jupyter) powergpu/base:cuda13 [GPUs & details](https://powergpu.ai/templates/nvidia-cuda) [Deploy](https://cloud.powergpu.ai/)
- **[PyTorch](https://powergpu.ai/templates/pytorch)** The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box. (CUDA 12.8) (ARM) (SSH) (Jupyter) powergpu/pytorch:2.6-cuda12.8 [GPUs & details](https://powergpu.ai/templates/pytorch) [Deploy](https://cloud.powergpu.ai/)
- **[TensorFlow CUDA](https://powergpu.ai/templates/tensorflow-cuda)** TensorFlow with GPU support, Keras and TensorBoard on a mapped port. (CUDA 12.1) (SSH) (Jupyter) powergpu/tensorflow:cuda12.1 [GPUs & details](https://powergpu.ai/templates/tensorflow-cuda) [Deploy](https://cloud.powergpu.ai/)
- **[PyTorch NGC](https://powergpu.ai/templates/pytorch-ngc)** NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards. (SSH) (VM) nvcr.io/nvidia/pytorch [GPUs & details](https://powergpu.ai/templates/pytorch-ngc) [Deploy](https://cloud.powergpu.ai/)
- **[NVIDIA RAPIDS](https://powergpu.ai/templates/nvidia-rapids)** GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook. (CUDA 13) (ARM) (SSH) (Jupyter) rapidsai/notebooks [GPUs & details](https://powergpu.ai/templates/nvidia-rapids) [Deploy](https://cloud.powergpu.ai/)
- **[All-in-One App Studio](https://powergpu.ai/templates/all-in-one-app-studio)** A launcher bundling the most-used AI apps behind one desktop — pick and run. (CUDA 12.9) (SSH) (Jupyter) powergpu/aio-studio [GPUs & details](https://powergpu.ai/templates/all-in-one-app-studio) [Deploy](https://cloud.powergpu.ai/)

## LLM serving & chat (9)

- **[vLLM](https://powergpu.ai/templates/vllm)** Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model. (CUDA 13) (ARM) (SSH) (Jupyter) powergpu/vllm [GPUs & details](https://powergpu.ai/templates/vllm) [Deploy](https://cloud.powergpu.ai/)
- **[vLLM Omni](https://powergpu.ai/templates/vllm-omni)** vLLM extended for multimodal models — vision and audio inputs on the same fast server. (CUDA 12.9) (SSH) (Jupyter) powergpu/vllm-omni [GPUs & details](https://powergpu.ai/templates/vllm-omni) [Deploy](https://cloud.powergpu.ai/)
- **[SGLang](https://powergpu.ai/templates/sglang)** High-throughput serving with RadixAttention — excels at structured and agentic workloads. (CUDA 13) (ARM) (SSH) (Jupyter) powergpu/sglang [GPUs & details](https://powergpu.ai/templates/sglang) [Deploy](https://cloud.powergpu.ai/)
- **[Llama.cpp](https://powergpu.ai/templates/llama-cpp)** GGUF inference with a built-in server — the lightest way to run quantized models. (CUDA 12.9) (ARM) (SSH) (Jupyter) powergpu/llama-cpp [GPUs & details](https://powergpu.ai/templates/llama-cpp) [Deploy](https://cloud.powergpu.ai/)
- **[Ollama](https://powergpu.ai/templates/ollama)** Pull and run quantized models in one command, with a clean REST API. (ARM) (SSH) (Jupyter) powergpu/ollama [GPUs & details](https://powergpu.ai/templates/ollama) [Deploy](https://cloud.powergpu.ai/)
- **[Open WebUI (Ollama)](https://powergpu.ai/templates/open-webui-ollama)** A full chat UI over Ollama — conversations, RAG and model management in the browser. (CUDA 12.9) (SSH) (Jupyter) powergpu/openwebui [GPUs & details](https://powergpu.ai/templates/open-webui-ollama) [Deploy](https://cloud.powergpu.ai/)
- **[Oobabooga Text Gen](https://powergpu.ai/templates/oobabooga-text-gen)** The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions. (CUDA 12.4) (SSH) (Jupyter) powergpu/oobabooga [GPUs & details](https://powergpu.ai/templates/oobabooga-text-gen) [Deploy](https://cloud.powergpu.ai/)
- **[HuggingFace TGI](https://powergpu.ai/templates/huggingface-tgi)** Hugging Face Text Generation Inference — battle-tested production serving with an API. (SSH) (Jupyter) ghcr.io/huggingface/text-generation-inference [GPUs & details](https://powergpu.ai/templates/huggingface-tgi) [Deploy](https://cloud.powergpu.ai/)
- **[Langflow (Ollama)](https://powergpu.ai/templates/langflow-ollama)** Visually build LLM pipelines and agents on top of a local Ollama backend. (CUDA 12.6) (SSH) (Jupyter) powergpu/langflow [GPUs & details](https://powergpu.ai/templates/langflow-ollama) [Deploy](https://cloud.powergpu.ai/)

## Training & fine-tuning (5)

- **[Axolotl — Fine Tuning](https://powergpu.ai/templates/axolotl-fine-tuning)** Fine-tune Llama, Qwen or Mistral from one YAML — LoRA, QLoRA and multi-GPU FSDP. (CUDA 12.6) (ARM) (SSH) (Jupyter) axolotlai/axolotl-cloud [GPUs & details](https://powergpu.ai/templates/axolotl-fine-tuning) [Deploy](https://cloud.powergpu.ai/)
- **[Unsloth Studio](https://powergpu.ai/templates/unsloth-studio)** 2× faster, lower-VRAM fine-tuning — Unsloth kernels with a notebook workflow. (CUDA 12.9) (SSH) (Jupyter) powergpu/unsloth-studio [GPUs & details](https://powergpu.ai/templates/unsloth-studio) [Deploy](https://cloud.powergpu.ai/)
- **[Kohya's GUI](https://powergpu.ai/templates/kohya-s-gui)** Train SDXL and Flux image LoRAs in a browser GUI — dataset in,.safetensors out. (CUDA 12.6) (SSH) (Jupyter) powergpu/kohyas-gui [GPUs & details](https://powergpu.ai/templates/kohya-s-gui) [Deploy](https://cloud.powergpu.ai/)
- **[Flux Gym](https://powergpu.ai/templates/flux-gym)** A dead-simple UI for training Flux LoRAs, built on the Kohya scripts. (ARM) (SSH) (Jupyter) powergpu/fluxgym [GPUs & details](https://powergpu.ai/templates/flux-gym) [Deploy](https://cloud.powergpu.ai/)
- **[Ostris AI Toolkit](https://powergpu.ai/templates/ostris-ai-toolkit)** The AI-Toolkit trainer for Flux and diffusion models, with a web UI. (CUDA 12.9) (SSH) (Jupyter) powergpu/ostris-ai-toolkit [GPUs & details](https://powergpu.ai/templates/ostris-ai-toolkit) [Deploy](https://cloud.powergpu.ai/)

## Image generation (6)

- **[ComfyUI](https://powergpu.ai/templates/comfyui)** Node-based image and video workflows with ComfyUI-Manager pre-installed. (CUDA 13) (ARM) (SSH) (Jupyter) powergpu/comfyui [GPUs & details](https://powergpu.ai/templates/comfyui) [Deploy](https://cloud.powergpu.ai/)
- **[InvokeAI](https://powergpu.ai/templates/invokeai)** A polished Stable Diffusion studio — canvas, layers and workflow nodes. (CUDA 12.9) (ARM) (SSH) (Jupyter) powergpu/invokeai [GPUs & details](https://powergpu.ai/templates/invokeai) [Deploy](https://cloud.powergpu.ai/)
- **[SD WebUI Forge](https://powergpu.ai/templates/sd-webui-forge)** The Forge fork of the SD WebUI — faster attention, lower VRAM, same extensions. (CUDA 12.8) (ARM) (SSH) (Jupyter) powergpu/sd-forge [GPUs & details](https://powergpu.ai/templates/sd-webui-forge) [Deploy](https://cloud.powergpu.ai/)
- **[SD WebUI A1111](https://powergpu.ai/templates/sd-webui-a1111)** The classic AUTOMATIC1111 Stable Diffusion WebUI with the full extension ecosystem. (CUDA 12.1) (ARM) (SSH) (Jupyter) powergpu/a1111 [GPUs & details](https://powergpu.ai/templates/sd-webui-a1111) [Deploy](https://cloud.powergpu.ai/)
- **[Fooocus](https://powergpu.ai/templates/fooocus)** Prompt-and-go image generation — Midjourney-style simplicity on SDXL. (CUDA 12.1) (SSH) (Jupyter) powergpu/fooocus [GPUs & details](https://powergpu.ai/templates/fooocus) [Deploy](https://cloud.powergpu.ai/)
- **[SwarmUI](https://powergpu.ai/templates/swarmui)** A ComfyUI-backed UI that scales generation across multiple GPUs. (CUDA 12.4) (SSH) (Jupyter) powergpu/swarmui [GPUs & details](https://powergpu.ai/templates/swarmui) [Deploy](https://cloud.powergpu.ai/)

## Video generation (2)

- **[Wan2GP](https://powergpu.ai/templates/wan2gp)** Run Wan and other video-generation models on modest VRAM, with a simple UI. (CUDA 12.9) (SSH) (Jupyter) powergpu/wan2gp [GPUs & details](https://powergpu.ai/templates/wan2gp) [Deploy](https://cloud.powergpu.ai/)
- **[Open-Sora](https://powergpu.ai/templates/open-sora)** The open text-to-video model, ready to generate and fine-tune. (CUDA 12.1) (SSH) (Jupyter) powergpu/open-sora [GPUs & details](https://powergpu.ai/templates/open-sora) [Deploy](https://cloud.powergpu.ai/)

## Audio & speech (3)

- **[Whisper WebUI & API](https://powergpu.ai/templates/whisper-webui-api)** Batch-transcribe audio with faster-whisper behind a UI and a REST endpoint. (CUDA 12.6) (ARM) (SSH) (Jupyter) powergpu/whisper [GPUs & details](https://powergpu.ai/templates/whisper-webui-api) [Deploy](https://cloud.powergpu.ai/)
- **[Voicebox TTS](https://powergpu.ai/templates/voicebox-tts)** Text-to-speech and voice cloning with a web interface and API. (CUDA 12.1) (SSH) (Jupyter) powergpu/voicebox [GPUs & details](https://powergpu.ai/templates/voicebox-tts) [Deploy](https://cloud.powergpu.ai/)
- **[ACE Step 1.5](https://powergpu.ai/templates/ace-step-1-5)** The ACE-Step music generation model with a UI and API. (CUDA 12.4) (SSH) (Jupyter) powergpu/acestep [GPUs & details](https://powergpu.ai/templates/ace-step-1-5) [Deploy](https://cloud.powergpu.ai/)

## Desktops & VMs (4)

- **[Linux Desktop](https://powergpu.ai/templates/linux-desktop)** A full GPU Linux desktop over the browser — XFCE, VNC/RDP, run GUI apps. (CUDA 12.1) (ARM) (SSH) (Jupyter) powergpu/linux-desktop [GPUs & details](https://powergpu.ai/templates/linux-desktop) [Deploy](https://cloud.powergpu.ai/)
- **[Pinokio](https://powergpu.ai/templates/pinokio)** The Pinokio 1-click app browser on a desktop — install AI apps with a click. (CUDA 13) (ARM) (SSH) (Jupyter) powergpu/pinokio [GPUs & details](https://powergpu.ai/templates/pinokio) [Deploy](https://cloud.powergpu.ai/)
- **[Ubuntu 22.04 VM](https://powergpu.ai/templates/ubuntu-22-04-vm)** A full KVM virtual machine — your own kernel, root, systemd, any stack. (SSH) (VM) powergpu/kvm:ubuntu-22.04 [GPUs & details](https://powergpu.ai/templates/ubuntu-22-04-vm) [Deploy](https://cloud.powergpu.ai/)
- **[Ubuntu Desktop VM](https://powergpu.ai/templates/ubuntu-desktop-vm)** The Ubuntu VM with a GNOME desktop over VNC/RDP — a full graphical workstation. (SSH) (VM) powergpu/kvm:ubuntu-desktop [GPUs & details](https://powergpu.ai/templates/ubuntu-desktop-vm) [Deploy](https://cloud.powergpu.ai/)

## Specialised (2)

- **[Unreal Pixel Streaming](https://powergpu.ai/templates/unreal-pixel-streaming)** Stream an Unreal Engine app to the browser — pixel streaming on a cloud GPU. (CUDA 12.8) (SSH) (Jupyter) powergpu/unreal-pixel-streaming [GPUs & details](https://powergpu.ai/templates/unreal-pixel-streaming) [Deploy](https://cloud.powergpu.ai/)
- **[Hashcat CUDA](https://powergpu.ai/templates/hashcat-cuda)** GPU password recovery for authorised security testing — CUDA-accelerated Hashcat. (SSH) dizcza/docker-hashcat [GPUs & details](https://powergpu.ai/templates/hashcat-cuda) [Deploy](https://cloud.powergpu.ai/)

## Bring your own image

Nothing here fits? The deploy form takes any OCI reference — we inject the NVIDIA runtime and run your entrypoint untouched.

- Public or private registries (credentials stored encrypted).
- Your ports, env vars and volume mounts, set at deploy.
- Pinned by tag — a redeploy is byte-identical next month.

*deploy — any image*

```
$ powergpu launch --gpu l40s \
    --image ghcr.io/acme/trainer:v14 \
    --ports 8080 --disk 100
✓ instance i-4fd02b11 running (31.2s)
```

## GPU templates FAQ

Environment variables, ports and custom images in the [template docs](https://powergpu.ai/docs/templates).

**What exactly is a template?**

A maintained container image plus sane launch defaults: exposed ports, volume mount points, health checks and environment variables. Pick one at deploy, override anything, and the instance boots straight into a working stack — usually in about 30 seconds.

**Can I use my own Docker image instead?**

Yes — the deploy form accepts any public or private OCI image reference (registry credentials are stored encrypted per-account). We inject the NVIDIA runtime; your entrypoint is untouched.

**What do the ARM, SSH, Jupyter and VM tags mean?**

They tell you how a template runs: ARM = also builds for ARM hosts, SSH = direct shell access, Jupyter = a notebook server on a mapped port, VM = a full virtual machine rather than a container.

**Do templates cost extra?**

No. You pay the GPU price — a ComfyUI session on an RTX 4090 is $0.327/hr, template included, billed per second.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/nvidia-cuda -->

---
title: "Run CUDA on a cloud GPU from $0.163/hr | PowerGPU"
description: "NVIDIA CUDA on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/nvidia-cuda
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Base & frameworks · CUDA 13

# Run CUDA on a cloud GPU, in 30 seconds

The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top. The bare CUDA image is the blank canvas: the NVIDIA driver stack, the CUDA 13 toolkit, nvcc, cuDNN, SSH and JupyterLab — nothing else. Use it to compile custom kernels, build your own framework stack, or reproduce a paper environment exactly, without fighting a pre-baked image. From **$0.163** /hr on an interruptible RTX 4090.

NVIDIA CUDA

powergpu/base:cuda13

(CUDA 13) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: NVIDIA CUDA on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for NVIDIA CUDA

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Cheap Ada card for compiling and testing kernels interactively. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.560 | $0.280 | Datacenter Ampere with FP64 and MIG for library development. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| (Best) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1.428 | $0.714 | Hopper features (FP8, TMA, thread-block clusters) for kernels that target the latest architecture. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy NVIDIA CUDA from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *NVIDIA CUDA* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "nvidia-cuda". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — nvidia-cuda*

```
$ powergpu launch --gpu rtx-4090 --template nvidia-cuda \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# NVIDIA CUDA · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the NVIDIA CUDA template

- **Image**: powergpu/base:cuda13
- **CUDA**: CUDA 13
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Base & frameworks](https://powergpu.ai/templates#base)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other base & frameworks templates

- [PyTorch on a cloud GPU](https://powergpu.ai/templates/pytorch) — The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box.
- [TensorFlow CUDA on a cloud GPU](https://powergpu.ai/templates/tensorflow-cuda) — TensorFlow with GPU support, Keras and TensorBoard on a mapped port.
- [PyTorch NGC on a cloud GPU](https://powergpu.ai/templates/pytorch-ngc) — NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards.
- [NVIDIA RAPIDS on a cloud GPU](https://powergpu.ai/templates/nvidia-rapids) — GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook.
- [All-in-One App Studio on a cloud GPU](https://powergpu.ai/templates/all-in-one-app-studio) — A launcher bundling the most-used AI apps behind one desktop — pick and run.
- [All 37 templates](https://powergpu.ai/templates)

## NVIDIA CUDA on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run NVIDIA CUDA on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does NVIDIA CUDA take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Which CUDA version does the template ship?**

CUDA 13 userspace on hosts running the current NVIDIA production driver. Templates pinned to CUDA 12.x exist for frameworks that need them; the offer card shows the maximum CUDA each machine supports.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/nvidia-cuda · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/pytorch -->

---
title: "Run PyTorch on a cloud GPU from $0.163/hr | PowerGPU"
description: "PyTorch on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/pytorch
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Base & frameworks · CUDA 12.8

# Run PyTorch on a cloud GPU, in 30 seconds

The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box. PyTorch 2.6 with CUDA 12.8, cuDNN, torchvision and torchaudio, JupyterLab on a mapped port and SSH — the environment most research code assumes. Datasets and checkpoints live on a volume, so the instance stays disposable and per-second billing does the rest. From **$0.163** /hr on an interruptible RTX 4090.

PyTorch

powergpu/pytorch:2.6-cuda12.8

(CUDA 12.8) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: PyTorch on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for PyTorch

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB and Ada tensor cores: the best price for single-GPU experiments and LoRA runs. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.560 | $0.280 | 80 GB HBM2e with NVLink for multi-GPU training up to 8× on one machine. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| (Best) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1.428 | $0.714 | FP8 and 3.35 TB/s for full fine-tunes and pre-training at the lowest cost per step. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy PyTorch from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *PyTorch* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "pytorch". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — pytorch*

```
$ powergpu launch --gpu rtx-4090 --template pytorch \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# PyTorch · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the PyTorch template

- **Image**: powergpu/pytorch:2.6-cuda12.8
- **CUDA**: CUDA 12.8
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Base & frameworks](https://powergpu.ai/templates#base)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other base & frameworks templates

- [NVIDIA CUDA on a cloud GPU](https://powergpu.ai/templates/nvidia-cuda) — The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top.
- [TensorFlow CUDA on a cloud GPU](https://powergpu.ai/templates/tensorflow-cuda) — TensorFlow with GPU support, Keras and TensorBoard on a mapped port.
- [PyTorch NGC on a cloud GPU](https://powergpu.ai/templates/pytorch-ngc) — NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards.
- [NVIDIA RAPIDS on a cloud GPU](https://powergpu.ai/templates/nvidia-rapids) — GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook.
- [All-in-One App Studio on a cloud GPU](https://powergpu.ai/templates/all-in-one-app-studio) — A launcher bundling the most-used AI apps behind one desktop — pick and run.
- [All 37 templates](https://powergpu.ai/templates)

## PyTorch on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run PyTorch on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does PyTorch take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Does it support torch.compile and FSDP?**

Yes — the image ships the matching Triton build for torch.compile, and FSDP/DDP work out of the box on multi-GPU machines (NCCL over NVLink where the hardware has it).

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/pytorch · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/tensorflow-cuda -->

---
title: "Run TensorFlow on a cloud GPU from $0.054/hr | PowerGPU"
description: "TensorFlow CUDA on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.054/hr on RTX 3090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/tensorflow-cuda
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Base & frameworks · CUDA 12.1

# Run TensorFlow on a cloud GPU, in 30 seconds

TensorFlow with GPU support, Keras and TensorBoard on a mapped port. TensorFlow with GPU support, Keras and TensorBoard exposed on its own port. Ideal for legacy training pipelines, TF-Serving experiments and courses that standardised on Keras. Save models to a volume; TensorBoard logs survive instance restarts. From **$0.054** /hr on an interruptible RTX 3090.

TensorFlow CUDA

powergpu/tensorflow:cuda12.1

(CUDA 12.1) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: TensorFlow CUDA on a RTX 3090 is $0.108/hr on-demand, $0.054/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for TensorFlow CUDA

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | Cheapest 24 GB for classic CNN/RNN training. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Ada tensor cores speed up mixed-precision Keras models. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB | $0.374 | $0.187 | 80 GB for large-batch training and TF-TRT inference. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-pcie) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy TensorFlow CUDA from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *TensorFlow CUDA* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "tensorflow-cuda". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — tensorflow-cuda*

```
$ powergpu launch --gpu rtx-3090 --template tensorflow-cuda \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# TensorFlow CUDA · RTX 3090 · $0.108/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the TensorFlow CUDA template

- **Image**: powergpu/tensorflow:cuda12.1
- **CUDA**: CUDA 12.1
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Base & frameworks](https://powergpu.ai/templates#base)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other base & frameworks templates

- [NVIDIA CUDA on a cloud GPU](https://powergpu.ai/templates/nvidia-cuda) — The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top.
- [PyTorch on a cloud GPU](https://powergpu.ai/templates/pytorch) — The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box.
- [PyTorch NGC on a cloud GPU](https://powergpu.ai/templates/pytorch-ngc) — NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards.
- [NVIDIA RAPIDS on a cloud GPU](https://powergpu.ai/templates/nvidia-rapids) — GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook.
- [All-in-One App Studio on a cloud GPU](https://powergpu.ai/templates/all-in-one-app-studio) — A launcher bundling the most-used AI apps behind one desktop — pick and run.
- [All 37 templates](https://powergpu.ai/templates)

## TensorFlow CUDA on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run TensorFlow CUDA on a cloud GPU?**

Only the GPU price — the template is free. From $0.054/hr on an interruptible RTX 3090, $0.108/hr on-demand on a RTX 3090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does TensorFlow CUDA take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Which TensorFlow version is installed?**

The current TensorFlow 2.x GPU build against CUDA 12.1 and matching cuDNN. Pin a different release with pip inside the instance, or bring your own image.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/tensorflow-cuda · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/pytorch-ngc -->

---
title: "Run NVIDIA NGC PyTorch on a cloud GPU from $0.280/hr | PowerGPU"
description: "PyTorch NGC on a fixed-price cloud GPU: one-click template, boots in minutes, from $0.280/hr on A100 SXM4. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/pytorch-ngc
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Base & frameworks

# Run NVIDIA NGC PyTorch on a cloud GPU, in minutes

NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards. NVIDIA's NGC PyTorch container is the tuned build: pre-compiled kernels for Hopper and Ampere, APEX, DALI, Transformer Engine and FP8 support already wired. It is what large-scale training teams run on datacenter cards; rent it here with the same image, no registry login required. From **$0.280** /hr on an interruptible A100 SXM4.

PyTorch NGC

nvcr.io/nvidia/pytorch

(SSH) (VM)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Boots in minutes

A full KVM virtual machine with the GPU passed through: your kernel, root, systemd.

### Template is free

You pay the GPU price only: PyTorch NGC on a A100 SXM4 is $0.560/hr on-demand, $0.280/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — SSH behind your own credentials.

## Best GPUs for PyTorch NGC

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.560 | $0.280 | Ampere datacenter card the NGC stack is tuned for. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| (Better) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1.428 | $0.714 | Transformer Engine FP8 paths shine on Hopper. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |
| (Best) | [H200](https://powergpu.ai/gpu/h200) | 141 GB | $2.791 | $1.395 | 141 GB for longer context and bigger micro-batches under the same tuned stack. | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy PyTorch NGC from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *PyTorch NGC* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "pytorch-ngc". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — pytorch-ngc*

```
$ powergpu launch --gpu a100-sxm4 --template pytorch-ngc \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (2m38s)
# PyTorch NGC · A100 SXM4 · $0.560/hr · per second
# https://i-7a41c0e2.powergpu.ai:8000 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the PyTorch NGC template

- **Image**: nvcr.io/nvidia/pytorch
- **CUDA**: inherits host driver (any 12.x / 13.x)
- **Access**: SSH shell · full virtual machine (KVM)
- **Category**: [Base & frameworks](https://powergpu.ai/templates#base)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other base & frameworks templates

- [NVIDIA CUDA on a cloud GPU](https://powergpu.ai/templates/nvidia-cuda) — The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top.
- [PyTorch on a cloud GPU](https://powergpu.ai/templates/pytorch) — The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box.
- [TensorFlow CUDA on a cloud GPU](https://powergpu.ai/templates/tensorflow-cuda) — TensorFlow with GPU support, Keras and TensorBoard on a mapped port.
- [NVIDIA RAPIDS on a cloud GPU](https://powergpu.ai/templates/nvidia-rapids) — GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook.
- [All-in-One App Studio on a cloud GPU](https://powergpu.ai/templates/all-in-one-app-studio) — A launcher bundling the most-used AI apps behind one desktop — pick and run.
- [All 37 templates](https://powergpu.ai/templates)

## PyTorch NGC on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run PyTorch NGC on a cloud GPU?**

Only the GPU price — the template is free. From $0.280/hr on an interruptible A100 SXM4, $0.560/hr on-demand on a A100 SXM4. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does PyTorch NGC take to start?**

Two to four minutes: this template is a full virtual machine that boots its own kernel. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Why choose the NGC build over plain PyTorch?**

On datacenter GPUs the NGC container is typically 10–30% faster on transformer training thanks to Transformer Engine, fused kernels and NCCL tuning — and it is validated release by release by NVIDIA.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/pytorch-ngc · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/nvidia-rapids -->

---
title: "Run RAPIDS on a cloud GPU from $0.140/hr | PowerGPU"
description: "NVIDIA RAPIDS on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.140/hr on RTX A6000. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/nvidia-rapids
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Base & frameworks · CUDA 13

# Run RAPIDS on a cloud GPU, in 30 seconds

GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook. RAPIDS turns pandas, scikit-learn and NetworkX workflows into GPU code: cuDF, cuML, cuGraph and Dask-CUDA inside a JupyterLab notebook. Load a 50 GB parquet dataset into GPU memory and run a groupby in milliseconds — then stop the instance and pay for the minutes you used. From **$0.140** /hr on an interruptible RTX A6000.

NVIDIA RAPIDS

rapidsai/notebooks

(CUDA 13) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: NVIDIA RAPIDS on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for NVIDIA RAPIDS

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB of fast GDDR6X for dataframes that no longer fit in pandas. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | $0.281 | $0.140 | 48 GB with ECC for larger tables and graph analytics. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a6000) |
| (Best) | [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | 80 GB | $0.374 | $0.187 | 80 GB HBM2e — whole datasets resident, Dask-CUDA across multiple cards. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-pcie) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy NVIDIA RAPIDS from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *NVIDIA RAPIDS* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "nvidia-rapids". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — nvidia-rapids*

```
$ powergpu launch --gpu rtx-4090 --template nvidia-rapids \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# NVIDIA RAPIDS · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the NVIDIA RAPIDS template

- **Image**: rapidsai/notebooks
- **CUDA**: CUDA 13
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Base & frameworks](https://powergpu.ai/templates#base)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other base & frameworks templates

- [NVIDIA CUDA on a cloud GPU](https://powergpu.ai/templates/nvidia-cuda) — The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top.
- [PyTorch on a cloud GPU](https://powergpu.ai/templates/pytorch) — The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box.
- [TensorFlow CUDA on a cloud GPU](https://powergpu.ai/templates/tensorflow-cuda) — TensorFlow with GPU support, Keras and TensorBoard on a mapped port.
- [PyTorch NGC on a cloud GPU](https://powergpu.ai/templates/pytorch-ngc) — NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards.
- [All-in-One App Studio on a cloud GPU](https://powergpu.ai/templates/all-in-one-app-studio) — A launcher bundling the most-used AI apps behind one desktop — pick and run.
- [All 37 templates](https://powergpu.ai/templates)

## NVIDIA RAPIDS on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run NVIDIA RAPIDS on a cloud GPU?**

Only the GPU price — the template is free. From $0.140/hr on an interruptible RTX A6000, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does NVIDIA RAPIDS take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I scale RAPIDS across several GPUs?**

Yes — Dask-CUDA is preinstalled; deploy a 2×–8× machine and a LocalCUDACluster spreads cuDF partitions across all cards on the host.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/nvidia-rapids · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/all-in-one-app-studio -->

---
title: "Run an all-in-one AI app studio on a cloud GPU | PowerGPU"
description: "All-in-One App Studio on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/all-in-one-app-studio
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Base & frameworks · CUDA 12.9

# Run an all-in-one AI app studio on a cloud GPU, in 30 seconds

A launcher bundling the most-used AI apps behind one desktop — pick and run. One desktop, many launchers: the App Studio bundles the most-used community AI apps behind a single web UI so you can start ComfyUI, an LLM chat, a TTS tool or a trainer without building an image for each. Perfect for exploring tools before committing a workflow. From **$0.163** /hr on an interruptible RTX 4090.

All-in-One App Studio

powergpu/aio-studio

(CUDA 12.9) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: All-in-One App Studio on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for All-in-One App Studio

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Runs every bundled app comfortably in 24 GB. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB of headroom for video and larger LLMs inside the studio. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | 96 GB — run several apps simultaneously with big models loaded. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy All-in-One App Studio from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *All-in-One App Studio* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "all-in-one-app-studio". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — all-in-one-app-studio*

```
$ powergpu launch --gpu rtx-4090 --template all-in-one-app-studio \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# All-in-One App Studio · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the All-in-One App Studio template

- **Image**: powergpu/aio-studio
- **CUDA**: CUDA 12.9
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Base & frameworks](https://powergpu.ai/templates#base)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other base & frameworks templates

- [NVIDIA CUDA on a cloud GPU](https://powergpu.ai/templates/nvidia-cuda) — The bare CUDA base image — drivers, toolkit, SSH and Jupyter. Build your own stack on top.
- [PyTorch on a cloud GPU](https://powergpu.ai/templates/pytorch) — The training and research default: PyTorch 2.6, cuDNN, JupyterLab and SSH out of the box.
- [TensorFlow CUDA on a cloud GPU](https://powergpu.ai/templates/tensorflow-cuda) — TensorFlow with GPU support, Keras and TensorBoard on a mapped port.
- [PyTorch NGC on a cloud GPU](https://powergpu.ai/templates/pytorch-ngc) — NVIDIA's NGC-optimised PyTorch build — tuned kernels for the datacenter cards.
- [NVIDIA RAPIDS on a cloud GPU](https://powergpu.ai/templates/nvidia-rapids) — GPU-accelerated data science — cuDF, cuML, cuGraph in a notebook.
- [All 37 templates](https://powergpu.ai/templates)

## All-in-One App Studio on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run All-in-One App Studio on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does All-in-One App Studio take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I keep the apps I installed between sessions?**

Mount a volume on the studio's data path: apps, models and outputs persist, and the next instance boots with everything in place.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/all-in-one-app-studio · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/vllm -->

---
title: "Run vLLM on a cloud GPU from $0.219/hr | PowerGPU"
description: "vLLM on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.219/hr on RTX 5090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/vllm
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat · CUDA 13

# Run vLLM on a cloud GPU, in 30 seconds

Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model. vLLM is the production standard for serving open LLMs: PagedAttention, continuous batching, tensor parallelism and an OpenAI-compatible API. The template exposes port 8000 with TLS, reads MODEL from an environment variable and caches weights on the instance disk or a mounted volume. From **$0.219** /hr on an interruptible RTX 5090.

vLLM

powergpu/vllm

(CUDA 13) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: vLLM on a RTX 5090 is $0.439/hr on-demand, $0.219/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for vLLM

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7: the lowest $/token for 7B–14B FP16 and 32B 4-bit models. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | 48 GB and datacenter cooling for 24/7 endpoints up to 32B. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| (Best) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB for quantized 70B on one card, FP8 for maximum throughput. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy vLLM from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *vLLM* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "vllm". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — vllm*

```
$ powergpu launch --gpu rtx-5090 --template vllm \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# vLLM · RTX 5090 · $0.439/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the vLLM template

- **Image**: powergpu/vllm
- **CUDA**: CUDA 13
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [Oobabooga Text Gen on a cloud GPU](https://powergpu.ai/templates/oobabooga-text-gen) — The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions.
- [All 37 templates](https://powergpu.ai/templates)

## vLLM on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run vLLM on a cloud GPU?**

Only the GPU price — the template is free. From $0.219/hr on an interruptible RTX 5090, $0.439/hr on-demand on a RTX 5090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does vLLM take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Is the endpoint really OpenAI-compatible?**

Yes — point any OpenAI SDK at https://<instance>.powergpu.ai:8000/v1 with the API key you set in VLLM_API_KEY. Chat, completions, embeddings and streaming all work unchanged.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/vllm · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/vllm-omni -->

---
title: "Run multimodal vLLM on a cloud GPU from $0.219/hr | PowerGPU"
description: "vLLM Omni on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.219/hr on RTX 5090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/vllm-omni
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat · CUDA 12.9

# Run multimodal vLLM on a cloud GPU, in 30 seconds

vLLM extended for multimodal models — vision and audio inputs on the same fast server. vLLM Omni extends the same fast server to vision and audio inputs: Qwen-VL, LLaVA-class and speech-capable models behind the OpenAI-compatible API, with the same continuous batching. Rent it when your product needs images or audio in the prompt without a second inference stack. From **$0.219** /hr on an interruptible RTX 5090.

vLLM Omni

powergpu/vllm-omni

(CUDA 12.9) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: vLLM Omni on a RTX 5090 is $0.439/hr on-demand, $0.219/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for vLLM Omni

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | Fast entry point for 7B-class vision-language models. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | 48 GB for larger multimodal models and longer image contexts. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| (Best) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB and FP8 for production multimodal throughput. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy vLLM Omni from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *vLLM Omni* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "vllm-omni". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — vllm-omni*

```
$ powergpu launch --gpu rtx-5090 --template vllm-omni \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# vLLM Omni · RTX 5090 · $0.439/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the vLLM Omni template

- **Image**: powergpu/vllm-omni
- **CUDA**: CUDA 12.9
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [Oobabooga Text Gen on a cloud GPU](https://powergpu.ai/templates/oobabooga-text-gen) — The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions.
- [All 37 templates](https://powergpu.ai/templates)

## vLLM Omni on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run vLLM Omni on a cloud GPU?**

Only the GPU price — the template is free. From $0.219/hr on an interruptible RTX 5090, $0.439/hr on-demand on a RTX 5090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does vLLM Omni take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Which multimodal models are supported?**

Anything in vLLM's multimodal registry — Qwen2.5-VL, LLaVA variants, Pixtral, Whisper-style audio encoders and more. Set MODEL to the Hugging Face id and the server loads the right processor.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/vllm-omni · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/sglang -->

---
title: "Run SGLang on a cloud GPU from $0.219/hr | PowerGPU"
description: "SGLang on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.219/hr on RTX 5090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/sglang
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat · CUDA 13

# Run SGLang on a cloud GPU, in 30 seconds

High-throughput serving with RadixAttention — excels at structured and agentic workloads. SGLang pairs RadixAttention prefix caching with a structured-generation frontend, which makes it the fastest server for agentic and JSON-heavy workloads where many requests share prompts. Same OpenAI-style API as vLLM, often higher throughput on multi-turn traffic. From **$0.219** /hr on an interruptible RTX 5090.

SGLang

powergpu/sglang

(CUDA 13) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: SGLang on a RTX 5090 is $0.439/hr on-demand, $0.219/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for SGLang

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | Cheap, fast card for 7B–14B agent backends. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | 48 GB for 32B-class models with long shared prefixes. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| (Best) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1.428 | $0.714 | Multi-GPU tensor parallel for 70B+ agent fleets. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy SGLang from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *SGLang* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "sglang". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — sglang*

```
$ powergpu launch --gpu rtx-5090 --template sglang \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# SGLang · RTX 5090 · $0.439/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the SGLang template

- **Image**: powergpu/sglang
- **CUDA**: CUDA 13
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [Oobabooga Text Gen on a cloud GPU](https://powergpu.ai/templates/oobabooga-text-gen) — The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions.
- [All 37 templates](https://powergpu.ai/templates)

## SGLang on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run SGLang on a cloud GPU?**

Only the GPU price — the template is free. From $0.219/hr on an interruptible RTX 5090, $0.439/hr on-demand on a RTX 5090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does SGLang take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**When is SGLang faster than vLLM?**

On workloads with heavy prompt reuse — agents, RAG with fixed system prompts, batch structured extraction — RadixAttention skips recomputation of shared prefixes and can double effective throughput.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/sglang · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/llama-cpp -->

---
title: "Run llama.cpp on a cloud GPU from $0.021/hr | PowerGPU"
description: "Llama.cpp on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.021/hr on RTX 3060. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/llama-cpp
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat · CUDA 12.9

# Run llama.cpp on a cloud GPU, in 30 seconds

GGUF inference with a built-in server — the lightest way to run quantized models. llama.cpp runs GGUF models with CUDA offload and ships a lightweight OpenAI-compatible server. It is the most memory-frugal way to serve quantized models and the natural home for Q4/Q5/Q6 GGUF files from Hugging Face — including models that do not fit vLLM's formats. From **$0.021** /hr on an interruptible RTX 3060.

Llama.cpp

powergpu/llama-cpp

(CUDA 12.9) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Llama.cpp on a RTX 3060 is $0.042/hr on-demand, $0.021/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Llama.cpp

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.042 | $0.021 | 12 GB is enough for 7B–8B Q4 GGUF — the cheapest chat endpoint on the sheet. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB runs 14B–32B quantized models with full GPU offload. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | $0.281 | $0.140 | 48 GB for 70B Q4 GGUF on a single card. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a6000) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Llama.cpp from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Llama.cpp* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "llama-cpp". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — llama-cpp*

```
$ powergpu launch --gpu rtx-3060 --template llama-cpp \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Llama.cpp · RTX 3060 · $0.042/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Llama.cpp template

- **Image**: powergpu/llama-cpp
- **CUDA**: CUDA 12.9
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [Oobabooga Text Gen on a cloud GPU](https://powergpu.ai/templates/oobabooga-text-gen) — The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions.
- [All 37 templates](https://powergpu.ai/templates)

## Llama.cpp on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Llama.cpp on a cloud GPU?**

Only the GPU price — the template is free. From $0.021/hr on an interruptible RTX 3060, $0.042/hr on-demand on a RTX 3060. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Llama.cpp take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I split a model between GPU and CPU RAM?**

Yes — set the number of offloaded layers (-ngl) at launch. Machines with large RAM let you run models bigger than VRAM at reduced speed, which is handy for occasional 70B use on a cheap card.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/llama-cpp · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/ollama -->

---
title: "Run Ollama on a cloud GPU from $0.021/hr | PowerGPU"
description: "Ollama on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.021/hr on RTX 3060. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/ollama
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat

# Run Ollama on a cloud GPU, in 30 seconds

Pull and run quantized models in one command, with a clean REST API. Ollama pulls and runs quantized models with one command and exposes a clean REST API on port 11434. It is the fastest way to get Llama, Qwen, Gemma or Mistral answering requests — and per-second billing turns a test into pennies. From **$0.021** /hr on an interruptible RTX 3060.

Ollama

powergpu/ollama

(ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Ollama on a RTX 3060 is $0.042/hr on-demand, $0.021/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Ollama

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.042 | $0.021 | 12 GB: 7B–8B models at the lowest price on the sheet. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB: 14B FP16 or 32B 4-bit models, fast. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7: the best tokens per dollar for mid-size models. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Ollama from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Ollama* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "ollama". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — ollama*

```
$ powergpu launch --gpu rtx-3060 --template ollama \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Ollama · RTX 3060 · $0.042/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Ollama template

- **Image**: powergpu/ollama
- **CUDA**: inherits host driver (any 12.x / 13.x)
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [Oobabooga Text Gen on a cloud GPU](https://powergpu.ai/templates/oobabooga-text-gen) — The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions.
- [All 37 templates](https://powergpu.ai/templates)

## Ollama on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Ollama on a cloud GPU?**

Only the GPU price — the template is free. From $0.021/hr on an interruptible RTX 3060, $0.042/hr on-demand on a RTX 3060. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Ollama take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How do I keep pulled models between instances?**

Mount a volume on /root/.ollama. Models download once and every future instance in the region starts with the library already present.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/ollama · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/open-webui-ollama -->

---
title: "Run Open WebUI with Ollama on a cloud GPU from $0.163/hr | PowerGPU"
description: "Open WebUI (Ollama) on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/open-webui-ollama
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat · CUDA 12.9

# Run Open WebUI with Ollama on a cloud GPU, in 30 seconds

A full chat UI over Ollama — conversations, RAG and model management in the browser. Open WebUI on top of Ollama is a private ChatGPT-style interface: multi-user chat, document RAG, model switching and prompt libraries, all served from your own GPU behind TLS. Teams rent it as an internal assistant that never sends data to a third party. From **$0.163** /hr on an interruptible RTX 4090.

Open WebUI (Ollama)

powergpu/openwebui

(CUDA 12.9) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Open WebUI (Ollama) on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Open WebUI (Ollama)

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB serves a 14B assistant to a small team. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for 32B 4-bit models with more concurrent users. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | 48 GB and passive cooling for an always-on team deployment. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Open WebUI (Ollama) from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Open WebUI (Ollama)* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "open-webui-ollama". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — open-webui-ollama*

```
$ powergpu launch --gpu rtx-4090 --template open-webui-ollama \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Open WebUI (Ollama) · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Open WebUI (Ollama) template

- **Image**: powergpu/openwebui
- **CUDA**: CUDA 12.9
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Oobabooga Text Gen on a cloud GPU](https://powergpu.ai/templates/oobabooga-text-gen) — The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions.
- [All 37 templates](https://powergpu.ai/templates)

## Open WebUI (Ollama) on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Open WebUI (Ollama) on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Open WebUI (Ollama) take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can several people use it at once?**

Yes — Open WebUI has accounts and roles, and Ollama queues concurrent requests. For heavy concurrency, pair it with the vLLM template as the backend instead.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/open-webui-ollama · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/oobabooga-text-gen -->

---
title: "Run text-generation-webui on a cloud GPU from $0.054/hr | PowerGPU"
description: "Oobabooga Text Gen on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.054/hr on RTX 3090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/oobabooga-text-gen
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat · CUDA 12.4

# Run text-generation-webui on a cloud GPU, in 30 seconds

The text-generation WebUI & API — load GPTQ, EXL2 or GGUF with extensions. The text-generation-webui (oobabooga) loads GPTQ, EXL2, AWQ and GGUF models with a rich chat UI, character cards, extensions and an API. It is the tinkerer's choice for testing quantization formats and sampling settings side by side. From **$0.054** /hr on an interruptible RTX 3090.

Oobabooga Text Gen

powergpu/oobabooga

(CUDA 12.4) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Oobabooga Text Gen on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Oobabooga Text Gen

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB: EXL2 13B–32B models with fast ExLlama kernels. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | Same VRAM at a lower rate for slower experiments. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Best) | [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | $0.281 | $0.140 | 48 GB for 70B EXL2 on one card. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a6000) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Oobabooga Text Gen from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Oobabooga Text Gen* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "oobabooga-text-gen". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — oobabooga-text-gen*

```
$ powergpu launch --gpu rtx-4090 --template oobabooga-text-gen \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Oobabooga Text Gen · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Oobabooga Text Gen template

- **Image**: powergpu/oobabooga
- **CUDA**: CUDA 12.4
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [All 37 templates](https://powergpu.ai/templates)

## Oobabooga Text Gen on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Oobabooga Text Gen on a cloud GPU?**

Only the GPU price — the template is free. From $0.054/hr on an interruptible RTX 3090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Oobabooga Text Gen take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Which loaders are included?**

Transformers, ExLlamaV2, llama.cpp, AutoGPTQ and AWQ — pick per model in the UI. The OpenAI-compatible API extension is enabled by default.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/oobabooga-text-gen · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/huggingface-tgi -->

---
title: "Run Text Generation Inference on a cloud GPU | PowerGPU"
description: "HuggingFace TGI on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.219/hr on RTX 5090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/huggingface-tgi
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat

# Run Text Generation Inference on a cloud GPU, in 30 seconds

Hugging Face Text Generation Inference — battle-tested production serving with an API. Hugging Face's TGI is a battle-tested serving stack: Flash Attention, quantization (bitsandbytes, GPTQ, AWQ), token streaming and a Messages API. The template runs the official ghcr.io image with MODEL_ID as an environment variable. From **$0.219** /hr on an interruptible RTX 5090.

HuggingFace TGI

ghcr.io/huggingface/text-generation-inference

(SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: HuggingFace TGI on a RTX 5090 is $0.439/hr on-demand, $0.219/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for HuggingFace TGI

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | Fast 32 GB card for 7B–14B endpoints. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | 48 GB for 32B-class serving with datacenter reliability. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| (Best) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB | $1.428 | $0.714 | Multi-GPU sharding for 70B+ with the highest throughput. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-sxm) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy HuggingFace TGI from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *HuggingFace TGI* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "huggingface-tgi". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — huggingface-tgi*

```
$ powergpu launch --gpu rtx-5090 --template huggingface-tgi \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# HuggingFace TGI · RTX 5090 · $0.439/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the HuggingFace TGI template

- **Image**: ghcr.io/huggingface/text-generation-inference
- **CUDA**: inherits host driver (any 12.x / 13.x)
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [All 37 templates](https://powergpu.ai/templates)

## HuggingFace TGI on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run HuggingFace TGI on a cloud GPU?**

Only the GPU price — the template is free. From $0.219/hr on an interruptible RTX 5090, $0.439/hr on-demand on a RTX 5090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does HuggingFace TGI take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**TGI or vLLM?**

Both are excellent. vLLM usually leads raw throughput; TGI integrates tightly with the Hugging Face Hub, Inference Endpoints conventions and the Messages API. Deploy both for ten minutes and benchmark — it costs cents.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/huggingface-tgi · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/langflow-ollama -->

---
title: "Run Langflow on a cloud GPU from $0.163/hr | PowerGPU"
description: "Langflow (Ollama) on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/langflow-ollama
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · LLM serving & chat · CUDA 12.6

# Run Langflow on a cloud GPU, in 30 seconds

Visually build LLM pipelines and agents on top of a local Ollama backend. Langflow is a visual builder for LLM pipelines and agents; this template backs it with a local Ollama so every node runs on your GPU. Drag a retriever, a prompt and a model together, export the flow as an API, keep the data on your instance. From **$0.163** /hr on an interruptible RTX 4090.

Langflow (Ollama)

powergpu/langflow

(CUDA 12.6) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Langflow (Ollama) on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Langflow (Ollama)

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB for 7B–14B agent backends during design. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for larger models behind the same flows. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | Always-on 48 GB card for flows served to a team. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Langflow (Ollama) from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Langflow (Ollama)* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "langflow-ollama". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — langflow-ollama*

```
$ powergpu launch --gpu rtx-4090 --template langflow-ollama \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Langflow (Ollama) · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Langflow (Ollama) template

- **Image**: powergpu/langflow
- **CUDA**: CUDA 12.6
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [LLM serving & chat](https://powergpu.ai/templates#llm)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other llm serving & chat templates

- [vLLM on a cloud GPU](https://powergpu.ai/templates/vllm) — Production LLM serving with PagedAttention — an OpenAI-compatible endpoint from any HF model.
- [vLLM Omni on a cloud GPU](https://powergpu.ai/templates/vllm-omni) — vLLM extended for multimodal models — vision and audio inputs on the same fast server.
- [SGLang on a cloud GPU](https://powergpu.ai/templates/sglang) — High-throughput serving with RadixAttention — excels at structured and agentic workloads.
- [Llama.cpp on a cloud GPU](https://powergpu.ai/templates/llama-cpp) — GGUF inference with a built-in server — the lightest way to run quantized models.
- [Ollama on a cloud GPU](https://powergpu.ai/templates/ollama) — Pull and run quantized models in one command, with a clean REST API.
- [Open WebUI (Ollama) on a cloud GPU](https://powergpu.ai/templates/open-webui-ollama) — A full chat UI over Ollama — conversations, RAG and model management in the browser.
- [All 37 templates](https://powergpu.ai/templates)

## Langflow (Ollama) on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Langflow (Ollama) on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Langflow (Ollama) take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I call flows from my app?**

Every flow exposes a REST endpoint on the Langflow port; TLS is terminated for you. Use the API key you set at deploy.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/langflow-ollama · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/axolotl-fine-tuning -->

---
title: "Run Axolotl fine-tuning on a cloud GPU from $0.163/hr | PowerGPU"
description: "Axolotl — Fine Tuning on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/axolotl-fine-tuning
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Training & fine-tuning · CUDA 12.6

# Run Axolotl fine-tuning on a cloud GPU, in 30 seconds

Fine-tune Llama, Qwen or Mistral from one YAML — LoRA, QLoRA and multi-GPU FSDP. Axolotl fine-tunes Llama, Qwen, Mistral and friends from one YAML: LoRA, QLoRA, full-parameter, DeepSpeed and FSDP, with dataset formats handled for you. The template is what our QLoRA walkthrough is written against — checkpoints go to a volume, the trainer instance is disposable. From **$0.163** /hr on an interruptible RTX 4090.

Axolotl — Fine Tuning

axolotlai/axolotl-cloud

(CUDA 12.6) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Axolotl — Fine Tuning on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Axolotl — Fine Tuning

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | QLoRA 8B–13B on 24 GB for a couple of dollars. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | 80 GB | $0.560 | $0.280 | 80 GB with NVLink for full fine-tunes and 8× FSDP. | [Deploy](https://cloud.powergpu.ai/?gpu=a100-sxm4) |
| (Best) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB + FP8 — full fine-tunes in roughly half the A100 wall-clock. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Axolotl — Fine Tuning from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Axolotl — Fine Tuning* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "axolotl-fine-tuning". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — axolotl-fine-tuning*

```
$ powergpu launch --gpu rtx-4090 --template axolotl-fine-tuning \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Axolotl — Fine Tuning · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Axolotl — Fine Tuning template

- **Image**: axolotlai/axolotl-cloud
- **CUDA**: CUDA 12.6
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Training & fine-tuning](https://powergpu.ai/templates#tune)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other training & fine-tuning templates

- [Unsloth Studio on a cloud GPU](https://powergpu.ai/templates/unsloth-studio) — 2× faster, lower-VRAM fine-tuning — Unsloth kernels with a notebook workflow.
- [Kohya's GUI on a cloud GPU](https://powergpu.ai/templates/kohya-s-gui) — Train SDXL and Flux image LoRAs in a browser GUI — dataset in,.safetensors out.
- [Flux Gym on a cloud GPU](https://powergpu.ai/templates/flux-gym) — A dead-simple UI for training Flux LoRAs, built on the Kohya scripts.
- [Ostris AI Toolkit on a cloud GPU](https://powergpu.ai/templates/ostris-ai-toolkit) — The AI-Toolkit trainer for Flux and diffusion models, with a web UI.
- [All 37 templates](https://powergpu.ai/templates)

## Axolotl — Fine Tuning on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Axolotl — Fine Tuning on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Axolotl — Fine Tuning take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How long does a QLoRA run take?**

About 1.5 hours for 10k instruction pairs on Llama 3.1 8B on one RTX 4090; a 70B QLoRA on an 80 GB card is an overnight job. The guide lists exact configs and bills.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/axolotl-fine-tuning · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/unsloth-studio -->

---
title: "Run Unsloth fine-tuning on a cloud GPU from $0.037/hr | PowerGPU"
description: "Unsloth Studio on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.037/hr on RTX 4060 Ti. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/unsloth-studio
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Training & fine-tuning · CUDA 12.9

# Run Unsloth fine-tuning on a cloud GPU, in 30 seconds

2× faster, lower-VRAM fine-tuning — Unsloth kernels with a notebook workflow. Unsloth rewrites the attention and MLP kernels so LoRA/QLoRA fine-tuning runs about 2× faster with up to 70% less VRAM. Unsloth Studio adds a notebook workflow on top: load a base model, attach a dataset, train, export GGUF or merged weights. From **$0.037** /hr on an interruptible RTX 4060 Ti.

Unsloth Studio

powergpu/unsloth-studio

(CUDA 12.9) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Unsloth Studio on a RTX 4060 Ti is $0.074/hr on-demand, $0.037/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Unsloth Studio

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4060 Ti](https://powergpu.ai/gpu/rtx-4060-ti) | 8 GB | $0.074 | $0.037 | 16 GB is enough for 8B QLoRA with Unsloth's memory savings. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4060-ti) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB handles 14B–32B QLoRA quickly. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7 for the fastest single-card fine-tunes. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Unsloth Studio from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Unsloth Studio* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "unsloth-studio". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — unsloth-studio*

```
$ powergpu launch --gpu rtx-4060-ti --template unsloth-studio \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Unsloth Studio · RTX 4060 Ti · $0.074/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Unsloth Studio template

- **Image**: powergpu/unsloth-studio
- **CUDA**: CUDA 12.9
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Training & fine-tuning](https://powergpu.ai/templates#tune)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other training & fine-tuning templates

- [Axolotl — Fine Tuning on a cloud GPU](https://powergpu.ai/templates/axolotl-fine-tuning) — Fine-tune Llama, Qwen or Mistral from one YAML — LoRA, QLoRA and multi-GPU FSDP.
- [Kohya's GUI on a cloud GPU](https://powergpu.ai/templates/kohya-s-gui) — Train SDXL and Flux image LoRAs in a browser GUI — dataset in,.safetensors out.
- [Flux Gym on a cloud GPU](https://powergpu.ai/templates/flux-gym) — A dead-simple UI for training Flux LoRAs, built on the Kohya scripts.
- [Ostris AI Toolkit on a cloud GPU](https://powergpu.ai/templates/ostris-ai-toolkit) — The AI-Toolkit trainer for Flux and diffusion models, with a web UI.
- [All 37 templates](https://powergpu.ai/templates)

## Unsloth Studio on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Unsloth Studio on a cloud GPU?**

Only the GPU price — the template is free. From $0.037/hr on an interruptible RTX 4060 Ti, $0.074/hr on-demand on a RTX 4060 Ti. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Unsloth Studio take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Which models does Unsloth support?**

Llama 3.x, Qwen 2.5/3, Gemma, Mistral, Phi and most decoder-only architectures on Hugging Face, plus vision-language variants in recent releases.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/unsloth-studio · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/kohya-s-gui -->

---
title: "Run Kohya LoRA training on a cloud GPU from $0.054/hr | PowerGPU"
description: "Kohya's GUI on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.054/hr on RTX 3090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/kohya-s-gui
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Training & fine-tuning · CUDA 12.6

# Run Kohya LoRA training on a cloud GPU, in 30 seconds

Train SDXL and Flux image LoRAs in a browser GUI — dataset in,.safetensors out. Kohya's GUI is the standard trainer for Stable Diffusion, SDXL and Flux LoRAs: dataset prep, captioning helpers, bucketed resolutions and a browser UI over the sd-scripts. Twenty to forty images in, a.safetensors LoRA out, usually within an hour on a 4090. From **$0.054** /hr on an interruptible RTX 3090.

Kohya's GUI

powergpu/kohyas-gui

(CUDA 12.6) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Kohya's GUI on a RTX 3090 is $0.108/hr on-demand, $0.054/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Kohya's GUI

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | 24 GB at the lowest rate for SDXL LoRAs. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | The community default: SDXL and Flux LoRAs in 30–60 minutes. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for Flux full-precision training and larger batch sizes. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Kohya's GUI from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Kohya's GUI* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "kohya-s-gui". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — kohya-s-gui*

```
$ powergpu launch --gpu rtx-3090 --template kohya-s-gui \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Kohya's GUI · RTX 3090 · $0.108/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Kohya's GUI template

- **Image**: powergpu/kohyas-gui
- **CUDA**: CUDA 12.6
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Training & fine-tuning](https://powergpu.ai/templates#tune)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other training & fine-tuning templates

- [Axolotl — Fine Tuning on a cloud GPU](https://powergpu.ai/templates/axolotl-fine-tuning) — Fine-tune Llama, Qwen or Mistral from one YAML — LoRA, QLoRA and multi-GPU FSDP.
- [Unsloth Studio on a cloud GPU](https://powergpu.ai/templates/unsloth-studio) — 2× faster, lower-VRAM fine-tuning — Unsloth kernels with a notebook workflow.
- [Flux Gym on a cloud GPU](https://powergpu.ai/templates/flux-gym) — A dead-simple UI for training Flux LoRAs, built on the Kohya scripts.
- [Ostris AI Toolkit on a cloud GPU](https://powergpu.ai/templates/ostris-ai-toolkit) — The AI-Toolkit trainer for Flux and diffusion models, with a web UI.
- [All 37 templates](https://powergpu.ai/templates)

## Kohya's GUI on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Kohya's GUI on a cloud GPU?**

Only the GPU price — the template is free. From $0.054/hr on an interruptible RTX 3090, $0.108/hr on-demand on a RTX 3090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Kohya's GUI take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How much does a style LoRA cost to train?**

A typical 30-image SDXL LoRA at 1,500 steps takes 30–45 minutes on an RTX 4090 — well under a dollar on interruptible pricing.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/kohya-s-gui · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/flux-gym -->

---
title: "Run Flux LoRA training on a cloud GPU from $0.163/hr | PowerGPU"
description: "Flux Gym on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/flux-gym
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Training & fine-tuning

# Run Flux LoRA training on a cloud GPU, in 30 seconds

A dead-simple UI for training Flux LoRAs, built on the Kohya scripts. Flux Gym is the simplest UI for training Flux LoRAs, built on the Kohya scripts with sensible defaults and low-VRAM modes. Upload images, write captions, pick a VRAM profile, press train. From **$0.163** /hr on an interruptible RTX 4090.

Flux Gym

powergpu/fluxgym

(ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Flux Gym on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Flux Gym

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB: the 20 GB profile trains a Flux dev LoRA in about an hour. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for faster runs and higher resolutions. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | 96 GB for full-precision Flux with big batches. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Flux Gym from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Flux Gym* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "flux-gym". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — flux-gym*

```
$ powergpu launch --gpu rtx-4090 --template flux-gym \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Flux Gym · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Flux Gym template

- **Image**: powergpu/fluxgym
- **CUDA**: inherits host driver (any 12.x / 13.x)
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Training & fine-tuning](https://powergpu.ai/templates#tune)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other training & fine-tuning templates

- [Axolotl — Fine Tuning on a cloud GPU](https://powergpu.ai/templates/axolotl-fine-tuning) — Fine-tune Llama, Qwen or Mistral from one YAML — LoRA, QLoRA and multi-GPU FSDP.
- [Unsloth Studio on a cloud GPU](https://powergpu.ai/templates/unsloth-studio) — 2× faster, lower-VRAM fine-tuning — Unsloth kernels with a notebook workflow.
- [Kohya's GUI on a cloud GPU](https://powergpu.ai/templates/kohya-s-gui) — Train SDXL and Flux image LoRAs in a browser GUI — dataset in,.safetensors out.
- [Ostris AI Toolkit on a cloud GPU](https://powergpu.ai/templates/ostris-ai-toolkit) — The AI-Toolkit trainer for Flux and diffusion models, with a web UI.
- [All 37 templates](https://powergpu.ai/templates)

## Flux Gym on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Flux Gym on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Flux Gym take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Does Flux training fit in 16 GB?**

Flux Gym's 12 GB and 16 GB profiles use quantized bases and gradient checkpointing. They work on a 16 GB card at reduced speed; 24 GB is the comfortable minimum.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/flux-gym · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/ostris-ai-toolkit -->

---
title: "Run AI Toolkit (Ostris) on a cloud GPU from $0.163/hr | PowerGPU"
description: "Ostris AI Toolkit on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/ostris-ai-toolkit
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Training & fine-tuning · CUDA 12.9

# Run AI Toolkit (Ostris) on a cloud GPU, in 30 seconds

The AI-Toolkit trainer for Flux and diffusion models, with a web UI. Ostris' AI Toolkit is the trainer of choice for Flux and newer diffusion models: LoRA training with a clean YAML, a web UI for monitoring, and support for the latest model releases before other tools catch up. From **$0.163** /hr on an interruptible RTX 4090.

Ostris AI Toolkit

powergpu/ostris-ai-toolkit

(CUDA 12.9) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Ostris AI Toolkit on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Ostris AI Toolkit

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Runs Flux dev LoRA training with the low-VRAM flag. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB of headroom for larger resolutions. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | 96 GB for full fine-tunes of diffusion models. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Ostris AI Toolkit from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Ostris AI Toolkit* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "ostris-ai-toolkit". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — ostris-ai-toolkit*

```
$ powergpu launch --gpu rtx-4090 --template ostris-ai-toolkit \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Ostris AI Toolkit · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Ostris AI Toolkit template

- **Image**: powergpu/ostris-ai-toolkit
- **CUDA**: CUDA 12.9
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Training & fine-tuning](https://powergpu.ai/templates#tune)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other training & fine-tuning templates

- [Axolotl — Fine Tuning on a cloud GPU](https://powergpu.ai/templates/axolotl-fine-tuning) — Fine-tune Llama, Qwen or Mistral from one YAML — LoRA, QLoRA and multi-GPU FSDP.
- [Unsloth Studio on a cloud GPU](https://powergpu.ai/templates/unsloth-studio) — 2× faster, lower-VRAM fine-tuning — Unsloth kernels with a notebook workflow.
- [Kohya's GUI on a cloud GPU](https://powergpu.ai/templates/kohya-s-gui) — Train SDXL and Flux image LoRAs in a browser GUI — dataset in,.safetensors out.
- [Flux Gym on a cloud GPU](https://powergpu.ai/templates/flux-gym) — A dead-simple UI for training Flux LoRAs, built on the Kohya scripts.
- [All 37 templates](https://powergpu.ai/templates)

## Ostris AI Toolkit on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Ostris AI Toolkit on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Ostris AI Toolkit take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I train on a mounted dataset volume?**

Yes — point the config's dataset path at the volume mount; outputs and samples land on the same volume so nothing is lost when the trainer instance is destroyed.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/ostris-ai-toolkit · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/comfyui -->

---
title: "Run ComfyUI on a cloud GPU from $0.054/hr | PowerGPU"
description: "ComfyUI on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.054/hr on RTX 3090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/comfyui
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Image generation · CUDA 13

# Run ComfyUI on a cloud GPU, in 30 seconds

Node-based image and video workflows with ComfyUI-Manager pre-installed. ComfyUI is the node-based workflow engine for Stable Diffusion, SDXL, Flux and today's video models. The template ships with ComfyUI-Manager for one-click custom nodes, exposes the UI on a TLS-terminated port and mounts a model folder you can put on a volume. Every workflow also exports as JSON and runs headless through the API. From **$0.054** /hr on an interruptible RTX 3090.

ComfyUI

powergpu/comfyui

(CUDA 13) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: ComfyUI on a RTX 3090 is $0.108/hr on-demand, $0.054/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for ComfyUI

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | 24 GB at the lowest price — the batch-farm workhorse for SDXL. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | The community default: Flux dev images in about 2 seconds each. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7 for full-precision Flux, video nodes and heavy upscale chains. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy ComfyUI from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *ComfyUI* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "comfyui". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — comfyui*

```
$ powergpu launch --gpu rtx-3090 --template comfyui \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# ComfyUI · RTX 3090 · $0.108/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the ComfyUI template

- **Image**: powergpu/comfyui
- **CUDA**: CUDA 13
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Image generation](https://powergpu.ai/templates#image)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other image generation templates

- [InvokeAI on a cloud GPU](https://powergpu.ai/templates/invokeai) — A polished Stable Diffusion studio — canvas, layers and workflow nodes.
- [SD WebUI Forge on a cloud GPU](https://powergpu.ai/templates/sd-webui-forge) — The Forge fork of the SD WebUI — faster attention, lower VRAM, same extensions.
- [SD WebUI A1111 on a cloud GPU](https://powergpu.ai/templates/sd-webui-a1111) — The classic AUTOMATIC1111 Stable Diffusion WebUI with the full extension ecosystem.
- [Fooocus on a cloud GPU](https://powergpu.ai/templates/fooocus) — Prompt-and-go image generation — Midjourney-style simplicity on SDXL.
- [SwarmUI on a cloud GPU](https://powergpu.ai/templates/swarmui) — A ComfyUI-backed UI that scales generation across multiple GPUs.
- [All 37 templates](https://powergpu.ai/templates)

## ComfyUI on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run ComfyUI on a cloud GPU?**

Only the GPU price — the template is free. From $0.054/hr on an interruptible RTX 3090, $0.108/hr on-demand on a RTX 3090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does ComfyUI take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Where do I put my checkpoints and LoRAs?**

Mount a volume on /workspace/ComfyUI/models. Checkpoints, LoRAs, VAEs and ControlNets persist across instances and load in seconds from NVMe instead of downloading every time.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/comfyui · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/invokeai -->

---
title: "Run InvokeAI on a cloud GPU from $0.037/hr | PowerGPU"
description: "InvokeAI on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.037/hr on RTX 4070. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/invokeai
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Image generation · CUDA 12.9

# Run InvokeAI on a cloud GPU, in 30 seconds

A polished Stable Diffusion studio — canvas, layers and workflow nodes. InvokeAI is the polished studio for image generation: an infinite canvas, layers, inpainting, workflow nodes and a model manager that understands SD 1.5, SDXL and Flux. It suits illustrators and teams who want a designed tool rather than a graph editor. From **$0.037** /hr on an interruptible RTX 4070.

InvokeAI

powergpu/invokeai

(CUDA 12.9) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: InvokeAI on a RTX 4070 is $0.075/hr on-demand, $0.037/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for InvokeAI

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | $0.075 | $0.037 | 12 GB runs SD 1.5 and SDXL with FP8 comfortably at low cost. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB for Flux and large canvases. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for multi-model pipelines and big batch renders. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy InvokeAI from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *InvokeAI* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "invokeai". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — invokeai*

```
$ powergpu launch --gpu rtx-4070 --template invokeai \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# InvokeAI · RTX 4070 · $0.075/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the InvokeAI template

- **Image**: powergpu/invokeai
- **CUDA**: CUDA 12.9
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Image generation](https://powergpu.ai/templates#image)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other image generation templates

- [ComfyUI on a cloud GPU](https://powergpu.ai/templates/comfyui) — Node-based image and video workflows with ComfyUI-Manager pre-installed.
- [SD WebUI Forge on a cloud GPU](https://powergpu.ai/templates/sd-webui-forge) — The Forge fork of the SD WebUI — faster attention, lower VRAM, same extensions.
- [SD WebUI A1111 on a cloud GPU](https://powergpu.ai/templates/sd-webui-a1111) — The classic AUTOMATIC1111 Stable Diffusion WebUI with the full extension ecosystem.
- [Fooocus on a cloud GPU](https://powergpu.ai/templates/fooocus) — Prompt-and-go image generation — Midjourney-style simplicity on SDXL.
- [SwarmUI on a cloud GPU](https://powergpu.ai/templates/swarmui) — A ComfyUI-backed UI that scales generation across multiple GPUs.
- [All 37 templates](https://powergpu.ai/templates)

## InvokeAI on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run InvokeAI on a cloud GPU?**

Only the GPU price — the template is free. From $0.037/hr on an interruptible RTX 4070, $0.075/hr on-demand on a RTX 4070. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does InvokeAI take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I import my Automatic1111 models?**

Yes — point the model manager at a mounted volume containing your.safetensors files; InvokeAI scans and registers them.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/invokeai · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/sd-webui-forge -->

---
title: "Run SD WebUI Forge on a cloud GPU from $0.021/hr | PowerGPU"
description: "SD WebUI Forge on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.021/hr on RTX 3060. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/sd-webui-forge
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Image generation · CUDA 12.8

# Run SD WebUI Forge on a cloud GPU, in 30 seconds

The Forge fork of the SD WebUI — faster attention, lower VRAM, same extensions. Forge is the performance-focused fork of the Automatic1111 WebUI: faster attention, lower VRAM, native Flux support and the same extension ecosystem. If you know A1111, you know Forge — it is simply quicker and lighter. From **$0.021** /hr on an interruptible RTX 3060.

SD WebUI Forge

powergpu/sd-forge

(CUDA 12.8) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: SD WebUI Forge on a RTX 3060 is $0.042/hr on-demand, $0.021/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for SD WebUI Forge

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.042 | $0.021 | 12 GB is enough for SD 1.5 and SDXL at modest sizes. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB: Flux dev and heavy ControlNet stacks. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for large batches and high-resolution hires-fix passes. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy SD WebUI Forge from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *SD WebUI Forge* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "sd-webui-forge". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — sd-webui-forge*

```
$ powergpu launch --gpu rtx-3060 --template sd-webui-forge \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# SD WebUI Forge · RTX 3060 · $0.042/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the SD WebUI Forge template

- **Image**: powergpu/sd-forge
- **CUDA**: CUDA 12.8
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Image generation](https://powergpu.ai/templates#image)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other image generation templates

- [ComfyUI on a cloud GPU](https://powergpu.ai/templates/comfyui) — Node-based image and video workflows with ComfyUI-Manager pre-installed.
- [InvokeAI on a cloud GPU](https://powergpu.ai/templates/invokeai) — A polished Stable Diffusion studio — canvas, layers and workflow nodes.
- [SD WebUI A1111 on a cloud GPU](https://powergpu.ai/templates/sd-webui-a1111) — The classic AUTOMATIC1111 Stable Diffusion WebUI with the full extension ecosystem.
- [Fooocus on a cloud GPU](https://powergpu.ai/templates/fooocus) — Prompt-and-go image generation — Midjourney-style simplicity on SDXL.
- [SwarmUI on a cloud GPU](https://powergpu.ai/templates/swarmui) — A ComfyUI-backed UI that scales generation across multiple GPUs.
- [All 37 templates](https://powergpu.ai/templates)

## SD WebUI Forge on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run SD WebUI Forge on a cloud GPU?**

Only the GPU price — the template is free. From $0.021/hr on an interruptible RTX 3060, $0.042/hr on-demand on a RTX 3060. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does SD WebUI Forge take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Do A1111 extensions work in Forge?**

Most do — ControlNet, ADetailer, Regional Prompter and the popular ones are supported natively or via the extension index. A few heavily patched extensions need Forge-specific forks.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/sd-webui-forge · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/sd-webui-a1111 -->

---
title: "Run A1111 Stable Diffusion WebUI on a cloud GPU | PowerGPU"
description: "SD WebUI A1111 on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.021/hr on RTX 3060. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/sd-webui-a1111
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Image generation · CUDA 12.1

# Run A1111 Stable Diffusion WebUI on a cloud GPU, in 30 seconds

The classic AUTOMATIC1111 Stable Diffusion WebUI with the full extension ecosystem. The classic AUTOMATIC1111 WebUI with the full extension ecosystem: txt2img, img2img, inpainting, ControlNet, scripts and the API. Rent it when a tutorial or a client workflow was written for A1111 and you want the exact same buttons. From **$0.021** /hr on an interruptible RTX 3060.

SD WebUI A1111

powergpu/a1111

(CUDA 12.1) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: SD WebUI A1111 on a RTX 3060 is $0.042/hr on-demand, $0.021/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for SD WebUI A1111

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.042 | $0.021 | SD 1.5 and SDXL at the lowest price on the sheet. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| (Better) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | 24 GB for SDXL with ControlNet and big batches. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Best) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Fastest samplers per dollar for interactive sessions. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy SD WebUI A1111 from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *SD WebUI A1111* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "sd-webui-a1111". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — sd-webui-a1111*

```
$ powergpu launch --gpu rtx-3060 --template sd-webui-a1111 \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# SD WebUI A1111 · RTX 3060 · $0.042/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the SD WebUI A1111 template

- **Image**: powergpu/a1111
- **CUDA**: CUDA 12.1
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Image generation](https://powergpu.ai/templates#image)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other image generation templates

- [ComfyUI on a cloud GPU](https://powergpu.ai/templates/comfyui) — Node-based image and video workflows with ComfyUI-Manager pre-installed.
- [InvokeAI on a cloud GPU](https://powergpu.ai/templates/invokeai) — A polished Stable Diffusion studio — canvas, layers and workflow nodes.
- [SD WebUI Forge on a cloud GPU](https://powergpu.ai/templates/sd-webui-forge) — The Forge fork of the SD WebUI — faster attention, lower VRAM, same extensions.
- [Fooocus on a cloud GPU](https://powergpu.ai/templates/fooocus) — Prompt-and-go image generation — Midjourney-style simplicity on SDXL.
- [SwarmUI on a cloud GPU](https://powergpu.ai/templates/swarmui) — A ComfyUI-backed UI that scales generation across multiple GPUs.
- [All 37 templates](https://powergpu.ai/templates)

## SD WebUI A1111 on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run SD WebUI A1111 on a cloud GPU?**

Only the GPU price — the template is free. From $0.021/hr on an interruptible RTX 3060, $0.042/hr on-demand on a RTX 3060. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does SD WebUI A1111 take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Is the API enabled?**

Yes — the template starts A1111 with --api and --listen; the /sdapi/v1 endpoints are reachable on the mapped port with TLS.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/sd-webui-a1111 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/fooocus -->

---
title: "Run Fooocus on a cloud GPU from $0.021/hr | PowerGPU"
description: "Fooocus on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.021/hr on RTX 3060. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/fooocus
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Image generation · CUDA 12.1

# Run Fooocus on a cloud GPU, in 30 seconds

Prompt-and-go image generation — Midjourney-style simplicity on SDXL. Fooocus wraps SDXL in a Midjourney-style interface: type a prompt, get great images, no sampler settings to learn. It is the template to hand to non-technical teammates — and per-second billing means an afternoon of prompting costs less than a coffee. From **$0.021** /hr on an interruptible RTX 3060.

Fooocus

powergpu/fooocus

(CUDA 12.1) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Fooocus on a RTX 3060 is $0.042/hr on-demand, $0.021/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Fooocus

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.042 | $0.021 | 12 GB runs Fooocus with its built-in low-VRAM mode. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| (Better) | [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | $0.075 | $0.037 | 12 GB GDDR6X and Ada speed at a low rate. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070) |
| (Best) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB for fast, large-batch generation. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Fooocus from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Fooocus* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "fooocus". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — fooocus*

```
$ powergpu launch --gpu rtx-3060 --template fooocus \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Fooocus · RTX 3060 · $0.042/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Fooocus template

- **Image**: powergpu/fooocus
- **CUDA**: CUDA 12.1
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Image generation](https://powergpu.ai/templates#image)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other image generation templates

- [ComfyUI on a cloud GPU](https://powergpu.ai/templates/comfyui) — Node-based image and video workflows with ComfyUI-Manager pre-installed.
- [InvokeAI on a cloud GPU](https://powergpu.ai/templates/invokeai) — A polished Stable Diffusion studio — canvas, layers and workflow nodes.
- [SD WebUI Forge on a cloud GPU](https://powergpu.ai/templates/sd-webui-forge) — The Forge fork of the SD WebUI — faster attention, lower VRAM, same extensions.
- [SD WebUI A1111 on a cloud GPU](https://powergpu.ai/templates/sd-webui-a1111) — The classic AUTOMATIC1111 Stable Diffusion WebUI with the full extension ecosystem.
- [SwarmUI on a cloud GPU](https://powergpu.ai/templates/swarmui) — A ComfyUI-backed UI that scales generation across multiple GPUs.
- [All 37 templates](https://powergpu.ai/templates)

## Fooocus on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Fooocus on a cloud GPU?**

Only the GPU price — the template is free. From $0.021/hr on an interruptible RTX 3060, $0.042/hr on-demand on a RTX 3060. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Fooocus take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I add my own SDXL checkpoints and LoRAs?**

Drop them on a volume mounted at the models path; Fooocus lists them in the advanced panel on the next start.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/fooocus · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/swarmui -->

---
title: "Run SwarmUI on a cloud GPU from $0.163/hr | PowerGPU"
description: "SwarmUI on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/swarmui
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Image generation · CUDA 12.4

# Run SwarmUI on a cloud GPU, in 30 seconds

A ComfyUI-backed UI that scales generation across multiple GPUs. SwarmUI (formerly StableSwarmUI) is a friendly front end over a ComfyUI backend that can drive several GPUs at once. One machine with 4× or 8× cards becomes a small generation farm with a single UI and API. From **$0.163** /hr on an interruptible RTX 4090.

SwarmUI

powergpu/swarmui

(CUDA 12.4) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: SwarmUI on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for SwarmUI

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Multi-GPU 4090 machines are the natural Swarm host. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB per card for Flux and video pipelines across the swarm. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | 48 GB datacenter cards for an always-on generation service. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy SwarmUI from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *SwarmUI* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "swarmui". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — swarmui*

```
$ powergpu launch --gpu rtx-4090 --template swarmui \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# SwarmUI · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the SwarmUI template

- **Image**: powergpu/swarmui
- **CUDA**: CUDA 12.4
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Image generation](https://powergpu.ai/templates#image)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other image generation templates

- [ComfyUI on a cloud GPU](https://powergpu.ai/templates/comfyui) — Node-based image and video workflows with ComfyUI-Manager pre-installed.
- [InvokeAI on a cloud GPU](https://powergpu.ai/templates/invokeai) — A polished Stable Diffusion studio — canvas, layers and workflow nodes.
- [SD WebUI Forge on a cloud GPU](https://powergpu.ai/templates/sd-webui-forge) — The Forge fork of the SD WebUI — faster attention, lower VRAM, same extensions.
- [SD WebUI A1111 on a cloud GPU](https://powergpu.ai/templates/sd-webui-a1111) — The classic AUTOMATIC1111 Stable Diffusion WebUI with the full extension ecosystem.
- [Fooocus on a cloud GPU](https://powergpu.ai/templates/fooocus) — Prompt-and-go image generation — Midjourney-style simplicity on SDXL.
- [All 37 templates](https://powergpu.ai/templates)

## SwarmUI on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run SwarmUI on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does SwarmUI take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Does it really use all GPUs on the machine?**

Yes — Swarm launches one ComfyUI backend per GPU and distributes queued generations across them automatically.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/swarmui · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/wan2gp -->

---
title: "Run Wan video generation on a cloud GPU from $0.163/hr | PowerGPU"
description: "Wan2GP on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/wan2gp
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Video generation · CUDA 12.9

# Run Wan video generation on a cloud GPU, in 30 seconds

Run Wan and other video-generation models on modest VRAM, with a simple UI. Wan2GP ("Wan for the GPU poor") runs Wan 2.x, Hunyuan and other text- and image-to-video models with aggressive offloading and quantization, so 720p clips are possible on 16–24 GB cards. A simple web UI hides the memory tricks. From **$0.163** /hr on an interruptible RTX 4090.

Wan2GP

powergpu/wan2gp

(CUDA 12.9) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Wan2GP on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Wan2GP

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB: Wan 2.1 14B at 480p–720p with offloading. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB GDDR7 — noticeably faster clips and fewer offload stalls. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB for full-precision video models without offloading. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Wan2GP from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Wan2GP* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "wan2gp". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — wan2gp*

```
$ powergpu launch --gpu rtx-4090 --template wan2gp \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Wan2GP · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Wan2GP template

- **Image**: powergpu/wan2gp
- **CUDA**: CUDA 12.9
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Video generation](https://powergpu.ai/templates#video)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other video generation templates

- [Open-Sora on a cloud GPU](https://powergpu.ai/templates/open-sora) — The open text-to-video model, ready to generate and fine-tune.
- [All 37 templates](https://powergpu.ai/templates)

## Wan2GP on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Wan2GP on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Wan2GP take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How long does a 5-second clip take?**

Roughly 8–15 minutes at 720p on an RTX 4090 with offloading, 4–8 minutes on an 80 GB H100 without it. Batch overnight on interruptible pricing to halve the cost per clip.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/wan2gp · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/open-sora -->

---
title: "Run Open-Sora on a cloud GPU from $0.219/hr | PowerGPU"
description: "Open-Sora on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.219/hr on RTX 5090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/open-sora
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Video generation · CUDA 12.1

# Run Open-Sora on a cloud GPU, in 30 seconds

The open text-to-video model, ready to generate and fine-tune. Open-Sora is the open text-to-video model and training codebase. The template ships inference scripts and the Gradio demo; the research-grade training pipeline is there for teams fine-tuning on their own footage. From **$0.219** /hr on an interruptible RTX 5090.

Open-Sora

powergpu/open-sora

(CUDA 12.1) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Open-Sora on a RTX 5090 is $0.439/hr on-demand, $0.219/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Open-Sora

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for inference at moderate resolutions. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Better) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB HBM for full-resolution generation. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |
| (Best) | [H200](https://powergpu.ai/gpu/h200) | 141 GB | $2.791 | $1.395 | 141 GB for training and long clips. | [Deploy](https://cloud.powergpu.ai/?gpu=h200) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Open-Sora from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Open-Sora* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "open-sora". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — open-sora*

```
$ powergpu launch --gpu rtx-5090 --template open-sora \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Open-Sora · RTX 5090 · $0.439/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Open-Sora template

- **Image**: powergpu/open-sora
- **CUDA**: CUDA 12.1
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Video generation](https://powergpu.ai/templates#video)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other video generation templates

- [Wan2GP on a cloud GPU](https://powergpu.ai/templates/wan2gp) — Run Wan and other video-generation models on modest VRAM, with a simple UI.
- [All 37 templates](https://powergpu.ai/templates)

## Open-Sora on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Open-Sora on a cloud GPU?**

Only the GPU price — the template is free. From $0.219/hr on an interruptible RTX 5090, $0.439/hr on-demand on a RTX 5090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Open-Sora take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I fine-tune Open-Sora here?**

Yes — the training scripts are included; put the dataset on a volume and use an 80 GB+ card or a multi-GPU machine.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/open-sora · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/whisper-webui-api -->

---
title: "Run Whisper transcription on a cloud GPU from $0.021/hr | PowerGPU"
description: "Whisper WebUI & API on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.021/hr on RTX 3060. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/whisper-webui-api
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Audio & speech · CUDA 12.6

# Run Whisper transcription on a cloud GPU, in 30 seconds

Batch-transcribe audio with faster-whisper behind a UI and a REST endpoint. faster-whisper behind a web UI and a REST endpoint: upload files or point it at a folder, get transcripts, subtitles and word timestamps in dozens of languages. A cheap card transcribes hours of audio per hour of GPU time. From **$0.021** /hr on an interruptible RTX 3060.

Whisper WebUI & API

powergpu/whisper

(CUDA 12.6) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Whisper WebUI & API on a RTX 3060 is $0.042/hr on-demand, $0.021/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Whisper WebUI & API

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.042 | $0.021 | 12 GB runs large-v3 with room to spare — the cheapest transcription box. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| (Better) | [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | $0.075 | $0.037 | Ada speed for batch jobs at a low rate. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070) |
| (Best) | [L4](https://powergpu.ai/gpu/l4) | 24 GB | $0.225 | $0.112 | 72 W datacenter card for an always-on transcription API. | [Deploy](https://cloud.powergpu.ai/?gpu=l4) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Whisper WebUI & API from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Whisper WebUI & API* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "whisper-webui-api". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — whisper-webui-api*

```
$ powergpu launch --gpu rtx-3060 --template whisper-webui-api \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Whisper WebUI & API · RTX 3060 · $0.042/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Whisper WebUI & API template

- **Image**: powergpu/whisper
- **CUDA**: CUDA 12.6
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Audio & speech](https://powergpu.ai/templates#audio)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other audio & speech templates

- [Voicebox TTS on a cloud GPU](https://powergpu.ai/templates/voicebox-tts) — Text-to-speech and voice cloning with a web interface and API.
- [ACE Step 1.5 on a cloud GPU](https://powergpu.ai/templates/ace-step-1-5) — The ACE-Step music generation model with a UI and API.
- [All 37 templates](https://powergpu.ai/templates)

## Whisper WebUI & API on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Whisper WebUI & API on a cloud GPU?**

Only the GPU price — the template is free. From $0.021/hr on an interruptible RTX 3060, $0.042/hr on-demand on a RTX 3060. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Whisper WebUI & API take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How fast is transcription?**

faster-whisper large-v3 transcribes roughly 20–40× real time on an RTX 4070 with batching — an hour of audio in about two minutes.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/whisper-webui-api · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/voicebox-tts -->

---
title: "Run voice cloning & TTS on a cloud GPU from $0.021/hr | PowerGPU"
description: "Voicebox TTS on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.021/hr on RTX 3060. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/voicebox-tts
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Audio & speech · CUDA 12.1

# Run voice cloning & TTS on a cloud GPU, in 30 seconds

Text-to-speech and voice cloning with a web interface and API. Voicebox bundles modern open TTS and voice-cloning models behind a web interface and an API: clone a voice from a short sample, synthesise long-form narration, export WAV. Rent a small card for interactive use or batch narrations overnight. From **$0.021** /hr on an interruptible RTX 3060.

Voicebox TTS

powergpu/voicebox

(CUDA 12.1) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Voicebox TTS on a RTX 3060 is $0.042/hr on-demand, $0.021/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Voicebox TTS

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | 12 GB | $0.042 | $0.021 | 12 GB is plenty for most TTS models. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3060) |
| (Better) | [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | $0.075 | $0.037 | Faster synthesis for batch narration. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070) |
| (Best) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | 24 GB for larger voice models and parallel requests. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Voicebox TTS from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Voicebox TTS* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "voicebox-tts". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — voicebox-tts*

```
$ powergpu launch --gpu rtx-3060 --template voicebox-tts \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Voicebox TTS · RTX 3060 · $0.042/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Voicebox TTS template

- **Image**: powergpu/voicebox
- **CUDA**: CUDA 12.1
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Audio & speech](https://powergpu.ai/templates#audio)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other audio & speech templates

- [Whisper WebUI & API on a cloud GPU](https://powergpu.ai/templates/whisper-webui-api) — Batch-transcribe audio with faster-whisper behind a UI and a REST endpoint.
- [ACE Step 1.5 on a cloud GPU](https://powergpu.ai/templates/ace-step-1-5) — The ACE-Step music generation model with a UI and API.
- [All 37 templates](https://powergpu.ai/templates)

## Voicebox TTS on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Voicebox TTS on a cloud GPU?**

Only the GPU price — the template is free. From $0.021/hr on an interruptible RTX 3060, $0.042/hr on-demand on a RTX 3060. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Voicebox TTS take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I use it as an API?**

Yes — the REST endpoint accepts text plus a reference sample and returns audio; it is reachable on the mapped port with TLS.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/voicebox-tts · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/ace-step-1-5 -->

---
title: "Run ACE-Step music generation on a cloud GPU | PowerGPU"
description: "ACE Step 1.5 on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.037/hr on RTX 4070. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/ace-step-1-5
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Audio & speech · CUDA 12.4

# Run ACE-Step music generation on a cloud GPU, in 30 seconds

The ACE-Step music generation model with a UI and API. ACE-Step generates full songs — vocals, instruments, structure — from text prompts and lyrics. The template exposes the Gradio UI and an API; a mid-range card produces a track in well under a minute. From **$0.037** /hr on an interruptible RTX 4070.

ACE Step 1.5

powergpu/acestep

(CUDA 12.4) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: ACE Step 1.5 on a RTX 4070 is $0.075/hr on-demand, $0.037/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for ACE Step 1.5

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4070](https://powergpu.ai/gpu/rtx-4070) | 12 GB | $0.075 | $0.037 | 12 GB at a low rate for experimentation. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4070) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Fast generation and room for higher-quality settings. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for batches and longer compositions. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy ACE Step 1.5 from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *ACE Step 1.5* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "ace-step-1-5". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — ace-step-1-5*

```
$ powergpu launch --gpu rtx-4070 --template ace-step-1-5 \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# ACE Step 1.5 · RTX 4070 · $0.075/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the ACE Step 1.5 template

- **Image**: powergpu/acestep
- **CUDA**: CUDA 12.4
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Audio & speech](https://powergpu.ai/templates#audio)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other audio & speech templates

- [Whisper WebUI & API on a cloud GPU](https://powergpu.ai/templates/whisper-webui-api) — Batch-transcribe audio with faster-whisper behind a UI and a REST endpoint.
- [Voicebox TTS on a cloud GPU](https://powergpu.ai/templates/voicebox-tts) — Text-to-speech and voice cloning with a web interface and API.
- [All 37 templates](https://powergpu.ai/templates)

## ACE Step 1.5 on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run ACE Step 1.5 on a cloud GPU?**

Only the GPU price — the template is free. From $0.037/hr on an interruptible RTX 4070, $0.075/hr on-demand on a RTX 4070. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does ACE Step 1.5 take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How long does a track take to generate?**

A 3-minute song renders in roughly 20–40 seconds on an RTX 4090.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/ace-step-1-5 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/linux-desktop -->

---
title: "Run a GPU Linux desktop in the cloud from $0.140/hr | PowerGPU"
description: "Linux Desktop on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.140/hr on RTX A6000. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/linux-desktop
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Desktops & VMs · CUDA 12.1

# Run a GPU Linux desktop in the cloud, in 30 seconds

A full GPU Linux desktop over the browser — XFCE, VNC/RDP, run GUI apps. A full XFCE desktop on a GPU, delivered to your browser or any VNC/RDP client. Run GUI tools — Blender, DaVinci Resolve, Unreal, scientific visualisers — on a rented card, with your files on a volume and the desktop gone the second you stop paying. From **$0.140** /hr on an interruptible RTX A6000.

Linux Desktop

powergpu/linux-desktop

(CUDA 12.1) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Linux Desktop on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Linux Desktop

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Fast interactive work at the best consumer price. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | 48 GB | $0.281 | $0.140 | 48 GB ECC with studio drivers for professional apps. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a6000) |
| (Best) | [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | 96 GB GDDR7 for the heaviest scenes and datasets. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Linux Desktop from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Linux Desktop* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "linux-desktop". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — linux-desktop*

```
$ powergpu launch --gpu rtx-4090 --template linux-desktop \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Linux Desktop · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Linux Desktop template

- **Image**: powergpu/linux-desktop
- **CUDA**: CUDA 12.1
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Desktops & VMs](https://powergpu.ai/templates#desktop)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other desktops & vms templates

- [Pinokio on a cloud GPU](https://powergpu.ai/templates/pinokio) — The Pinokio 1-click app browser on a desktop — install AI apps with a click.
- [Ubuntu 22.04 VM on a cloud GPU](https://powergpu.ai/templates/ubuntu-22-04-vm) — A full KVM virtual machine — your own kernel, root, systemd, any stack.
- [Ubuntu Desktop VM on a cloud GPU](https://powergpu.ai/templates/ubuntu-desktop-vm) — The Ubuntu VM with a GNOME desktop over VNC/RDP — a full graphical workstation.
- [All 37 templates](https://powergpu.ai/templates)

## Linux Desktop on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Linux Desktop on a cloud GPU?**

Only the GPU price — the template is free. From $0.140/hr on an interruptible RTX A6000, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Linux Desktop take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Is the desktop encrypted in transit?**

Browser access goes through the TLS-terminated port; native VNC/RDP should be tunnelled over SSH, which the template exposes as well.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/linux-desktop · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/pinokio -->

---
title: "Run Pinokio on a cloud GPU from $0.163/hr | PowerGPU"
description: "Pinokio on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/pinokio
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Desktops & VMs · CUDA 13

# Run Pinokio on a cloud GPU, in 30 seconds

The Pinokio 1-click app browser on a desktop — install AI apps with a click. Pinokio is the one-click installer for community AI apps — hundreds of scripts that set up ComfyUI, TTS tools, LLM UIs and research demos with a single button. On a rented GPU it becomes a disposable playground: install, try, destroy. From **$0.163** /hr on an interruptible RTX 4090.

Pinokio

powergpu/pinokio

(CUDA 13) (ARM) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Pinokio on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Pinokio

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Runs nearly every Pinokio script in 24 GB. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | 32 GB for the video and large-model scripts. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |
| (Best) | [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | 96 GB when you want several apps loaded at once. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Pinokio from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Pinokio* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "pinokio". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — pinokio*

```
$ powergpu launch --gpu rtx-4090 --template pinokio \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Pinokio · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Pinokio template

- **Image**: powergpu/pinokio
- **CUDA**: CUDA 13
- **Access**: also builds for ARM hosts · SSH shell · JupyterLab on a mapped port
- **Category**: [Desktops & VMs](https://powergpu.ai/templates#desktop)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other desktops & vms templates

- [Linux Desktop on a cloud GPU](https://powergpu.ai/templates/linux-desktop) — A full GPU Linux desktop over the browser — XFCE, VNC/RDP, run GUI apps.
- [Ubuntu 22.04 VM on a cloud GPU](https://powergpu.ai/templates/ubuntu-22-04-vm) — A full KVM virtual machine — your own kernel, root, systemd, any stack.
- [Ubuntu Desktop VM on a cloud GPU](https://powergpu.ai/templates/ubuntu-desktop-vm) — The Ubuntu VM with a GNOME desktop over VNC/RDP — a full graphical workstation.
- [All 37 templates](https://powergpu.ai/templates)

## Pinokio on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Pinokio on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Pinokio take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Do installed apps survive a restart?**

Mount a volume on the Pinokio home directory and every installed app, model and output persists across instances.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/pinokio · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/ubuntu-22-04-vm -->

---
title: "Run an Ubuntu GPU virtual machine from $0.163/hr | PowerGPU"
description: "Ubuntu 22.04 VM on a fixed-price cloud GPU: one-click template, boots in minutes, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/ubuntu-22-04-vm
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Desktops & VMs

# Run an Ubuntu GPU virtual machine, in minutes

A full KVM virtual machine — your own kernel, root, systemd, any stack. A real KVM virtual machine with the GPU passed through: Ubuntu 22.04, your own kernel, systemd, root, nested Docker, custom drivers. It is the template for software that refuses to live in a container — or for teams that simply want a server. From **$0.163** /hr on an interruptible RTX 4090.

Ubuntu 22.04 VM

powergpu/kvm:ubuntu-22.04

(SSH) (VM)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Boots in minutes

A full KVM virtual machine with the GPU passed through: your kernel, root, systemd.

### Template is free

You pay the GPU price only: Ubuntu 22.04 VM on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — SSH behind your own credentials.

## Best GPUs for Ubuntu 22.04 VM

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | A fast consumer card for a personal GPU server. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | Datacenter card for a long-running VM under sustained load. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |
| (Best) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB | $1.867 | $0.933 | 80 GB Hopper for a VM-based training or serving host. | [Deploy](https://cloud.powergpu.ai/?gpu=h100-pcie) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Ubuntu 22.04 VM from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Ubuntu 22.04 VM* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "ubuntu-22-04-vm". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — ubuntu-22-04-vm*

```
$ powergpu launch --gpu rtx-4090 --template ubuntu-22-04-vm \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (2m38s)
# Ubuntu 22.04 VM · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8000 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Ubuntu 22.04 VM template

- **Image**: powergpu/kvm:ubuntu-22.04
- **CUDA**: inherits host driver (any 12.x / 13.x)
- **Access**: SSH shell · full virtual machine (KVM)
- **Category**: [Desktops & VMs](https://powergpu.ai/templates#desktop)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other desktops & vms templates

- [Linux Desktop on a cloud GPU](https://powergpu.ai/templates/linux-desktop) — A full GPU Linux desktop over the browser — XFCE, VNC/RDP, run GUI apps.
- [Pinokio on a cloud GPU](https://powergpu.ai/templates/pinokio) — The Pinokio 1-click app browser on a desktop — install AI apps with a click.
- [Ubuntu Desktop VM on a cloud GPU](https://powergpu.ai/templates/ubuntu-desktop-vm) — The Ubuntu VM with a GNOME desktop over VNC/RDP — a full graphical workstation.
- [All 37 templates](https://powergpu.ai/templates)

## Ubuntu 22.04 VM on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Ubuntu 22.04 VM on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Ubuntu 22.04 VM take to start?**

Two to four minutes: this template is a full virtual machine that boots its own kernel. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How long does a VM take to boot?**

Two to four minutes for a fresh VM versus about 30 seconds for a container template; stop/start of an existing VM is faster.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/ubuntu-22-04-vm · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/ubuntu-desktop-vm -->

---
title: "Run an Ubuntu desktop GPU VM from $0.163/hr | PowerGPU"
description: "Ubuntu Desktop VM on a fixed-price cloud GPU: one-click template, boots in minutes, from $0.163/hr on RTX 4090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/ubuntu-desktop-vm
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Desktops & VMs

# Run an Ubuntu desktop GPU VM, in minutes

The Ubuntu VM with a GNOME desktop over VNC/RDP — a full graphical workstation. The Ubuntu VM with a GNOME desktop over VNC/RDP: a full graphical workstation with hardware GPU passthrough. Install anything — proprietary renderers, CAD, research tools with dongles on reserved machines — and work as if the card were under your desk. From **$0.163** /hr on an interruptible RTX 4090.

Ubuntu Desktop VM

powergpu/kvm:ubuntu-desktop

(SSH) (VM)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Boots in minutes

A full KVM virtual machine with the GPU passed through: your kernel, root, systemd.

### Template is free

You pay the GPU price only: Ubuntu Desktop VM on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — SSH behind your own credentials.

## Best GPUs for Ubuntu Desktop VM

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | Consumer speed for interactive creative work. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) | 48 GB | $0.467 | $0.233 | 48 GB ECC and studio drivers for CAD and rendering. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-6000ada) |
| (Best) | [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | 96 GB for production scenes and huge datasets. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-pro-6000-ws) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Ubuntu Desktop VM from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Ubuntu Desktop VM* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "ubuntu-desktop-vm". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — ubuntu-desktop-vm*

```
$ powergpu launch --gpu rtx-4090 --template ubuntu-desktop-vm \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (2m38s)
# Ubuntu Desktop VM · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8000 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Ubuntu Desktop VM template

- **Image**: powergpu/kvm:ubuntu-desktop
- **CUDA**: inherits host driver (any 12.x / 13.x)
- **Access**: SSH shell · full virtual machine (KVM)
- **Category**: [Desktops & VMs](https://powergpu.ai/templates#desktop)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other desktops & vms templates

- [Linux Desktop on a cloud GPU](https://powergpu.ai/templates/linux-desktop) — A full GPU Linux desktop over the browser — XFCE, VNC/RDP, run GUI apps.
- [Pinokio on a cloud GPU](https://powergpu.ai/templates/pinokio) — The Pinokio 1-click app browser on a desktop — install AI apps with a click.
- [Ubuntu 22.04 VM on a cloud GPU](https://powergpu.ai/templates/ubuntu-22-04-vm) — A full KVM virtual machine — your own kernel, root, systemd, any stack.
- [All 37 templates](https://powergpu.ai/templates)

## Ubuntu Desktop VM on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Ubuntu Desktop VM on a cloud GPU?**

Only the GPU price — the template is free. From $0.163/hr on an interruptible RTX 4090, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Ubuntu Desktop VM take to start?**

Two to four minutes: this template is a full virtual machine that boots its own kernel. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Can I run Windows instead?**

Containers are Linux-only, but VMs can boot a Windows image you licence yourself (BYOL) via the custom qcow2 import.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/ubuntu-desktop-vm · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/unreal-pixel-streaming -->

---
title: "Run Unreal Pixel Streaming on a cloud GPU from $0.080/hr | PowerGPU"
description: "Unreal Pixel Streaming on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.080/hr on RTX A5000. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/unreal-pixel-streaming
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Specialised · CUDA 12.8

# Run Unreal Pixel Streaming on a cloud GPU, in 30 seconds

Stream an Unreal Engine app to the browser — pixel streaming on a cloud GPU. Stream a packaged Unreal Engine application to any browser: the template runs the signalling server and your Linux build with hardware encode on the GPU. Demos, configurators and virtual showrooms scale by adding instances, not by shipping installers. From **$0.080** /hr on an interruptible RTX A5000.

Unreal Pixel Streaming

powergpu/unreal-pixel-streaming

(CUDA 12.8) (SSH) (Jupyter)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Unreal Pixel Streaming on a RTX 4090 is $0.327/hr on-demand, $0.163/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — Jupyter and SSH behind your own credentials.

## Best GPUs for Unreal Pixel Streaming

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | High frame rates with a strong NVENC at consumer prices. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Better) | [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) | 24 GB | $0.161 | $0.080 | Blower-cooled workstation card for long-running streams. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-a5000) |
| (Best) | [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | Datacenter Ada with dual encoders for many concurrent viewers. | [Deploy](https://cloud.powergpu.ai/?gpu=l40s) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Unreal Pixel Streaming from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Unreal Pixel Streaming* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "unreal-pixel-streaming". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — unreal-pixel-streaming*

```
$ powergpu launch --gpu rtx-4090 --template unreal-pixel-streaming \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Unreal Pixel Streaming · RTX 4090 · $0.327/hr · per second
# https://i-7a41c0e2.powergpu.ai:8888 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Unreal Pixel Streaming template

- **Image**: powergpu/unreal-pixel-streaming
- **CUDA**: CUDA 12.8
- **Access**: SSH shell · JupyterLab on a mapped port
- **Category**: [Specialised](https://powergpu.ai/templates#tools)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other specialised templates

- [Hashcat CUDA on a cloud GPU](https://powergpu.ai/templates/hashcat-cuda) — GPU password recovery for authorised security testing — CUDA-accelerated Hashcat.
- [All 37 templates](https://powergpu.ai/templates)

## Unreal Pixel Streaming on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Unreal Pixel Streaming on a cloud GPU?**

Only the GPU price — the template is free. From $0.080/hr on an interruptible RTX A5000, $0.327/hr on-demand on a RTX 4090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Unreal Pixel Streaming take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**How many viewers per GPU?**

One interactive session per GPU for a full 3D scene is the safe assumption; lighter scenes can share a card. Scale horizontally by launching more instances behind your matchmaker.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/unreal-pixel-streaming · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/templates/hashcat-cuda -->

---
title: "Run Hashcat on a cloud GPU from $0.054/hr | PowerGPU"
description: "Hashcat CUDA on a fixed-price cloud GPU: one-click template, running in ~30 s, from $0.054/hr on RTX 3090. Best GPUs, volumes, per-second billing."
url: https://powergpu.ai/templates/hashcat-cuda
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Template · Specialised

# Run Hashcat on a cloud GPU, in 30 seconds

GPU password recovery for authorised security testing — CUDA-accelerated Hashcat. CUDA-accelerated Hashcat for authorised password audits and recovery of your own credentials. Rent a fast consumer card for an afternoon instead of building a cracking rig — subject to the acceptable use policy, which permits security testing only on systems you own or are contracted to test. From **$0.054** /hr on an interruptible RTX 3090.

Hashcat CUDA

dizcza/docker-hashcat

(SSH)

Pinned image, rebuilt on upstream releases. Override env, ports and disk at deploy.

### Running in ~30 s

Pre-cached image, ports mapped, TLS terminated — the stack is working before you finish reading this.

### Template is free

You pay the GPU price only: Hashcat CUDA on a RTX 3090 is $0.108/hr on-demand, $0.054/hr interruptible, billed per second.

### Volumes for state

Models, datasets and outputs live on a $0.08/GB/mo volume; the instance stays disposable.

### Private by default

Dedicated GPU, encrypted disk, crypto payments, no KYC, no stored IPs — SSH behind your own credentials.

## Best GPUs for Hashcat CUDA

Three price points that run this template well — Good, Better, Best. Every model in the catalogue works; these are the value picks.

| Tier | GPU | VRAM | On-demand | Interruptible | Why this card |  |
| --- | --- | --- | --- | --- | --- | --- |
| (Good) | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | High hash rates per dollar for classic algorithms. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-3090) |
| (Better) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | The fastest single-card hash rates in the consumer line. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-4090) |
| (Best) | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | Blackwell throughput for the largest audits. | [Deploy](https://cloud.powergpu.ai/?gpu=rtx-5090) |

Need more VRAM? The [full catalogue](https://powergpu.ai/gpus) lists all 80 models with live availability; the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) sizes models to cards.

## Deploy Hashcat CUDA from the console, CLI or API

Pick the template in the [console](https://cloud.powergpu.ai/) deploy bar, or script it:

- **Console** — filter by GPU, choose *Hashcat CUDA* in the template picker, set disk and env, deploy.
- **CLI** — pip install powergpu, then the command on the right. [CLI reference](https://powergpu.ai/cli).
- **API** — POST /v1/instances with "template": "hashcat-cuda". [REST reference](https://powergpu.ai/api).
- **Own image** — any OCI reference works too; we inject the NVIDIA runtime. [Template docs](https://powergpu.ai/docs/templates).

*deploy — hashcat-cuda*

```
$ powergpu launch --gpu rtx-3090 --template hashcat-cuda \
    --disk 100 --volume models:/workspace/models
✓ instance i-7a41c0e2 running (27.9s)
# Hashcat CUDA · RTX 3090 · $0.108/hr · per second
# https://i-7a41c0e2.powergpu.ai:8000 (TLS)
$ powergpu stop i-7a41c0e2   # billing ends this second
```

## What is inside the Hashcat CUDA template

- **Image**: dizcza/docker-hashcat
- **CUDA**: inherits host driver (any 12.x / 13.x)
- **Access**: SSH shell
- **Category**: [Specialised](https://powergpu.ai/templates#tools)
- **Storage**: Instance NVMe disk (sized at deploy) + optional [network volumes](https://powergpu.ai/products/volumes)
- **Billing**: GPU price only, per second — no template fee, no setup fee

## Other specialised templates

- [Unreal Pixel Streaming on a cloud GPU](https://powergpu.ai/templates/unreal-pixel-streaming) — Stream an Unreal Engine app to the browser — pixel streaming on a cloud GPU.
- [All 37 templates](https://powergpu.ai/templates)

## Hashcat CUDA on a cloud GPU: FAQ

Environment variables, ports and custom images are covered in the [template docs](https://powergpu.ai/docs/templates).

**How much does it cost to run Hashcat CUDA on a cloud GPU?**

Only the GPU price — the template is free. From $0.054/hr on an interruptible RTX 3090, $0.108/hr on-demand on a RTX 3090. Billing is per second, so an hour of tinkering costs an hour, not a day. Storage is $0.08/GB/month.

**How long does Hashcat CUDA take to start?**

About 30 seconds from the deploy click: the image is pre-cached on hosts, ports are mapped and TLS is terminated for you. Restarting a stopped instance is faster, and your disk is exactly as you left it.

**Can I keep my models and outputs between sessions?**

Yes — attach a volume at deploy. Everything on the volume survives instance destruction and mounts on the next instance in the region in seconds, at $0.08/GB/month. The instance disk itself survives stop/start but not destroy.

**Is this allowed under the acceptable use policy?**

Yes for audits of systems you own or hold written authorisation to test, and for recovering your own passwords. Attacks on third parties are prohibited and lead to immediate suspension.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/templates/hashcat-cuda · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives -->

---
title: "GPU Cloud Alternatives (2026): PowerGPU vs Vast.ai, RunPod & 32 More"
description: "Looking for a Vast.ai, RunPod, Lambda, CoreWeave or AWS GPU alternative? 34 head-to-heads: public list prices with their check date, billing, payment, migration map."
url: https://powergpu.ai/alternatives
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternatives · 34 providers · public list prices, dated on each page

# GPU cloud alternatives: the same NVIDIA cards, 30% under the market

Honest head-to-heads with the clouds you already know — public list prices, billing rules, payment and identity requirements, and a migration map for each. Where a provider beats us on a card, the page says so.

- [RunPod alternative Developer GPU cloud: Pods, Serverless endpoints, Community and Secure tiers **H100 SXM**: RunPod $3.29 · PowerGPU $1.428 (−57%)](https://powergpu.ai/alternatives/runpod)
- [Lambda Labs alternative ML-focused cloud: on-demand instances, 1-Click Clusters, Lambda Stack images **H100 SXM**: Lambda $3.99 · PowerGPU $1.428 (−64%)](https://powergpu.ai/alternatives/lambda)
- [CoreWeave alternative Enterprise AI hyperscaler: Kubernetes-native, 8-GPU HGX nodes, reserved-first **H100 SXM**: CoreWeave $6.16 · PowerGPU $1.428 (−77%)](https://powergpu.ai/alternatives/coreweave)
- [Paperspace alternative Now part of DigitalOcean: GPU Droplets, Paperspace machines and Gradient notebooks **H100 SXM**: Paperspace $4.41 · PowerGPU $1.428 (−68%)](https://powergpu.ai/alternatives/paperspace)
- [TensorDock alternative GPU marketplace: independent hosts set prices across 100+ locations **H100 SXM**: TensorDock $2.25 · PowerGPU $1.428 (−37%)](https://powergpu.ai/alternatives/tensordock)
- [Vultr GPU alternative General cloud with GPU instances and bare metal in 32 regions **H100 SXM**: Vultr $2.99 · PowerGPU $1.428 (−52%)](https://powergpu.ai/alternatives/vultr)
- [Hyperstack alternative European GPU cloud with per-minute billing and NVLink options **H200**: Hyperstack $3.99 · PowerGPU $2.791 (−30%)](https://powergpu.ai/alternatives/hyperstack)
- [SaladCloud alternative Distributed consumer GPUs on volunteer PCs — the lowest hourly prices, with the trade-offs **RTX 5090**: SaladCloud $0.25 · PowerGPU $0.439 (+76%)](https://powergpu.ai/alternatives/salad)
- [io.net alternative Decentralized GPU network: clusters aggregated from independent suppliers, paid in crypto **H100 SXM**: io.net $2.10 · PowerGPU $1.428 (−32%)](https://powergpu.ai/alternatives/io-net)
- [Modal alternative Python-native serverless: functions on GPUs, per-second, scale to zero **B200**: Modal $6.25 · PowerGPU $5.425 (−13%)](https://powergpu.ai/alternatives/modal)
- [Jarvislabs alternative India- and EU-based GPU cloud with per-minute billing and pause/resume **H200**: Jarvislabs $3.99 · PowerGPU $2.791 (−30%)](https://powergpu.ai/alternatives/jarvislabs)
- [AWS GPU alternative Amazon EC2 GPU instances: P5 (H100), P4d (A100), G6e (L40S), G5 (A10G), G4dn (T4) **H100 SXM**: AWS $6.88 · PowerGPU $1.428 (−79%)](https://powergpu.ai/alternatives/aws)
- [Google Cloud GPU alternative Compute Engine A3 (H100), A2 (A100), G2 (L4) machine types and Vertex AI **H100 SXM**: Google Cloud $11.06 · PowerGPU $1.428 (−87%)](https://powergpu.ai/alternatives/google-cloud)
- [Azure GPU alternative ND H100 v5, NC A100 v4 and NC T4 virtual machines with enterprise agreements **H100 SXM**: Microsoft Azure $12.29 · PowerGPU $1.428 (−88%)](https://powergpu.ai/alternatives/azure)
- [Akash Network alternative Decentralized GPU marketplace: permissionless providers bid, settled on-chain in AKT or USDC **H100 SXM**: Akash Network $2.04 · PowerGPU $1.428 (−30%)](https://powergpu.ai/alternatives/akash)
- [Crusoe alternative AI cloud built on stranded and flared energy: H100, H200, A100, L40S on a published rate card **H200**: Crusoe Cloud $4.29 · PowerGPU $2.791 (−35%)](https://powergpu.ai/alternatives/crusoe)
- [Cudo Compute alternative NVIDIA Cloud Partner marketplace; live pricing page is quote-only, not self-serve **H100 SXM**: Cudo Compute $2.25 · PowerGPU $1.428 (−37%)](https://powergpu.ai/alternatives/cudo-compute)
- [DataCrunch alternative Renamed Verda in Nov. 2025: own Nordic datacenters, H100 through B300, no crypto payment **B300**: DataCrunch $7.50 · PowerGPU $6.737 (−10%)](https://powergpu.ai/alternatives/datacrunch)
- [FluidStack alternative Neocloud turned AI-factory builder: dedicated, multi-year contracts for a handful of large labs](https://powergpu.ai/alternatives/fluidstack)
- [Genesis Cloud alternative Munich GPU cloud on Nordic capacity; GmbH in formal liquidation since August 2025 **H100 SXM**: Genesis Cloud $2.19 · PowerGPU $1.428 (−35%)](https://powergpu.ai/alternatives/genesis-cloud)
- [Hetzner GPU alternative Dedicated GPU servers by the month (or hour): RTX PRO 4000/6000 Blackwell, Germany/Finland **RTX PRO 4000**: Hetzner $0.34 · PowerGPU $0.183 (−46%)](https://powergpu.ai/alternatives/hetzner)
- [Hyperbolic alternative GPU marketplace and inference API: weekly-refreshed on-demand rates, reserved and private-cloud tiers **H100 SXM**: Hyperbolic $3.19 · PowerGPU $1.428 (−55%)](https://powergpu.ai/alternatives/hyperbolic)
- [Latitude.sh alternative Bare metal GPU by the hour or the month: H100, RTX PRO 6000, HGX B300 across the Americas, Europe and Asia **H100 PCIE**: Latitude.sh $1.68 · PowerGPU $1.867 (+11%)](https://powergpu.ai/alternatives/latitude-sh)
- [Lightning AI alternative Cloud IDE with built-in GPUs: Studios, per-hour billing, monthly free credits **Tesla T4**: Lightning AI $0.41 · PowerGPU $0.103 (−75%)](https://powergpu.ai/alternatives/lightning-ai)
- [Massed Compute alternative NVIDIA Preferred Partner: on-demand GPU VMs, bare metal, US-only, per-minute billing **B300**: Massed Compute $6.60 · PowerGPU $6.737 (+2%)](https://powergpu.ai/alternatives/massed-compute)
- [Nebius AI Cloud alternative Ex-Yandex hyperscaler-scale cloud: owned Finland/France datacenters, H100 through B300 **B300**: Nebius $7.85 · PowerGPU $6.737 (−14%)](https://powergpu.ai/alternatives/nebius)
- [Novita AI alternative GPU instances, serverless endpoints and hosted model APIs, strong APAC reach **RTX 4090**: Novita AI $0.61 · PowerGPU $0.327 (−46%)](https://powergpu.ai/alternatives/novita)
- [Oracle Cloud GPU alternative OCI bare metal and VMs: H100, H200, A100, L40S, A10, preemptible at 50% off **H100 SXM**: Oracle Cloud $10.00 · PowerGPU $1.428 (−86%)](https://powergpu.ai/alternatives/oracle-cloud)
- [OVHcloud GPU alternative European sovereign cloud: hourly Public Cloud GPUs, Scale/HGR-AI bare metal by quote **H100 PCIE**: OVHcloud $2.99 · PowerGPU $1.867 (−38%)](https://powergpu.ai/alternatives/ovhcloud)
- [Prime Intellect alternative Multi-provider compute exchange: aggregated on-demand GPUs, multi-node clusters, decentralized training runs **H100 SXM**: Prime Intellect $1.65 · PowerGPU $1.428 (−13%)](https://powergpu.ai/alternatives/prime-intellect)
- [Replicate alternative Hosted models via API, Cog containers, billed per second of GPU time **Tesla T4**: Replicate $0.81 · PowerGPU $0.103 (−87%)](https://powergpu.ai/alternatives/replicate)
- [Scaleway GPU alternative European GPU cloud: H100 SXM, L40S, L4 by the hour, free egress, Paris/Amsterdam/Warsaw **H100 SXM**: Scaleway $3.17 · PowerGPU $1.428 (−55%)](https://powergpu.ai/alternatives/scaleway)
- [Together AI alternative Inference and fine-tuning API, plus dedicated GPU Clusters on H100, H200, B200 **H100 SXM**: Together AI $3.99 · PowerGPU $1.428 (−64%)](https://powergpu.ai/alternatives/together-ai)
- [Vast.ai alternative GPU marketplace: independent hosts, DLPerf scores, verified-datacenter tier **H100 SXM**: Vast.ai $2.21 · PowerGPU $1.428 (−35%)](https://powergpu.ai/alternatives/vast-ai)

Competitor names and logos belong to their owners and identify the compared service only. List prices are quoted from each provider's public pricing page on the date shown on that page; they change — verify before deciding.

## Choosing a GPU cloud: FAQ

Side-by-side hardware decisions live in [GPU comparisons](https://powergpu.ai/compare); full-month budgets in the [cost calculator](https://powergpu.ai/calculator).

**How are competitor prices on these pages collected?**

Each page quotes the public list price shown on the provider's own pricing page on the date printed at the top of that page, on-demand, per GPU-hour, before taxes, with the configuration in a note. When a provider is cheaper on a card, the page says so. PowerGPU prices are live from the sheet.

**What makes PowerGPU different from all of them?**

One public rule instead of a price list: every on-demand price is the public marketplace median × 0.70, re-checked weekly. Add per-second billing, crypto-only payment with no KYC, verified datacenters with a 99.9% SLA, and a flat −50% interruptible tier.

**How often are these pages updated?**

PowerGPU prices on every page are live: they follow the weekly re-check of the market median. Competitor list prices are re-read when a page is revisited; the date at the top of each page says when. If you spot a change we have not caught, the provider's own pricing page wins.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/runpod -->

---
title: "RunPod Alternative (2026): GPU Prices vs PowerGPU"
description: "RunPod vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/runpod
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# RunPod alternative: same NVIDIA GPUs, fixed prices 30% under the market

RunPod is one of the most popular developer GPU clouds: container "Pods" on its own Secure Cloud datacenters or on third-party Community Cloud hosts, Serverless endpoints with fast cold starts, network volumes, 30+ regions and a big template library. Billing is per second and payment is by card.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## RunPod vs PowerGPU at a glance

RunPod

Developer GPU cloud: Pods, Serverless endpoints, Community and Secure tiers

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | RunPod | PowerGPU |
| --- | --- | --- |
| Billing | Per second | Per second, no minimum, price locked at deploy |
| Payment | Card (Stripe), prepaid balance | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card details required | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Community Cloud (third-party hosts) and spot pods | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 30+ | 32 regions on 5 continents, verified datacenters only |
| Minimums | None | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Container disk $0.10/GB/mo · network volume $0.10 running, $0.20 idle | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not itemised on the pricing page | $0.01/GB in and out, every region |
| Access | Containers (Pods), Serverless, SSH/Jupyter, API, runpodctl | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

RunPod column: as published on runpod.io on 2026-09-03. Details change — verify before you decide.

## RunPod vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on RunPod's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | RunPod list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (Secure Cloud · Community $2.69) | $3.29 | **$1.428** | $0.714 | −57% on PowerGPU |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (Secure · Community $1.99) | $2.89 | **$1.867** | $0.933 | −35% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (Secure · Community $3.59) | $4.59 | **$2.791** | $1.395 | −39% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (Secure · Community $5.98) | $6.79 | **$5.425** | $2.712 | −20% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (Secure · Community $1.39) | $1.59 | **$0.560** | $0.280 | −65% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (Secure · Community $0.79) | $0.99 | **$0.514** | $0.257 | −48% on PowerGPU |
| [L4](https://powergpu.ai/gpu/l4) (Secure · Community $0.44) | $0.49 | **$0.225** | $0.112 | −54% on PowerGPU |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (Secure · Community $0.34) | $0.74 | **$0.327** | $0.163 | −56% on PowerGPU |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (Secure · Community $0.69) | $0.99 | **$0.439** | $0.219 | −56% on PowerGPU |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) (Secure · Community $0.33) | $0.53 | **$0.281** | $0.140 | −47% on PowerGPU |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) (Secure · Community $0.74) | $0.84 | **$0.467** | $0.233 | −44% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which RunPod price is quoted. Negative differences mean RunPod is cheaper on that card.

## When to stay with RunPod

- Serverless with sub-second cold starts and FlashBoot is your whole product — RunPod's serverless tooling is more mature.
- You depend on one of RunPod's specific regions or on a community template that only exists there.
- Finance insists on card or invoice billing; PowerGPU is crypto-settled by design.

## Why teams switch to PowerGPU

- Secure-tier reliability at below Community-tier prices: every PowerGPU machine sits in a verified datacenter, yet the H100 SXM is fixed at market median × 0.70.
- No card, no KYC, no stored IP addresses — top up in USDT or Monero and deploy.
- Storage that does not double when idle: $0.08/GB/month, running or stopped, versus $0.10/$0.20 on RunPod volumes.
- A flat −50% interruptible tier instead of a fluctuating spot market — and 32 regions instead of choosing between Secure and Community availability.

## Switching from RunPod: what maps to what

| On RunPod | On PowerGPU |
| --- | --- |
| Pod (Secure Cloud) | Instance — every machine is a verified datacenter host |
| Pod (Community Cloud) | Interruptible instance, flat −50% |
| Network volume | Volume, $0.08/GB/month, no idle rate |
| Serverless endpoint | Serverless endpoint (vLLM / ComfyUI / any container, scale to zero) |
| Template | One of 37 templates or any OCI image |
| runpodctl / REST API | powergpu CLI / REST API with pg_live_ keys |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your RunPod machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | RunPod | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $2,402 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $526 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Container disk $0.10/GB/mo · network volume $0.10 running, $0.20 idle | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## RunPod alternative: FAQ

**Is PowerGPU cheaper than RunPod?**

On every datacenter card on the sheet, yes: PowerGPU prices are fixed at the public market median × 0.70, while RunPod Secure Cloud lists an H100 SXM at $3.29/hr and an A100 80 GB at $1.59/hr (checked 2026-09-03). RunPod Community Cloud is cheaper than Secure but runs on third-party hosts; PowerGPU interruptible capacity is usually below Community prices with datacenter reliability.

**Can I run my RunPod template on PowerGPU?**

Yes. Deploy any public or private OCI image; ports, environment variables and volume mounts are set at deploy exactly like a Pod template. The official PyTorch, vLLM, ComfyUI and Ollama images are already in the template picker.

**Does PowerGPU have serverless like RunPod?**

Yes — autoscaling endpoints with scale-to-zero, cached weights and the same fixed per-second GPU prices. RunPod's cold-start tooling is more mature; for steady traffic a reserved instance at −35% is usually cheaper than any serverless.

**How do I move data off a RunPod network volume?**

Start a small PowerGPU instance with a volume attached, then rsync or rclone from the RunPod pod over SSH. Inbound bandwidth costs $0.01/GB here; RunPod does not itemise egress.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on RunPod, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

RunPod and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with RunPod. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/runpod · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/lambda -->

---
title: "Lambda Labs Alternative (2026): GPU Prices vs PowerGPU"
description: "Lambda vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/lambda
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Lambda Labs alternative: same NVIDIA GPUs, fixed prices 30% under the market

Lambda (formerly Lambda Labs) runs its own datacenters and sells on-demand GPU instances, 1-Click Clusters with InfiniBand and large reserved contracts, all with the Lambda Stack ML image preinstalled. Billing is per minute, egress is free, payment is by card, and single-GPU capacity is often limited.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Lambda vs PowerGPU at a glance

Lambda

ML-focused cloud: on-demand instances, 1-Click Clusters, Lambda Stack images

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Lambda | PowerGPU |
| --- | --- | --- |
| Billing | Per minute | Per second, no minimum, price locked at deploy |
| Payment | Card | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account verification; business details for reserved capacity | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Reserved contracts via sales; no spot tier | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Primarily United States | 32 regions on 5 continents, verified datacenters only |
| Minimums | None on-demand; 1-Click Clusters start at 16 GPUs | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Persistent filesystems billed per GB/month | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free | $0.01/GB in and out, every region |
| Access | VMs with Lambda Stack, SSH, Jupyter, API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Lambda column: as published on lambda.ai on 2026-09-03. Details change — verify before you decide.

## Lambda vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Lambda's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Lambda list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (1×–8× SXM instances, up to $4.29 by config) | $3.99 | **$1.428** | $0.714 | −64% on PowerGPU |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | $3.29 | **$1.867** | $0.933 | −43% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (1×; 8× from $6.69) | $6.99 | **$5.425** | $2.712 | −22% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (80 GB; up to $2.79 by config) | $1.99 | **$0.560** | $0.280 | −72% on PowerGPU |
| [A10](https://powergpu.ai/gpu/a10) | $1.29 | **$0.168** | $0.084 | −87% on PowerGPU |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | $1.09 | **$0.281** | $0.140 | −74% on PowerGPU |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) | $0.79 | **$0.130** | $0.065 | −84% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Lambda price is quoted. Negative differences mean Lambda is cheaper on that card.

## When to stay with Lambda

- You need a 16–512 GPU InfiniBand cluster on a long reservation with hands-on ML support — that is Lambda's core business.
- Your team is standardised on the Lambda Stack image and its exact driver/CUDA combinations.
- You only ever rent datacenter cards and never need a consumer GPU.

## Why teams switch to PowerGPU

- Price: the H100 SXM is fixed at market median × 0.70 here versus $3.99/hr at Lambda; the A100 80 GB gap is similar.
- Availability: single GPUs and 2×/4× machines are always in stock across 32 regions — Lambda regularly shows "unavailable" on 1× instances.
- Consumer cards: RTX 5090, 4090 and 3090 for diffusion and quantized LLMs — Lambda sells none.
- Crypto, no KYC, per-second billing and a flat −50% interruptible tier Lambda does not offer.

## Switching from Lambda: what maps to what

| On Lambda | On PowerGPU |
| --- | --- |
| On-demand instance (VM) | Instance — container in 30 s or full KVM VM |
| Lambda Stack | PyTorch / PyTorch NGC / NVIDIA CUDA templates |
| Persistent filesystem | Volume, $0.08/GB/month |
| 1-Click Cluster | Clusters — 16 to 512 GPUs over InfiniBand, quoted from the public sheet |
| Lambda Cloud API | REST API, CLI, Python SDK |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Lambda machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Lambda | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $2,913 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $638 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Persistent filesystems billed per GB/month | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Lambda Labs alternative: FAQ

**Is PowerGPU cheaper than Lambda?**

Yes on every overlapping card: Lambda lists the H100 SXM from $3.99/hr and the A100 80 GB from $1.99/hr (checked 2026-09-03); PowerGPU fixes both at the public market median × 0.70, re-checked weekly. Lambda's free egress is matched closely by a flat $0.01/GB here.

**Does PowerGPU offer InfiniBand clusters like Lambda's 1-Click Clusters?**

Yes — multi-node pods of 8× SXM machines with NVLink in-node and InfiniBand between nodes, from 16 to 512 GPUs, provisioned in 1–3 business days and priced per GPU from the same sheet.

**Can I keep using the Lambda Stack environment?**

The PyTorch and NVIDIA NGC templates carry the same frameworks, drivers and CUDA versions; or bring your own image if you have pinned a specific stack.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Lambda, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Lambda and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Lambda. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/lambda · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/coreweave -->

---
title: "CoreWeave Alternative (2026): GPU Prices vs PowerGPU"
description: "CoreWeave vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/coreweave
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# CoreWeave alternative: same NVIDIA GPUs, fixed prices 30% under the market

CoreWeave is an AI-specialised hyperscaler: Kubernetes-native platform (CKS), Slurm on Kubernetes, HGX H100/H200/B200 nodes with InfiniBand, and most capacity sold on multi-year reservations to large labs. On-demand is priced per 8-GPU node and billed hourly; onboarding is an enterprise sales process.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## CoreWeave vs PowerGPU at a glance

CoreWeave

Enterprise AI hyperscaler: Kubernetes-native, 8-GPU HGX nodes, reserved-first

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | CoreWeave | PowerGPU |
| --- | --- | --- |
| Billing | Hourly | Per second, no minimum, price locked at deploy |
| Payment | Invoicing and contracts; credit approval | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Enterprise onboarding (KYB) | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot (roughly −60%) and reservations up to −60% | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | North America and Europe | 32 regions on 5 continents, verified datacenters only |
| Minimums | Whole 8-GPU nodes; most capacity under contract | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Object/file storage from $0.015 to $0.06/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free | $0.01/GB in and out, every region |
| Access | Kubernetes (CKS), Slurm, bare-metal nodes | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

CoreWeave column: as published on coreweave.com on 2026-09-03. Details change — verify before you decide.

## CoreWeave vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on CoreWeave's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | CoreWeave list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (8× HGX H100 node $49.24/hr ÷ 8) | $6.16 | **$1.428** | $0.714 | −77% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (8× HGX H200 node $50.44/hr ÷ 8) | $6.31 | **$2.791** | $1.395 | −56% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (8× HGX B200 node $68.80/hr ÷ 8) | $8.60 | **$5.425** | $2.712 | −37% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (8× A100 node $21.60/hr ÷ 8) | $2.70 | **$0.560** | $0.280 | −79% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (8× node $18.00/hr ÷ 8) | $2.25 | **$0.514** | $0.257 | −77% on PowerGPU |
| [L40](https://powergpu.ai/gpu/l40) (8× node $10.00/hr ÷ 8) | $1.25 | **$0.235** | $0.117 | −81% on PowerGPU |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) (RTX PRO 6000 (high memory), 8× node $20.00/hr ÷ 8) | $2.50 | **$1.040** | $0.520 | −58% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which CoreWeave price is quoted. Negative differences mean CoreWeave is cheaper on that card.

## When to stay with CoreWeave

- You are reserving thousands of GPUs for years with a Kubernetes-native platform team and enterprise SLAs.
- Your workloads are already CKS/Slurm-on-K8s manifests and you value that control plane.
- You need GB200 NVL72 or B300-class systems at rack scale.

## Why teams switch to PowerGPU

- You need one to eight GPUs today, not an 8-node contract: PowerGPU deploys single cards in 30 seconds with no sales call.
- Per-GPU price: the H100 SXM is fixed at market median × 0.70 here versus $6.16/GPU-hour on CoreWeave on-demand nodes.
- Per-second billing, crypto settlement and no procurement process — the sheet is the quote.
- Consumer and workstation cards CoreWeave does not sell (RTX 5090, 4090, RTX 6000 Ada, A6000).

## Switching from CoreWeave: what maps to what

| On CoreWeave | On PowerGPU |
| --- | --- |
| CKS pod / bare-metal node | Instance (container or full VM), 1× to 8× GPUs |
| Slurm on Kubernetes | Clusters with Slurm preinstalled on request |
| Object / file storage | Volumes at $0.08/GB/month |
| Reserved contract | Reserved instances (−35%) and enterprise fleets, quoted from the public sheet |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your CoreWeave machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | CoreWeave | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $4,497 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $986 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Object/file storage from $0.015 to $0.06/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## CoreWeave alternative: FAQ

**Is PowerGPU cheaper than CoreWeave?**

On-demand, by a wide margin: CoreWeave prices whole 8× HGX H100 nodes at $49.24/hr ($6.16 per GPU, checked 2026-09-03); PowerGPU fixes the H100 SXM at the public market median × 0.70 per GPU and lets you rent one card. CoreWeave's large reservations can undercut its own on-demand by up to 60%.

**Can PowerGPU replace a CoreWeave cluster?**

For 16–512 GPU training pods with InfiniBand, yes — quoted per GPU from the public sheet and provisioned in days. For thousand-GPU multi-year deployments, CoreWeave remains the specialist.

**Do I need an enterprise contract on PowerGPU?**

No. Create an account with an email, top up in crypto and deploy. Enterprise terms (fenced capacity, named support, sub-accounts) exist for larger fleets but are optional.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on CoreWeave, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

CoreWeave and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with CoreWeave. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/coreweave · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/paperspace -->

---
title: "Paperspace Alternative (2026): GPU Prices vs PowerGPU"
description: "Paperspace vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/paperspace
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Paperspace alternative: same NVIDIA GPUs, fixed prices 30% under the market

Paperspace was acquired by DigitalOcean; its Gradient notebooks and Core machines now sit next to DigitalOcean GPU Droplets (H100, H200, L40S, RTX Ada, AMD MI300X). Billing is per second with a five-minute minimum, payment by card or PayPal, and the GPU line-up is datacenter cards only.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Paperspace vs PowerGPU at a glance

Paperspace

Now part of DigitalOcean: GPU Droplets, Paperspace machines and Gradient notebooks

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Paperspace | PowerGPU |
| --- | --- | --- |
| Billing | Per second, 5-minute minimum | Per second, no minimum, price locked at deploy |
| Payment | Card, PayPal, Google Pay, Apple Pay | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card and account verification | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Promotional credits for new accounts (varies) | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Reservations via sales; free Gradient notebook tiers | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Selected DigitalOcean datacenters (US / Canada) | 32 regions on 5 continents, verified datacenters only |
| Minimums | None | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Boot and scratch NVMe included; volumes ~$0.10/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | 10–15 TB included per month, then per GB | $0.01/GB in and out, every region |
| Access | VMs (Droplets), Jupyter notebooks, API, doctl | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Paperspace column: as published on digitalocean.com on 2026-09-03. Details change — verify before you decide.

## Paperspace vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Paperspace's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Paperspace list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (GPU Droplet, 1× H100) | $4.41 | **$1.428** | $0.714 | −68% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (GPU Droplet, 1× H200) | $4.47 | **$2.791** | $1.395 | −38% on PowerGPU |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) (GPU Droplet) | $1.57 | **$0.467** | $0.233 | −70% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (GPU Droplet) | $1.57 | **$0.514** | $0.257 | −67% on PowerGPU |
| [RTX 4000Ada](https://powergpu.ai/gpu/rtx-4000ada) (GPU Droplet) | $0.76 | **$0.128** | $0.064 | −83% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Paperspace price is quoted. Negative differences mean Paperspace is cheaper on that card.

## When to stay with Paperspace

- Your stack already lives on DigitalOcean (Droplets, Spaces, Kubernetes) and you want one bill.
- Gradient notebooks with a free tier are exactly the workflow you teach or learn on.
- You need AMD MI300X or MI325X capacity.

## Why teams switch to PowerGPU

- Price: the H100 is $4.41/hr on GPU Droplets versus market median × 0.70 fixed here — and the L40S gap is similar.
- Consumer cards for diffusion and quantized LLMs (RTX 5090, 4090, 3090), which DigitalOcean does not offer.
- No five-minute minimum, no card, no KYC: per-second billing settled in crypto.
- 32 regions and interruptible capacity at a flat −50%.

## Switching from Paperspace: what maps to what

| On Paperspace | On PowerGPU |
| --- | --- |
| GPU Droplet | Instance (container in 30 s or full KVM VM) |
| Gradient notebook | PyTorch template with JupyterLab on a mapped port |
| Block storage volume | Volume, $0.08/GB/month |
| doctl / API | powergpu CLI / REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Paperspace machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Paperspace | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $3,219 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $706 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Boot and scratch NVMe included; volumes ~$0.10/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Paperspace alternative: FAQ

**Is PowerGPU cheaper than Paperspace / DigitalOcean GPU Droplets?**

Yes: DigitalOcean lists the H100 at $4.41/hr and the L40S at $1.57/hr (checked 2026-09-03); PowerGPU fixes both at the public market median × 0.70, billed per second with no five-minute minimum.

**I used Paperspace Gradient notebooks — what is the equivalent?**

The PyTorch, TensorFlow or NVIDIA RAPIDS templates start JupyterLab on a TLS-terminated port in about 30 seconds; keep datasets and notebooks on a volume so the GPU instance stays disposable.

**Does PowerGPU include bandwidth like DigitalOcean?**

Bandwidth is a flat $0.01/GB in and out rather than a bundled allowance — cheaper for most GPU workloads, which move gigabytes, not terabytes.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Paperspace, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Paperspace and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Paperspace. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/paperspace · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/tensordock -->

---
title: "TensorDock Alternative (2026): GPU Prices vs PowerGPU"
description: "TensorDock vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/tensordock
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# TensorDock alternative: same NVIDIA GPUs, fixed prices 30% under the market

TensorDock is a marketplace where independent hosts list GPU machines at their own prices across 100+ locations. It is cheap and flexible — per-second billing, VMs and containers — but vCPU, RAM and storage are billed as separate lines and reliability depends on the host.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## TensorDock vs PowerGPU at a glance

TensorDock

GPU marketplace: independent hosts set prices across 100+ locations

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | TensorDock | PowerGPU |
| --- | --- | --- |
| Billing | Per second (GPU + vCPU + RAM + storage billed separately) | Per second, no minimum, price locked at deploy |
| Payment | Card, prepaid balance | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Email and card | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot on some hosts | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 100+ locations (independent hosts) | 32 regions on 5 continents, verified datacenters only |
| Minimums | None | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | NVMe ~$0.036/GB/mo + $0.003/vCPU-hr + $0.002/GB RAM-hr | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not itemised | $0.01/GB in and out, every region |
| Access | VMs and containers, SSH, API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

TensorDock column: as published on tensordock.com on 2026-09-03. Details change — verify before you decide.

## TensorDock vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on TensorDock's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | TensorDock list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (from; hosts set prices) | $2.25 | **$1.428** | $0.714 | −37% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (from) | $1.80 | **$0.560** | $0.280 | −69% on PowerGPU |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) (from) | $1.50 | **$0.374** | $0.187 | −75% on PowerGPU |
| [L40](https://powergpu.ai/gpu/l40) (from) | $0.95 | **$0.235** | $0.117 | −75% on PowerGPU |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) (from) | $0.75 | **$0.467** | $0.233 | −38% on PowerGPU |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) (from) | $0.45 | **$0.281** | $0.140 | −38% on PowerGPU |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (from) | $0.35 | **$0.327** | $0.163 | −7% on PowerGPU |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) (from) | $0.20 | **$0.108** | $0.054 | −46% on PowerGPU |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) (from) | $0.10 | **$0.071** | $0.035 | −29% on PowerGPU |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) (from) | $0.17 | **$0.130** | $0.065 | −24% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which TensorDock price is quoted. Negative differences mean TensorDock is cheaper on that card.

## When to stay with TensorDock

- You need a machine in one very specific city and a host happens to be there.
- You want the absolute floor price on an old card (A4000, V100) and accept host-dependent reliability.
- You are already comfortable with marketplace-style variance in hardware and uptime.

## Why teams switch to PowerGPU

- Fixed, all-inclusive prices: no separate vCPU, RAM or disk lines — the per-GPU rate includes the machine.
- Verified datacenters only, with a 99.9% SLA, instead of host-by-host reliability.
- Crypto payments with no card or KYC, and a flat −50% interruptible tier rather than per-host spot.
- Datacenter flagships (H100, H200, B200) always in stock at market median × 0.70, plus RTX 5090s.

## Switching from TensorDock: what maps to what

| On TensorDock | On PowerGPU |
| --- | --- |
| Host machine (VM) | Instance on a verified datacenter machine — same CPU/RAM class shown on every offer |
| Spot host | Interruptible instance, flat −50% |
| Separate vCPU / RAM / NVMe lines | Included in the per-GPU price; storage $0.08/GB/month |
| TensorDock API | REST API, CLI, Python SDK |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your TensorDock machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | TensorDock | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $1,643 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $360 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | NVMe ~$0.036/GB/mo + $0.003/vCPU-hr + $0.002/GB RAM-hr | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## TensorDock alternative: FAQ

**Is PowerGPU cheaper than TensorDock?**

On flagships, yes: TensorDock hosts list the H100 SXM from $2.25/hr and the A100 SXM from $1.80/hr before vCPU, RAM and disk (checked 2026-09-03); PowerGPU fixes both at market median × 0.70 all-inclusive. On the cheapest legacy cards TensorDock hosts can be a few cents lower per hour.

**Are PowerGPU machines from independent hosts?**

No. Every machine sits in a Tier-III colocation facility with redundant power and network, burned in for 72 hours before it is listed. There are no residential hosts.

**Do you charge separately for CPU and RAM?**

No — the per-GPU price includes the vCPUs and RAM shown on the offer. Only storage and bandwidth are separate lines.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on TensorDock, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

TensorDock and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with TensorDock. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/tensordock · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/vultr -->

---
title: "Vultr GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "Vultr vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/vultr
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Vultr GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

Vultr is a general-purpose cloud (compute, Kubernetes, storage) that also sells GPU cloud instances and 8× HGX bare-metal servers in 32 regions, with fractional GPU slices and one- to three-year committed-use discounts. Billing is hourly and payment is by card or PayPal.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Vultr vs PowerGPU at a glance

Vultr

General cloud with GPU instances and bare metal in 32 regions

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Vultr | PowerGPU |
| --- | --- | --- |
| Billing | Hourly | Per second, no minimum, price locked at deploy |
| Payment | Card, PayPal | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card and account verification | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Promotional credits | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Committed-use discounts (1–3 years) | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 32 | 32 regions on 5 continents, verified datacenters only |
| Minimums | None (fractional GPUs available) | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Block storage ~$0.10/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Bandwidth allowance, then per GB | $0.01/GB in and out, every region |
| Access | VMs, bare metal, Kubernetes, API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Vultr column: as published on vultr.com on 2026-09-03. Details change — verify before you decide.

## Vultr vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Vultr's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Vultr list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (8× HGX H100 bare metal, $23.92/hr ÷ 8) | $2.99 | **$1.428** | $0.714 | −52% on PowerGPU |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) (Cloud GPU, 80 GB) | $2.40 | **$0.374** | $0.187 | −84% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (Cloud GPU) | $1.67 | **$0.514** | $0.257 | −69% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Vultr price is quoted. Negative differences mean Vultr is cheaper on that card.

## When to stay with Vultr

- You want compute, object storage, load balancers and GPUs on one Vultr account.
- A fractional GPU slice is all your inference service needs.
- You can commit to three-year terms for the deepest Vultr discounts.

## Why teams switch to PowerGPU

- Per-second billing and fixed per-GPU prices at market median × 0.70 — the H100 SXM is well under Vultr's $2.99 bare-metal rate, and no 8× minimum.
- Consumer cards (RTX 5090, 4090) and workstation cards Vultr does not carry.
- Crypto, no KYC, and a flat −50% interruptible tier without a multi-year commitment.

## Switching from Vultr: what maps to what

| On Vultr | On PowerGPU |
| --- | --- |
| Cloud GPU instance | Instance (container or full VM) |
| 8× HGX bare metal | 8× SXM machine at the same per-GPU price, or a cluster |
| Block storage | Volume, $0.08/GB/month |
| Committed-use discount | Reserved instances, −35% from 3 months |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Vultr machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Vultr | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $2,183 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $478 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Block storage ~$0.10/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Vultr GPU alternative: FAQ

**Is PowerGPU cheaper than Vultr for GPUs?**

Yes: Vultr's self-serve H100 is an 8× bare-metal server at $23.92/hr ($2.99 per GPU) and the A100 80 GB PCIe lists at $2.40/hr (checked 2026-09-03); PowerGPU fixes both at the public market median × 0.70 and rents single cards per second.

**Does PowerGPU offer fractional GPUs like Vultr?**

No — every instance gets whole, dedicated GPUs. The cheapest whole cards (RTX 3060, A2000, T4) start at a few cents per hour, which covers most fractional use cases.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Vultr, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Vultr and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Vultr. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/vultr · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/hyperstack -->

---
title: "Hyperstack Alternative (2026): GPU Prices vs PowerGPU"
description: "Hyperstack vs PowerGPU: H200 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/hyperstack
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Hyperstack alternative: same NVIDIA GPUs, fixed prices 30% under the market

Hyperstack (by NexGen Cloud) sells on-demand and reserved datacenter GPUs — H100 PCIe/NVLink/SXM, H200, A100, L40, RTX A-series — with per-minute billing, free egress and renewable-powered European and North American datacenters. Payment is by card through Stripe; reservations are invoiced.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Hyperstack vs PowerGPU at a glance

Hyperstack

European GPU cloud with per-minute billing and NVLink options

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Hyperstack | PowerGPU |
| --- | --- | --- |
| Billing | Per minute | Per second, no minimum, price locked at deploy |
| Payment | Card (Stripe); invoices for reservations | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Reservations (e.g. H100 SXM from $2.72) | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Europe and North America | 32 regions on 5 continents, verified datacenters only |
| Minimums | None | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | ~$0.07/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free | $0.01/GB in and out, every region |
| Access | VMs, SSH, API, Terraform | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Hyperstack column: as published on hyperstack.cloud on 2026-09-03. Details change — verify before you decide.

## Hyperstack vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Hyperstack's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Hyperstack list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H200](https://powergpu.ai/gpu/h200) (SXM) | $3.99 | **$2.791** | $1.395 | −30% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (SXM · reserved from $2.72) | $3.20 | **$1.428** | $0.714 | −55% on PowerGPU |
| [H100 NVL](https://powergpu.ai/gpu/h100-nvl) (NVLink) | $2.60 | **$1.811** | $0.905 | −30% on PowerGPU |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (reserved from $1.75) | $2.50 | **$1.867** | $0.933 | −25% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (SXM) | $1.60 | **$0.560** | $0.280 | −65% on PowerGPU |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) | $1.35 | **$0.374** | $0.187 | −72% on PowerGPU |
| [L40](https://powergpu.ai/gpu/l40) | $1.00 | **$0.235** | $0.117 | −77% on PowerGPU |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | $0.50 | **$0.281** | $0.140 | −44% on PowerGPU |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) | $0.15 | **$0.071** | $0.035 | −53% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Hyperstack price is quoted. Negative differences mean Hyperstack is cheaper on that card.

## When to stay with Hyperstack

- EU data residency on renewable-powered datacenters is a hard requirement.
- You want NVLink or InfiniBand-connected H100 fleets on a reservation with invoicing.
- Terraform-managed VM fleets are your operating model.

## Why teams switch to PowerGPU

- Price: Hyperstack lists the H100 SXM at $3.20/hr and the A100 SXM at $1.60/hr; PowerGPU fixes both at market median × 0.70 and re-checks weekly.
- Consumer cards (RTX 5090, 4090, 3090) for diffusion and quantized LLM work, absent from Hyperstack.
- Per-second billing, crypto with no KYC, and 32 regions including Asia-Pacific and South America.

## Switching from Hyperstack: what maps to what

| On Hyperstack | On PowerGPU |
| --- | --- |
| VM | Instance (container or full KVM VM) |
| Volume | Volume, $0.08/GB/month |
| Reservation | Reserved instance (−35%, capacity held) |
| Hyperstack API / Terraform | REST API, CLI, Python SDK |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h200 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $2.791/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Hyperstack machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Hyperstack | PowerGPU |
| --- | --- | --- |
| 1× H200, 730 hours on-demand | $2,913 | **$2,037** |
| 1× H200, 8 h/day × 20 days | $638 | **$447** |
| Same 160 hours, PowerGPU interruptible | — | **$223** |
| 500 GB of storage, one month | ~$0.07/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Hyperstack alternative: FAQ

**Is PowerGPU cheaper than Hyperstack?**

On-demand, yes on every overlapping card (Hyperstack H100 SXM $3.20, A100 SXM $1.60, checked 2026-09-03 — PowerGPU fixes them at market median × 0.70). Hyperstack's reserved H100 SXM at $2.72 is still above PowerGPU on-demand.

**Does PowerGPU have European regions?**

Ten: Amsterdam, London, Paris, Frankfurt, Warsaw, Stockholm, Helsinki, Milan, Madrid and Bucharest — all Tier-III facilities.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Hyperstack, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Hyperstack and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Hyperstack. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/hyperstack · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/salad -->

---
title: "SaladCloud Alternative (2026): GPU Prices vs PowerGPU"
description: "SaladCloud vs PowerGPU: RTX 5090 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/salad
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# SaladCloud alternative: same NVIDIA GPUs, fixed prices 30% under the market

SaladCloud runs containers on thousands of consumer PCs whose owners rent out idle GPUs. That makes it the cheapest RTX 4090 and 5090 by the hour, at the cost of node churn, no SSH, no persistent disks and residential-grade networking. Payment is by card; pricing is announced to change on 12 September 2026.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## SaladCloud vs PowerGPU at a glance

SaladCloud

Distributed consumer GPUs on volunteer PCs — the lowest hourly prices, with the trade-offs

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | SaladCloud | PowerGPU |
| --- | --- | --- |
| Billing | Usage-based per node-hour | Per second, no minimum, price locked at deploy |
| Payment | Card | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Priority tiers (batch vs high) | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Distributed residential nodes worldwide | 32 regions on 5 continents, verified datacenters only |
| Minimums | None | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Ephemeral only — no persistent volumes | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Included | $0.01/GB in and out, every region |
| Access | Containers only, no SSH; API and portal | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

SaladCloud column: as published on salad.com on 2026-09-03. Details change — verify before you decide.

## SaladCloud vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on SaladCloud's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | SaladCloud list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (consumer node, 32 GB) | $0.25 | **$0.439** | $0.219 | +76% on PowerGPU |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (consumer node, 24 GB) | $0.16 | **$0.327** | $0.163 | +104% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which SaladCloud price is quoted. Negative differences mean SaladCloud is cheaper on that card.

## When to stay with SaladCloud

- Your job is stateless, embarrassingly parallel and tolerant of nodes disappearing mid-task (batch image generation, transcription farms).
- The absolute lowest hourly price on a 4090/5090 matters more than reliability, and your container survives cold restarts.

## Why teams switch to PowerGPU

- Honest math: SaladCloud is cheaper per hour on the RTX 4090 ($0.16 vs our on-demand rate) — but PowerGPU interruptible capacity on the same card is priced below Salad's rate, on datacenter hardware.
- SSH, Jupyter, persistent volumes, full VMs and datacenter cards (H100, H200, L40S) — none of which exist on a residential node network.
- Deterministic hardware: the machine you deploy on is the machine you get, with a known CPU, RAM, disk and uplink.
- Crypto payments with no KYC.

## Switching from SaladCloud: what maps to what

| On SaladCloud | On PowerGPU |
| --- | --- |
| Container group on consumer nodes | Interruptible instance on a datacenter machine, flat −50% |
| Ephemeral storage | Instance disk that survives stop/start + volumes |
| Job queue / batch priority | Interruptible queue with auto-requeue and volume checkpoints |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu rtx-5090 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $0.439/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your SaladCloud machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | SaladCloud | PowerGPU |
| --- | --- | --- |
| 1× RTX 5090, 730 hours on-demand | $183 | **$320** |
| 1× RTX 5090, 8 h/day × 20 days | $40 | **$70** |
| Same 160 hours, PowerGPU interruptible | — | **$35** |
| 500 GB of storage, one month | Ephemeral only — no persistent volumes | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## SaladCloud alternative: FAQ

**Is SaladCloud cheaper than PowerGPU?**

Per hour on consumer cards, yes: Salad lists the RTX 4090 at $0.16/hr and the RTX 5090 at $0.25/hr (checked 2026-09-03). PowerGPU's interruptible RTX 4090 and 5090 are priced at half of on-demand — below Salad on the 4090 — on datacenter hardware with SSH and persistent storage. On datacenter GPUs there is no comparison: Salad does not offer them.

**Can I get SSH access on SaladCloud?**

No — Salad runs containers on residential nodes without shell access or persistent disks. Every PowerGPU instance has SSH, mapped ports with TLS, optional Jupyter and volumes.

**Which is more reliable?**

PowerGPU on-demand instances run on Tier-III datacenter machines with a 99.9% SLA; Salad nodes are volunteer PCs that can leave at any moment. Salad is designed around that churn; stateful or latency-sensitive work is not.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on SaladCloud, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

SaladCloud and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with SaladCloud. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/salad · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/io-net -->

---
title: "io.net Alternative (2026): GPU Prices vs PowerGPU"
description: "io.net vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/io-net
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# io.net alternative: same NVIDIA GPUs, fixed prices 30% under the market

io.net aggregates GPUs from independent suppliers into a decentralized network and sells clusters by the hour, settled in USDC or its IO token. Prices vary by supplier and location; the pitch is scale and crypto-native payments rather than datacenter guarantees.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## io.net vs PowerGPU at a glance

io.net

Decentralized GPU network: clusters aggregated from independent suppliers, paid in crypto

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | io.net | PowerGPU |
| --- | --- | --- |
| Billing | Hourly, prepaid | Per second, no minimum, price locked at deploy |
| Payment | Crypto (USDC, IO) | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account and wallet | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Supplier-set prices vary | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Supplier locations worldwide | 32 regions on 5 continents, verified datacenters only |
| Minimums | Cluster deployments, hourly | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Depends on supplier | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not itemised | $0.01/GB in and out, every region |
| Access | Clusters (Ray/Kubernetes-style), portal, API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

io.net column: as published on io.net on 2026-09-03. Details change — verify before you decide.

## io.net vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on io.net's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | io.net list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (from; varies by supplier and location) | $2.10 | **$1.428** | $0.714 | −32% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which io.net price is quoted. Negative differences mean io.net is cheaper on that card.

## When to stay with io.net

- You want very large aggregated clusters and are happy to pay in IO tokens.
- Supplier-to-supplier variability in hardware, network and uptime is acceptable for your batch jobs.

## Why teams switch to PowerGPU

- Same crypto-native payments (USDT, BTC, Monero, Solana…) but on verified Tier-III datacenter machines with a 99.9% SLA.
- Fixed, published prices at market median × 0.70 instead of supplier-set rates — and per-second billing.
- Single GPUs in 30 seconds, persistent volumes, SSH/Jupyter, full VMs, 76 models from RTX 3060 to B300.

## Switching from io.net: what maps to what

| On io.net | On PowerGPU |
| --- | --- |
| Cluster of supplier GPUs | Instance (1–8 GPUs on one verified machine) or InfiniBand cluster |
| USDC / IO payment | USDT, BTC, XMR, LTC, ETH, TRX or SOL top-up |
| Supplier storage | Volumes, $0.08/GB/month |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your io.net machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | io.net | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $1,533 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $336 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Depends on supplier | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## io.net alternative: FAQ

**Is PowerGPU cheaper than io.net?**

io.net advertises H100 SXM capacity from about $2.10/hr depending on supplier (checked 2026-09-03); PowerGPU fixes the H100 SXM at the public market median × 0.70 on datacenter machines, with a flat −50% interruptible tier below that.

**Can I pay PowerGPU in crypto like io.net?**

Yes — USDT (TRC-20 or ERC-20), Bitcoin, Monero, Litecoin, Ethereum, TRON and Solana, with no KYC. The balance is kept in USD and drawn down per second.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on io.net, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

io.net and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with io.net. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/io-net · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/modal -->

---
title: "Modal Alternative (2026): GPU Prices vs PowerGPU"
description: "Modal vs PowerGPU: B200 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/modal
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Modal alternative: same NVIDIA GPUs, fixed prices 30% under the market

Modal is a serverless platform where Python functions decorated with a GPU spec run in containers that scale from zero to thousands and bill per second — GPU, CPU and memory as separate meters. It is excellent for bursty inference and jobs; it is not a place to keep a machine, SSH in and leave Jupyter running.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Modal vs PowerGPU at a glance

Modal

Python-native serverless: functions on GPUs, per-second, scale to zero

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Modal | PowerGPU |
| --- | --- | --- |
| Billing | Per second — GPU + CPU ($0.047/core-hr) + memory ($0.008/GiB-hr) separately | Per second, no minimum, price locked at deploy |
| Payment | Card | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card or GitHub account | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | $30/month (Starter), $100/month (Team) | None — every hour is ≥30% under the market median instead |
| Cheaper tier | None | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | United States (+ selected) | 32 regions on 5 continents, verified datacenters only |
| Minimums | None | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Volumes $0.09/GiB/mo, first 1 TiB free | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free | $0.01/GB in and out, every region |
| Access | Python SDK only (no SSH, no long-lived VMs) | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Modal column: as published on modal.com on 2026-09-03. Details change — verify before you decide.

## Modal vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Modal's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Modal list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [B200](https://powergpu.ai/gpu/b200) ($0.001736/s) | $6.25 | **$5.425** | $2.712 | −13% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) ($0.001261/s) | $4.54 | **$2.791** | $1.395 | −39% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) ($0.001097/s) | $3.95 | **$1.428** | $0.714 | −64% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (80 GB, $0.000694/s) | $2.50 | **$0.560** | $0.280 | −78% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) ($0.000542/s) | $1.95 | **$0.514** | $0.257 | −74% on PowerGPU |
| [A10](https://powergpu.ai/gpu/a10) ($0.000306/s) | $1.10 | **$0.168** | $0.084 | −85% on PowerGPU |
| [L4](https://powergpu.ai/gpu/l4) ($0.000222/s) | $0.80 | **$0.225** | $0.112 | −72% on PowerGPU |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) ($0.000164/s) | $0.59 | **$0.103** | $0.051 | −83% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Modal price is quoted. Negative differences mean Modal is cheaper on that card.

## When to stay with Modal

- Your workload is bursty inference written in Python and you want autoscaling without operating anything.
- The $30/month free credit covers your usage.
- You never need a shell, a desktop, a VM or a long-running notebook.

## Why teams switch to PowerGPU

- Long-running machines: SSH, Jupyter, full KVM VMs, desktops — Modal has no persistent instances.
- One price per GPU-hour that includes the CPU and RAM; Modal meters GPU, CPU and memory separately.
- Fixed rates at market median × 0.70 (Modal H100 $3.95, A100 80 GB $2.50), interruptible −50%, reserved −35%.
- Crypto settlement with no card or KYC — and a serverless endpoint product when you do want scale-to-zero.

## Switching from Modal: what maps to what

| On Modal | On PowerGPU |
| --- | --- |
| @app.function(gpu="H100") | Serverless endpoint (vLLM / ComfyUI / any container) or an instance you keep |
| Modal Volume | Volume, $0.08/GB/month |
| Container image definition | Any OCI image or one of 37 templates |
| Modal CLI | powergpu CLI, REST API, Python SDK |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu b200 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $5.425/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Modal machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Modal | PowerGPU |
| --- | --- | --- |
| 1× B200, 730 hours on-demand | $4,563 | **$3,960** |
| 1× B200, 8 h/day × 20 days | $1,000 | **$868** |
| Same 160 hours, PowerGPU interruptible | — | **$434** |
| 500 GB of storage, one month | Volumes $0.09/GiB/mo, first 1 TiB free | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Modal alternative: FAQ

**Is PowerGPU cheaper than Modal?**

Per GPU-hour, yes: Modal bills the H100 at $3.95/hr and the A100 80 GB at $2.50/hr plus CPU and memory meters (checked 2026-09-03); PowerGPU fixes them at the public market median × 0.70, CPU and RAM included. Modal's $30 monthly credit can make tiny workloads free.

**Does PowerGPU have scale-to-zero like Modal?**

Yes — serverless endpoints with queue-based autoscaling, cached weights and per-second billing on the same fixed prices. The difference is that PowerGPU also offers instances, VMs and SSH when a function is not the right shape.

**Can I keep using Python to drive PowerGPU?**

Yes: pip install powergpu gives a typed SDK with a context manager that guarantees the meter stops when your block exits.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Modal, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Modal and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Modal. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/modal · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/jarvislabs -->

---
title: "Jarvislabs Alternative (2026): GPU Prices vs PowerGPU"
description: "Jarvislabs vs PowerGPU: H200 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/jarvislabs
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Jarvislabs alternative: same NVIDIA GPUs, fixed prices 30% under the market

Jarvislabs sells on-demand and spot GPU instances — H100, H200, A100, RTX PRO 6000, L4 — with per-minute billing, pause/resume, templates and serverless, from datacenters in India and Europe. Payment is by card, UPI or PayPal.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Jarvislabs vs PowerGPU at a glance

Jarvislabs

India- and EU-based GPU cloud with per-minute billing and pause/resume

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Jarvislabs | PowerGPU |
| --- | --- | --- |
| Billing | Per minute | Per second, no minimum, price locked at deploy |
| Payment | Card, UPI, PayPal | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot up to −56% | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | India (Noida) and Europe | 32 regions on 5 continents, verified datacenters only |
| Minimums | None | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | $0.10/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free in-region | $0.01/GB in and out, every region |
| Access | VMs, templates, Jupyter, SSH, API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Jarvislabs column: as published on jarvislabs.ai on 2026-09-03. Details change — verify before you decide.

## Jarvislabs vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Jarvislabs's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Jarvislabs list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H200](https://powergpu.ai/gpu/h200) (SXM · spot up to −56%) | $3.99 | **$2.791** | $1.395 | −30% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (spot up to −56%) | $2.69 | **$1.428** | $0.714 | −47% on PowerGPU |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) (RTX PRO 6000 Blackwell) | $1.89 | **$1.040** | $0.520 | −45% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (80 GB) | $1.49 | **$0.560** | $0.280 | −62% on PowerGPU |
| [L4](https://powergpu.ai/gpu/l4) | $0.44 | **$0.225** | $0.112 | −49% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Jarvislabs price is quoted. Negative differences mean Jarvislabs is cheaper on that card.

## When to stay with Jarvislabs

- You are in India and need a local region with UPI payments.
- Pause/resume of a single workstation is your main workflow and you like Jarvislabs' simple UI.

## Why teams switch to PowerGPU

- Price: H100 SXM $2.69 and A100 80 GB $1.49 at Jarvislabs versus market median × 0.70 fixed here.
- 32 regions on five continents, consumer cards (RTX 5090, 4090, 3090) and 76 models in total.
- Crypto payments with no KYC, per-second billing, flat −50% interruptible instead of spot.

## Switching from Jarvislabs: what maps to what

| On Jarvislabs | On PowerGPU |
| --- | --- |
| Instance (pause/resume) | Instance (stop keeps the disk, start resumes) |
| Template | One of 37 templates or any image |
| Spot instance | Interruptible instance, flat −50% |
| Storage | Volume, $0.08/GB/month |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h200 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $2.791/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Jarvislabs machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Jarvislabs | PowerGPU |
| --- | --- | --- |
| 1× H200, 730 hours on-demand | $2,913 | **$2,037** |
| 1× H200, 8 h/day × 20 days | $638 | **$447** |
| Same 160 hours, PowerGPU interruptible | — | **$223** |
| 500 GB of storage, one month | $0.10/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Jarvislabs alternative: FAQ

**Is PowerGPU cheaper than Jarvislabs?**

Yes on every overlapping card: Jarvislabs lists the H100 SXM at $2.69/hr and the A100 80 GB at $1.49/hr (checked 2026-09-03); PowerGPU fixes them at the public market median × 0.70, and its interruptible tier is a flat −50%.

**Does PowerGPU support pause and resume?**

Yes — stop an instance and GPU billing ends that second while the disk waits at $0.08/GB/month; start resumes on the same data.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Jarvislabs, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Jarvislabs and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Jarvislabs. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/jarvislabs · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/aws -->

---
title: "AWS GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "AWS vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/aws
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# AWS GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

Amazon EC2 sells GPUs as instance families — P5 (8× H100), P4d (8× A100 40 GB), G6e (L40S), G5 (A10G), G4dn (T4) — billed per second with a one-minute minimum, plus Spot, Savings Plans and prepaid Capacity Blocks. The integration with S3, IAM, VPC and SageMaker is unmatched; the list prices and quota process are the cost.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## AWS vs PowerGPU at a glance

AWS

Amazon EC2 GPU instances: P5 (H100), P4d (A100), G6e (L40S), G5 (A10G), G4dn (T4)

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | AWS | PowerGPU |
| --- | --- | --- |
| Billing | Per second, 60-second minimum | Per second, no minimum, price locked at deploy |
| Payment | Card, invoicing | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card and identity verification; GPU quotas must be requested | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Free tier excludes GPU instances | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot (−50–70%, interruptible), Savings Plans, Capacity Blocks | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 30+ | 32 regions on 5 continents, verified datacenters only |
| Minimums | P5 and P4d are 8-GPU instances only | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | EBS gp3 $0.08/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | $0.09/GB after the first 100 GB | $0.01/GB in and out, every region |
| Access | VMs, SageMaker, EKS, full AWS API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

AWS column: as published on aws.amazon.com on 2026-09-03. Details change — verify before you decide.

## AWS vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on AWS's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | AWS list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (p5.48xlarge on-demand, us-east-1, per GPU · Capacity Blocks ~$5.19) | $6.88 | **$1.428** | $0.714 | −79% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (p4d.24xlarge, 40 GB per GPU, per GPU) | $4.10 | **$0.560** | $0.280 | −86% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (g6e.xlarge) | $1.86 | **$0.514** | $0.257 | −72% on PowerGPU |
| [A10](https://powergpu.ai/gpu/a10) (g5.xlarge (A10G 24 GB)) | $1.01 | **$0.168** | $0.084 | −83% on PowerGPU |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) (g4dn.xlarge) | $0.53 | **$0.103** | $0.051 | −81% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which AWS price is quoted. Negative differences mean AWS is cheaper on that card.

## When to stay with AWS

- Your data lives in S3 with IAM policies and compliance frameworks you cannot re-certify elsewhere.
- You use SageMaker, Bedrock or EKS as the platform, not just the GPU.
- An enterprise agreement already discounts your list prices heavily.

## Why teams switch to PowerGPU

- Price: an H100 SXM on p5.48xlarge is $6.88 per GPU-hour on-demand; PowerGPU fixes it at the public market median × 0.70 — a several-fold gap on the same silicon.
- One GPU when you need one GPU: P5 and P4d only come as 8-GPU instances; PowerGPU rents 1× to 8× at the same per-GPU price.
- No quota requests, no card, no KYC: deploy in 30 seconds after a crypto top-up.
- Egress at $0.01/GB versus $0.09/GB — a 10 TB result set costs $100 here, about $900 on AWS.

## Switching from AWS: what maps to what

| On AWS | On PowerGPU |
| --- | --- |
| p5.48xlarge (8× H100) | 8× H100 SXM machine, or 1×/2×/4× at the same per-GPU price |
| g6e / g5 instance | L40S / A10 instance, or an RTX 5090 at a fraction of the price |
| EBS volume | Volume, $0.08/GB/month |
| Spot instance | Interruptible instance, flat −50%, auto-requeue |
| Deep Learning AMI | PyTorch / NGC / CUDA templates or a full Ubuntu VM |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your AWS machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | AWS | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $5,022 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $1,101 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | EBS gp3 $0.08/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## AWS GPU alternative: FAQ

**How much cheaper is PowerGPU than AWS for an H100?**

AWS lists p5.48xlarge at $55.04/hr on-demand in us-east-1 — $6.88 per H100 (checked 2026-09-03); PowerGPU fixes the H100 SXM at the public market median × 0.70. Even AWS Capacity Blocks (~$5.19/GPU-hr, prepaid) stay far above.

**Can I rent a single H100 on PowerGPU?**

Yes — 1×, 2×, 4× or 8× per machine at the same per-GPU price, deployed in about 30 seconds. On AWS the P5 family only exists as 8-GPU instances.

**What about data in S3?**

Pull it once onto a PowerGPU volume (AWS egress applies on their side), then every instance mounts it from NVMe. Results go back for $0.01/GB.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on AWS, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

AWS and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with AWS. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/aws · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/google-cloud -->

---
title: "Google Cloud GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "Google Cloud vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/google-cloud
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Google Cloud GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

Google Cloud sells GPUs as accelerator-optimized machine types — A3 (8× H100), A2 (A100), G2 (L4) — billed per second with a one-minute minimum, discounted by Spot VMs and committed-use contracts, and wrapped by Vertex AI. Quotas start at zero and must be requested.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Google Cloud vs PowerGPU at a glance

Google Cloud

Compute Engine A3 (H100), A2 (A100), G2 (L4) machine types and Vertex AI

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Google Cloud | PowerGPU |
| --- | --- | --- |
| Billing | Per second, 1-minute minimum | Per second, no minimum, price locked at deploy |
| Payment | Card, invoicing | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card and verification; GPU quota requests | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | $300 trial credit (GPU quota usually 0) | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot VMs (−60–90%), committed use (−37–55%) | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 40+ | 32 regions on 5 continents, verified datacenters only |
| Minimums | A3 = 8-GPU VMs | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Persistent Disk / Hyperdisk from $0.10/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | $0.08–0.12/GB | $0.01/GB in and out, every region |
| Access | VMs, GKE, Vertex AI, full GCP API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Google Cloud column: as published on cloud.google.com on 2026-09-03. Details change — verify before you decide.

## Google Cloud vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Google Cloud's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Google Cloud list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (a3-highgpu-8g, us-central1 list, per GPU) | $11.06 | **$1.428** | $0.714 | −87% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (a2-highgpu-1g, 40 GB per GPU) | $3.67 | **$0.560** | $0.280 | −85% on PowerGPU |
| [L4](https://powergpu.ai/gpu/l4) (g2-standard-4) | $0.70 | **$0.225** | $0.112 | −68% on PowerGPU |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) (T4 attached to an N1 VM, GPU only) | $0.35 | **$0.103** | $0.051 | −71% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Google Cloud price is quoted. Negative differences mean Google Cloud is cheaper on that card.

## When to stay with Google Cloud

- Vertex AI, BigQuery and GCS are the backbone of your pipeline.
- You have committed-use contracts or TPU workloads that only exist on Google Cloud.
- Your compliance posture is certified against GCP specifically.

## Why teams switch to PowerGPU

- Price: A3 H100 lists at about $11 per GPU-hour; PowerGPU fixes the H100 SXM at the public market median × 0.70.
- Single GPUs without quota tickets — A3 is 8-GPU only and new projects start with a GPU quota of zero.
- Crypto, no KYC, per-second billing, flat egress at $0.01/GB.

## Switching from Google Cloud: what maps to what

| On Google Cloud | On PowerGPU |
| --- | --- |
| a3-highgpu-8g | 8× H100 SXM machine, or 1×–4× at the same per-GPU price |
| a2 / g2 VM | A100 / L4 instance |
| Spot VM | Interruptible instance, flat −50% |
| Persistent Disk | Volume, $0.08/GB/month |
| Deep Learning VM image | PyTorch / NGC templates or a full Ubuntu VM |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Google Cloud machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Google Cloud | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $8,074 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $1,770 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Persistent Disk / Hyperdisk from $0.10/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Google Cloud GPU alternative: FAQ

**How does PowerGPU compare with Google Cloud A3 pricing?**

A3 H100 machine types list at roughly $11 per GPU-hour on-demand in us-central1 (checked 2026-09-03); PowerGPU fixes the H100 SXM at the public market median × 0.70 and rents from one card. Google's Spot VMs narrow the gap but can be preempted with 30 seconds' notice.

**Do I need a quota increase on PowerGPU?**

No. New accounts have modest default limits that raise automatically with history; nothing starts at zero.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Google Cloud, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Google Cloud and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Google Cloud. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/google-cloud · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/azure -->

---
title: "Azure GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "Microsoft Azure vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-03."
url: https://powergpu.ai/alternatives/azure
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-03

# Azure GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

Azure sells GPUs as VM series — ND H100 v5 (8× H100 with InfiniBand), NC A100 v4, NC T4 v3 — billed per second, with Spot VMs and one- to three-year reservations, and enterprise agreements for large customers. GPU quotas are zero by default and require a support request.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Microsoft Azure vs PowerGPU at a glance

Microsoft Azure

ND H100 v5, NC A100 v4 and NC T4 virtual machines with enterprise agreements

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Microsoft Azure | PowerGPU |
| --- | --- | --- |
| Billing | Per second | Per second, no minimum, price locked at deploy |
| Payment | Card, enterprise agreement / invoicing | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card and identity verification; GPU quota requests | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | $200 trial credit (GPU quota usually 0) | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot VMs, reservations (1–3 years) | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 60+ | 32 regions on 5 continents, verified datacenters only |
| Minimums | ND H100 v5 = 8 GPUs | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Managed disks from ~$0.08/GB/mo | $0.08/GB/month NVMe, volumes survive instances |
| Egress | $0.087/GB after 100 GB | $0.01/GB in and out, every region |
| Access | VMs, AKS, Azure ML, full Azure API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Microsoft Azure column: as published on azure.microsoft.com on 2026-09-03. Details change — verify before you decide.

## Microsoft Azure vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Microsoft Azure's pricing page (2026-09-03) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Microsoft Azure list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (ND H100 v5, 8 GPUs = $98.32/hr, East US, per GPU) | $12.29 | **$1.428** | $0.714 | −88% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (NC24ads A100 v4, 80 GB) | $3.67 | **$0.560** | $0.280 | −85% on PowerGPU |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) (NC4as T4 v3) | $0.53 | **$0.103** | $0.051 | −81% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Microsoft Azure price is quoted. Negative differences mean Microsoft Azure is cheaper on that card.

## When to stay with Microsoft Azure

- Azure ML, Entra ID and enterprise agreements are non-negotiable in your organisation.
- You are consuming Azure OpenAI or other Azure-only services next to the GPUs.
- A multi-year reservation already discounts your rates.

## Why teams switch to PowerGPU

- Price: ND H100 v5 lists at $98.32/hr for eight GPUs — $12.29 per H100 — versus market median × 0.70 fixed here.
- Single GPUs in 30 seconds, no quota request, no card, no KYC.
- Consumer and workstation cards Azure does not offer, per-second billing and a flat −50% interruptible tier.

## Switching from Microsoft Azure: what maps to what

| On Microsoft Azure | On PowerGPU |
| --- | --- |
| ND H100 v5 (8×) | 8× H100 SXM machine or 1×–4× at the same per-GPU price |
| NC A100 v4 / NC T4 v3 | A100 / T4 instance |
| Spot VM | Interruptible instance, flat −50% |
| Managed disk | Volume, $0.08/GB/month |
| Azure ML compute | PyTorch / NGC templates, Jupyter, REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Microsoft Azure machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Microsoft Azure | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $8,972 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $1,966 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Managed disks from ~$0.08/GB/mo | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Azure GPU alternative: FAQ

**How much cheaper is PowerGPU than Azure for H100s?**

Azure ND H100 v5 lists at $98.32/hr for 8 GPUs in East US — $12.29 per H100 (checked 2026-09-03); PowerGPU fixes the H100 SXM at the public market median × 0.70, several times lower, and rents from a single card.

**Can I get InfiniBand clusters outside Azure?**

Yes — PowerGPU clusters connect 8× SXM nodes over InfiniBand from 16 to 512 GPUs, quoted per GPU from the public sheet, without an enterprise agreement.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Microsoft Azure, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Microsoft Azure and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Microsoft Azure. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/azure · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/akash -->

---
title: "Akash Network Alternative (2026): GPU Prices vs PowerGPU"
description: "Akash Network vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/akash
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Akash Network alternative: same NVIDIA GPUs, fixed prices 30% under the market

Akash Network is a decentralized, blockchain-based marketplace (Cosmos SDK, its own AKT token) where independent, permissionless providers bid to run workloads — GPU rentals included — with no company operating the datacenters itself. A deployment is described in an SDL (YAML) file, funded from a Cosmos wallet (typically Keplr) and settled on-chain in AKT or USDC through an escrow contract; the lowest qualifying bid usually wins. No account or identity check exists beyond connecting a wallet, and GPU listings — H100, H200, A100 and consumer cards — move in the Akash Console as providers join, leave or requote.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Akash Network vs PowerGPU at a glance

Akash Network

Decentralized GPU marketplace: permissionless providers bid, settled on-chain in AKT or USDC

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Akash Network | PowerGPU |
| --- | --- | --- |
| Billing | Hourly, on-chain escrow funds the lease; live marketplace bid price | Per second, no minimum, price locked at deploy |
| Payment | AKT or USDC on-chain, via a Cosmos wallet (Keplr); no card | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | None — permissionless, wallet-funded; no KYC to deploy or to provide | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None published | None — every hour is ≥30% under the market median instead |
| Cheaper tier | None beyond the bidding itself — the live price already reflects provider competition; no separate spot or reserved tier | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | No fixed region list, provider location shown per listing; roughly 60-90 active providers network-wide (third-party trackers, mid-2026), 7 for the A100 specifically | 32 regions on 5 continents, verified datacenters only |
| Minimums | None stated; whole GPUs specified in the SDL deployment file | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Provider-set, not centrally published | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not itemised on public pages | $0.01/GB in and out, every region |
| Access | Console (web), CLI, SDL (YAML) deployment spec | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Akash Network column: as published on akash.network on 2026-09-04. Details change — verify before you decide.

## Akash Network vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Akash Network's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Akash Network list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (from; mixed form factors, marketplace bid, ~$2.52/hr average) | $2.04 | **$1.428** | $0.714 | −30% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (from; 141 GB HBM3e SXM, lowest rate found across providers surveyed) | $4.45 | **$2.791** | $1.395 | −37% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (from; 80 GB HBM2e, marketplace bid, average $1.54/hr, up to $1.83) | $1.07 | **$0.560** | $0.280 | −48% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Akash Network price is quoted. Negative differences mean Akash Network is cheaper on that card.

## When to stay with Akash Network

- You are already holding AKT, or want exposure to a Cosmos-native marketplace with no company in the loop.
- A wallet and an SDL file are an acceptable deployment workflow for your team, in place of a dashboard or REST API.
- The cheapest live bid on a given GPU, moving with the market, matters more to you than a fixed published price.

## Why teams switch to PowerGPU

- A fixed price at the public market median × 0.70 instead of a live bid that read $1.07-$1.83/hr on the A100 (avg $1.54, checked 2026-09-04) and $2.04-$2.52/hr on the H100 (checked 2026-09-04) inside the same month.
- Email sign-up and a crypto top-up rather than a Cosmos wallet (Keplr) and an SDL/YAML file as the only way in.
- SSH, Jupyter, a dashboard, CLI and REST API on every instance — Akash publishes no access path beyond its own console and CLI.
- Datacenter-verified machines with a 99.9% SLA across 32 fixed regions, against the provider-to-provider variance a bidding model does not screen for.

## Switching from Akash Network: what maps to what

| On Akash Network | On PowerGPU |
| --- | --- |
| Provider bid (lease) | Instance on a verified Tier-III datacenter — the same fixed price on every machine |
| SDL (YAML) deployment | CLI / REST API / console deploy, no manifest required |
| AKT / USDC escrow | USDT, BTC, XMR, LTC, ETH, TRX or SOL top-up |
| Akash Console | powergpu dashboard / CLI / REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Akash Network machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Akash Network | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $1,489 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $326 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Provider-set, not centrally published | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Akash Network alternative: FAQ

**Is PowerGPU cheaper than Akash?**

It depends on the moment: Akash is a live bidding market, and its own guides show the A100 80 GB moving between $1.07 and $1.83/hr (average $1.54) and the H100 between about $2.04 and $2.52/hr within August 2026 (checked 2026-09-04). PowerGPU fixes both at the public market median × 0.70 and does not change while an instance is running, or when the next one is deployed.

**Do I need a crypto wallet to use PowerGPU the way I do on Akash?**

No Cosmos wallet or SDL file: create an account with an email, top up in USDT, BTC, Monero, Litecoin, Ethereum, TRON or Solana, and deploy from a dashboard, CLI or REST API. Payment is still crypto-settled and KYC-free, matching Akash's no-KYC deployment model.

**Does PowerGPU verify its datacenters the way a centralized cloud does?**

Yes — every machine sits in an audited Tier-III facility with a 99.9% SLA. Akash is intentionally the opposite: any permissionless provider can list capacity, which is what keeps its bid prices low and its hardware guarantees variable.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Akash Network, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Akash Network and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Akash Network. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/akash · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/crusoe -->

---
title: "Crusoe Alternative (2026): GPU Prices vs PowerGPU"
description: "Crusoe Cloud vs PowerGPU: H200 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/crusoe
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Crusoe alternative: same NVIDIA GPUs, fixed prices 30% under the market

Crusoe (founded 2018 by Chase Lochmiller and Cully Cavness) began as a Digital Flare Mitigation business, converting natural gas that oil wells would otherwise burn off into on-site power for mobile data centers, and used that power first to mine bitcoin before selling the mining/DFM unit to NYDIG in late 2024 to focus entirely on AI cloud. Crusoe Cloud now sells on-demand GPU instances — H100, H200, A100 and L40S on a published rate card, plus AMD MI300X — while Blackwell-class B200, GB200 NVL72 and AMD's MI355X are contact-sales only, alongside Managed Kubernetes, Managed Slurm and S3-compatible storage. The company is also building gigawatt-scale "AI Factory" campuses, including a large site in Abilene, Texas, sold separately as dedicated capacity.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Crusoe Cloud vs PowerGPU at a glance

Crusoe Cloud

AI cloud built on stranded and flared energy: H100, H200, A100, L40S on a published rate card

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Crusoe Cloud | PowerGPU |
| --- | --- | --- |
| Billing | Hourly on-demand; reserved/committed capacity quoted by sales | Per second, no minimum, price locked at deploy |
| Payment | Not published on the pricing or product pages | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Not published beyond standard account sign-up | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | $5 credit for Crusoe Intelligence Foundry (serverless fine-tuning/inference) — not a general compute credit | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Reserved capacity via sales, described as the lowest rates; spot listed as a capacity option with no published discount | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 5 regions — Houston, Virginia and Nevada (US), plus Iceland and Norway (Europe); gigawatt-scale campuses such as Abilene, TX are sold as separate dedicated deals | 32 regions on 5 continents, verified datacenters only |
| Minimums | None published for on-demand GPU instances | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Persistent disk $0.08/GiB/month · shared disk $0.07/GiB/month · S3-compatible object storage $0.06/GiB/month | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free — no charge for network ingress or egress | $0.01/GB in and out, every region |
| Access | Console, CLI, REST API, Terraform and SDKs; Crusoe Managed Kubernetes and Managed Slurm for clusters | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Crusoe Cloud column: as published on crusoe.ai on 2026-09-04. Details change — verify before you decide.

## Crusoe Cloud vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Crusoe Cloud's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Crusoe Cloud list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H200](https://powergpu.ai/gpu/h200) (HGX H200, 141 GB) | $4.29 | **$2.791** | $1.395 | −35% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (HGX H100, 80 GB) | $3.90 | **$1.428** | $0.714 | −63% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (80 GB SXM) | $2.30 | **$0.560** | $0.280 | −76% on PowerGPU |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) (80 GB PCIe) | $2.00 | **$0.374** | $0.187 | −81% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (48 GB) | $1.50 | **$0.514** | $0.257 | −66% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Crusoe Cloud price is quoted. Negative differences mean Crusoe Cloud is cheaper on that card.

## When to stay with Crusoe Cloud

- You need AMD MI300X ($3.45/hr published) or MI355X, or Blackwell-class B200/GB200 NVL72 rack-scale capacity — PowerGPU's catalog is NVIDIA GPU instances up to B300, not AMD accelerators or Grace-Blackwell superchip racks.
- Unconditional free bandwidth, not just a low flat rate, is a hard requirement for a very egress-heavy workload.
- Sourcing compute from stranded and flared energy specifically matters for your own sustainability reporting.

## Why teams switch to PowerGPU

- Price: the H100 HGX is fixed at market median × 0.70 here versus $3.90/hr at Crusoe on-demand, and the H200 gap is wider still ($4.29/hr).
- No card or KYC anywhere in Crusoe's public pricing or product pages — top up in crypto and deploy.
- 32 regions on five continents versus Crusoe's 5, concentrated in the US and two Nordic countries.
- A flat −50% interruptible tier and a published −35% reserved tier from 3 months, instead of a "contact sales for our lowest rates" reserved tier.

## Switching from Crusoe Cloud: what maps to what

| On Crusoe Cloud | On PowerGPU |
| --- | --- |
| On-demand GPU instance | Instance, 1× to 8× GPUs at the same per-GPU price |
| Reserved/committed capacity (sales quote) | Reserved instance, flat −35% from 3 months, no sales call |
| Crusoe Managed Kubernetes / Managed Slurm | Clusters — InfiniBand-connected multi-node pods on request |
| Persistent disk / object storage | Volume, $0.08/GB/month |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h200 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $2.791/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Crusoe Cloud machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Crusoe Cloud | PowerGPU |
| --- | --- | --- |
| 1× H200, 730 hours on-demand | $3,132 | **$2,037** |
| 1× H200, 8 h/day × 20 days | $686 | **$447** |
| Same 160 hours, PowerGPU interruptible | — | **$223** |
| 500 GB of storage, one month | Persistent disk $0.08/GiB/month · shared disk $0.07/GiB/month · S3-compatible object storage $0.06/GiB/month | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Crusoe alternative: FAQ

**Is PowerGPU cheaper than Crusoe Cloud?**

Yes on every published overlapping card: Crusoe prices the H100 HGX at $3.90/hr, the H200 HGX at $4.29/hr, the A100 SXM at $2.30/hr and the A100 PCIe at $2.00/hr on-demand (checked 2026-09-04); PowerGPU fixes all of them at the public market median × 0.70. Crusoe's B200 and GB200 NVL72 are contact-sales only, with no published rate to compare.

**Does Crusoe really not charge for bandwidth?**

Correct — Crusoe Cloud states it does not charge for network ingress or egress (checked 2026-09-04), the one place its published terms beat PowerGPU's flat $0.01/GB.

**What happened to Crusoe's bitcoin-mining business?**

Sold to NYDIG in late 2024. Crusoe started in 2018 by burning stranded, otherwise-flared oil-well gas to power mining rigs; after the sale it kept the power strategy and put it entirely behind AI cloud and dedicated "AI Factory" campuses instead.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Crusoe Cloud, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Crusoe Cloud and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Crusoe Cloud. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/crusoe · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/cudo-compute -->

---
title: "Cudo Compute Alternative (2026): GPU Prices vs PowerGPU"
description: "Cudo Compute vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/cudo-compute
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Cudo Compute alternative: same NVIDIA GPUs, fixed prices 30% under the market

Cudo Compute positions itself as an NVIDIA Cloud Partner selling GPU VMs across partner datacenters, from RTX-class cards up to H100, H200, B200, GB200 NVL72 and B300. It previously published self-serve per-GPU rates; as of this check every GPU on its live pricing page shows "Quote on request" instead of a checkout price. The figures below come from Cudo's own 2025 blog post comparing hyperscaler costs, not from the current page, and should be read as historical rather than confirmed today.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Cudo Compute vs PowerGPU at a glance

Cudo Compute

NVIDIA Cloud Partner marketplace; live pricing page is quote-only, not self-serve

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Cudo Compute | PowerGPU |
| --- | --- | --- |
| Billing | Historically hourly against a prepaid credit balance; the live pricing page (checked 2026-09-04) is quote-based for every GPU rather than self-checkout | Per second, no minimum, price locked at deploy |
| Payment | Card only — Visa/Mastercard, per Cudo's own documentation; Apple Pay and Google Pay explicitly not supported, no crypto option published | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Billing account with a card on file before deploying | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Not found published | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Not published — no self-serve spot/interruptible tier found | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Partner datacenters in multiple countries; not itemised on the current quote-gated pricing page | 32 regions on 5 continents, verified datacenters only |
| Minimums | Not published (pricing is quote-based) | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not published on the current pricing page; a 2025 source notes storage is bundled into cluster-rate quotes | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not published | $0.01/GB in and out, every region |
| Access | VMs, API, per Cudo's documentation | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Cudo Compute column: as published on cudocompute.com on 2026-09-04. Details change — verify before you decide.

## Cudo Compute vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Cudo Compute's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Cudo Compute list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (per Cudo's own 2025 blog post, flat on-demand; the live pricing page (checked 2026-09-04) shows "Quote on request" instead) | $2.25 | **$1.428** | $0.714 | −37% on PowerGPU |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (per Cudo's own 2025 blog post; live pricing page (checked 2026-09-04) shows "Quote on request") | $2.47 | **$1.867** | $0.933 | −24% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Cudo Compute price is quoted. Negative differences mean Cudo Compute is cheaper on that card.

## When to stay with Cudo Compute

- You specifically need Cudo's partner-datacenter marketplace model, or already have a negotiated quote or contract with them.
- You need GB200 NVL72-class rack systems and are fine going through a sales-quote process to get them.
- Your billing is already integrated with Cudo's account/API tooling.

## Why teams switch to PowerGPU

- A published, self-serve price you can act on today at market median × 0.70 — Cudo's live pricing page (checked 2026-09-04) shows "Quote on request" for every single GPU, with no checkout price at all.
- On the only numbers Cudo has published anywhere (its own 2025 blog: $2.25/hr H100 SXM, $2.47/hr H100 PCIe), PowerGPU's market-median × 0.70 pricing is openly dated and re-checked, not a year-old blog figure.
- Crypto with no KYC, versus Cudo's card-only billing — Visa/Mastercard only, no Apple Pay, no Google Pay, no cryptocurrency despite the unrelated CUDOS-named token.
- Consumer cards (RTX 5090, 4090, 3090) that a partner-datacenter enterprise catalog like Cudo's does not carry.

## Switching from Cudo Compute: what maps to what

| On Cudo Compute | On PowerGPU |
| --- | --- |
| GPU VM (quote) | Instance, self-serve at a published price |
| Billing account + card on file | Crypto top-up — no card, no KYC |
| Cluster quote | Clusters (16–512 GPUs), priced per GPU from the public sheet |
| Cudo REST API | powergpu CLI / REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Cudo Compute machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Cudo Compute | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $1,643 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $360 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Not published on the current pricing page; a 2025 source notes storage is bundled into cluster-rate quotes | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Cudo Compute alternative: FAQ

**Does Cudo Compute take crypto payments?**

No — per Cudo's own documentation, billing accounts take Visa/Mastercard only; Apple Pay and Google Pay are explicitly not supported, and no cryptocurrency option is published (checked 2026-09-04), despite the unrelated CUDOS token sharing the company's name. PowerGPU is crypto-only: USDT, BTC, XMR, LTC, ETH, TRX, SOL, no card, no KYC.

**Is PowerGPU cheaper than Cudo Compute?**

The only Cudo H100 prices published anywhere come from its own 2025 blog post — $2.25/hr SXM, $2.47/hr PCIe — since the live pricing page (checked 2026-09-04) shows "Quote on request" for every GPU instead. PowerGPU fixes the H100 SXM at the public market median × 0.70 and publishes it, no quote required.

**Why does Cudo Compute show "Quote on request" instead of prices?**

Not stated on the site; it reads as a shift toward enterprise, sales-assisted pricing on what used to be a self-serve marketplace (checked 2026-09-04). PowerGPU's full price sheet stays public and dated on every alternative page.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Cudo Compute, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Cudo Compute and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Cudo Compute. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/cudo-compute · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/datacrunch -->

---
title: "DataCrunch Alternative (2026): GPU Prices vs PowerGPU"
description: "DataCrunch vs PowerGPU: B300 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/datacrunch
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# DataCrunch alternative: same NVIDIA GPUs, fixed prices 30% under the market

DataCrunch renamed itself Verda in November 2025; datacrunch.io now redirects to verda.com, and the team, infrastructure and customer accounts carried over unchanged. Founded in Helsinki in 2018, Verda owns and operates its own datacenters rather than reselling capacity, with two sites in Helsinki and one in Reykjanesbaer, Iceland, on renewable power. It raised $117M in 2026 to build what it calls a European hyperscaler and expand into the UK, the US and Asia; a fourth Nordic site (Akaa, Finland) is announced but not yet live.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## DataCrunch vs PowerGPU at a glance

DataCrunch

Renamed Verda in Nov. 2025: own Nordic datacenters, H100 through B300, no crypto payment

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | DataCrunch | PowerGPU |
| --- | --- | --- |
| Billing | Hourly rate shown per GPU, pay-as-you-go; sub-hour billing increment not published on Verda's own pricing or terms pages | Per second, no minimum, price locked at deploy |
| Payment | Card or debit by default; invoicing available at Verda's discretion; no cryptocurrency documented | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | No KYC clause in the Terms and Conditions; billing name, email, phone and address required to set up an account | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None advertised on the current pricing or company pages | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot tier at roughly −50% versus on-demand; reserved capacity from a 1-month minimum term | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 3 own-operated sites: Helsinki ×2 (Finland), Reykjanesbaer ×1 (Iceland); a fourth (Akaa, Finland) and international offices announced, none live yet | 32 regions on 5 continents, verified datacenters only |
| Minimums | None for pay-as-you-go; reserved capacity starts at 1 month | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | $0.20/GiB/month NVMe | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not published on Verda's pricing or FAQ pages | $0.01/GB in and out, every region |
| Access | Self-serve console and API, no sales call for the instances listed above | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

DataCrunch column: as published on datacrunch.io on 2026-09-04. Details change — verify before you decide.

## DataCrunch vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on DataCrunch's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | DataCrunch list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [B300](https://powergpu.ai/gpu/b300) (1× B300 SXM6 268GB) | $7.50 | **$6.737** | $3.368 | −10% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (1× B200 SXM6 180GB) | $6.11 | **$5.425** | $2.712 | −11% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (1× H200 SXM5 141GB) | $4.00 | **$2.791** | $1.395 | −30% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (1× H100 SXM5 80GB) | $3.25 | **$1.428** | $0.714 | −56% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (1× A100 SXM4 80GB; 40GB variant $1.29/h) | $1.79 | **$0.560** | $0.280 | −69% on PowerGPU |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) (1× RTX PRO 6000 96GB) | $1.89 | **$1.040** | $0.520 | −45% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (1× L40S 48GB) | $1.37 | **$0.514** | $0.257 | −62% on PowerGPU |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) (1× RTX 6000 Ada 48GB) | $1.04 | **$0.467** | $0.233 | −55% on PowerGPU |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) (1× RTX A6000 48GB) | $0.61 | **$0.281** | $0.140 | −54% on PowerGPU |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) (1× Tesla V100 16GB) | $0.17 | **$0.130** | $0.065 | −24% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which DataCrunch price is quoted. Negative differences mean DataCrunch is cheaper on that card.

## When to stay with DataCrunch

- Verda owns its Nordic datacenters end to end, power to network, rather than reselling capacity, which some buyers want for pricing and control stability.
- Blackwell-generation cards (B300, B200) are already listed alongside H200 and H100, useful if the newest silicon matters more than price.
- EU or Nordic data residency and hydro or renewable-powered sites are a hard requirement for compliance or sustainability reporting.

## Why teams switch to PowerGPU

- Price: Verda's H100 SXM is $3.25/hr and its A100 SXM4 80GB is $1.79/hr; PowerGPU fixes both at the public market median × 0.70.
- Geography: Verda runs 3 sites, all in Finland and Iceland, against 32 PowerGPU regions on five continents.
- Payment: Verda bills by card by default with no cryptocurrency option; PowerGPU is crypto-settled with no card and no KYC.
- A published egress rate, $0.01/GB here, where Verda's own pricing and FAQ pages currently list none.

## Switching from DataCrunch: what maps to what

| On DataCrunch | On PowerGPU |
| --- | --- |
| On-demand instance (verda.com/pricing) | Instance, container in 30 s or full KVM VM |
| Spot tier, about −50% | Interruptible flat −50% |
| Reserved, 1-month minimum | Reserved −35% (3+ months) |
| NVMe volume, $0.20/GiB/month | Volume, $0.08/GB/month |
| Card, or invoice by request | Crypto top-up, no card, no KYC |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu b300 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $6.737/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your DataCrunch machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | DataCrunch | PowerGPU |
| --- | --- | --- |
| 1× B300, 730 hours on-demand | $5,475 | **$4,918** |
| 1× B300, 8 h/day × 20 days | $1,200 | **$1,078** |
| Same 160 hours, PowerGPU interruptible | — | **$539** |
| 500 GB of storage, one month | $0.20/GiB/month NVMe | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## DataCrunch alternative: FAQ

**Is PowerGPU cheaper than Verda (formerly DataCrunch)?**

On every overlapping card, yes: Verda lists the H100 SXM at $3.25/hr and the A100 SXM4 80GB at $1.79/hr (checked 2026-09-04); PowerGPU fixes both at the public market median × 0.70. Verda's own spot tier, about −50%, narrows the gap but still sits on top of a higher on-demand rate.

**Is DataCrunch still around, or is it Verda now?**

Same company. DataCrunch renamed to Verda in November 2025; datacrunch.io now redirects to verda.com and existing accounts moved over automatically. This fiche uses Verda's current pricing and terms, checked 2026-09-04.

**Does Verda charge for bandwidth?**

Not published: neither Verda's pricing page nor its FAQ list an egress rate as of 2026-09-04. PowerGPU publishes a flat $0.01/GB in and out.

**Can I pay Verda in crypto?**

No. Verda's terms specify card or debit by default, with invoicing available at its discretion; no cryptocurrency option is documented (checked 2026-09-04). PowerGPU is crypto-only: USDT, BTC, XMR, LTC, ETH, TRX, SOL, with no card and no KYC.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on DataCrunch, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

DataCrunch and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with DataCrunch. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/datacrunch · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/fluidstack -->

---
title: "FluidStack Alternative (2026): GPU Prices vs PowerGPU"
description: "FluidStack vs PowerGPU: billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/fluidstack
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# FluidStack alternative: same NVIDIA GPUs, fixed prices 30% under the market

FluidStack (founded 2017 in London) started as a GPU rental broker but has pivoted hard toward large, invoiced infrastructure contracts: by the end of 2024 the company said about 62% of its revenue already came from multi-year "Private Cloud" deals versus 38% from its original on-demand marketplace, and 2026 reporting centers on a reported compute build-out tied to Anthropic and on capacity work referenced alongside Google, funded by a $1.5B round that valued the company at about $18B. Sites are reported in the UK, France, Iceland, Norway, Argentina and the US, sized for named-customer contracts. As of the date checked, fluidstack.io carries no rate card: the pricing page returns HTTP 404 and the homepage is careers-and-news content built around "gigawatts of compute in 6 months," not a self-serve console.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## FluidStack vs PowerGPU at a glance

FluidStack

Neocloud turned AI-factory builder: dedicated, multi-year contracts for a handful of large labs

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | FluidStack | PowerGPU |
| --- | --- | --- |
| Billing | Not published — no live self-serve rate card as of the date checked | Per second, no minimum, price locked at deploy |
| Payment | Not published for self-serve; large deals are contracted and invoiced | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Enterprise sales process for capacity deals; no public self-serve sign-up found | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None published | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Not applicable — there is no published on-demand tier to discount | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Sites reported in the UK, France, Iceland, Norway, Argentina and the US, built for named-customer contracts rather than self-serve zones | 32 regions on 5 continents, verified datacenters only |
| Minimums | Multi-year dedicated-cluster contracts, by the company's own public description of its business | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not published | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not published | $0.01/GB in and out, every region |
| Access | Sales-led onboarding; no public self-serve console or API documentation found at the date checked | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

FluidStack column: as published on fluidstack.io on 2026-09-04. Details change — verify before you decide.

## FluidStack vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on FluidStack's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

Configurations differ (node sizes, tiers, regions); the note beside each card says which FluidStack price is quoted. Negative differences mean FluidStack is cheaper on that card.

## When to stay with FluidStack

- You are, or are becoming, one of the small number of named customers FluidStack builds gigawatt-scale dedicated capacity for — that relationship is not something a per-GPU-hour public cloud replaces.
- An invoiced, multi-year, capex-style infrastructure contract is what your finance and procurement teams actually want.

## Why teams switch to PowerGPU

- You need a GPU today, not a sales process: PowerGPU is a self-serve console with instances live in about 30 seconds, no contract and no minimum term.
- A published rate card anyone can read before signing up, fixed at the public market median × 0.70, instead of no public pricing at all.
- 32 self-serve regions on five continents versus a handful of dedicated sites built for named contracts.
- Crypto payment with no KYC and no procurement cycle — the price sheet is the quote.

## Switching from FluidStack: what maps to what

| On FluidStack | On PowerGPU |
| --- | --- |
| Private Cloud / dedicated cluster contract (sales-led) | Reserved instance, flat −35% from 3 months, or a multi-node InfiniBand cluster quoted from the public sheet — no sales call for a handful of GPUs |
| (no published self-serve tier) | On-demand instance, per second, fixed at market median × 0.70, live in about 30 seconds |
| Enterprise / KYB onboarding | Email sign-up, crypto top-up, deploy |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your FluidStack machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

## FluidStack alternative: FAQ

**Is PowerGPU cheaper than FluidStack?**

There is nothing public to compare against: fluidstack.io's pricing page returned HTTP 404 and its homepage carried no rate card when checked 2026-09-04 — the company has pivoted to invoiced, multi-year Private Cloud contracts, reported at about 62% of revenue by the end of 2024 versus 38% for the original on-demand marketplace. PowerGPU publishes on-demand prices fixed at the public market median × 0.70 before you sign up for anything.

**Can I still rent a single FluidStack GPU by the hour?**

No self-serve page, console, sign-up flow or API documentation was reachable from fluidstack.io as of 2026-09-04. Treat any hourly figure quoted for FluidStack elsewhere as unconfirmed against the company's current site.

**What does FluidStack sell now?**

Dedicated, contracted AI-factory capacity for a small number of large customers, most visibly reported alongside Anthropic's compute build-out, across sites in the US, UK, France, Iceland, Norway and Argentina, funded by a $1.5B round that valued the company at about $18B (2026).

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on FluidStack, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

FluidStack and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with FluidStack. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/fluidstack · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/genesis-cloud -->

---
title: "Genesis Cloud Alternative (2026): GPU Prices vs PowerGPU"
description: "Genesis Cloud vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/genesis-cloud
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Genesis Cloud alternative: same NVIDIA GPUs, fixed prices 30% under the market

Genesis Cloud is a Munich-based GPU cloud provider that built its capacity on Nordic sites, first in Iceland, then Kristiansand, Norway with Bulk Infrastructure Group. Genesis Cloud GmbH (Handelsregister Munich, HRB 250051) entered formal liquidation under German law on 7 August 2025 and remains registered "GmbH i.L." (in liquidation), with a continuation filing as recently as 25 June 2026; third-party pricing trackers still listed its capacity for sale through 2026. genesiscloud.com could not be reached directly for this fiche (persistent TLS error, 3 attempts, checked 2026-09-04), so its current operating status beyond the registry filing is not independently confirmed here.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Genesis Cloud vs PowerGPU at a glance

Genesis Cloud

Munich GPU cloud on Nordic capacity; GmbH in formal liquidation since August 2025

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Genesis Cloud | PowerGPU |
| --- | --- | --- |
| Billing | Reported as per second or per hour depending on instance type (secondhand, not confirmed on genesiscloud.com directly) | Per second, no minimum, price locked at deploy |
| Payment | Not confirmed: genesiscloud.com could not be reached directly and no secondary source states accepted payment methods | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Not confirmed, for the same reason | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Not confirmed: none found in the sources checked | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Not confirmed: no spot or preemptible tier found in the sources checked | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Company blog posts describe Kristiansand, Norway and an earlier Iceland site; not independently confirmed as currently live | 32 regions on 5 continents, verified datacenters only |
| Minimums | At least the H100 SXM is sold by the 8-GPU node rather than per card, per third-party pricing trackers | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not confirmed: none found in the sources checked | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not confirmed: none found in the sources checked | $0.01/GB in and out, every region |
| Access | A self-serve console is implied by Genesis Cloud's own per-GPU product pages (URLs found by search), not independently confirmed | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Genesis Cloud column: as published on genesiscloud.com on 2026-09-04. Details change — verify before you decide.

## Genesis Cloud vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Genesis Cloud's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Genesis Cloud list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (8× node $17.52/hr ÷ 8; secondhand, not confirmed on genesiscloud.com directly) | $2.19 | **$1.428** | $0.714 | −35% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (from; secondhand, not confirmed on genesiscloud.com directly) | $2.80 | **$2.791** | $1.395 | +0% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (from; secondhand, not confirmed on genesiscloud.com directly) | $3.75 | **$5.425** | $2.712 | +45% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Genesis Cloud price is quoted. Negative differences mean Genesis Cloud is cheaper on that card.

## When to stay with Genesis Cloud

- Data, images or a running job are already on Genesis Cloud, and the near-term switching cost is judged lower than the continuity risk described below.
- The reported 8× H100 node at $17.52/hr, about $2.19/GPU-hr, is a genuinely low headline rate if a workload can use all 8 GPUs at once.

## Why teams switch to PowerGPU

- Continuity: Genesis Cloud GmbH (Munich, HRB 250051) has been in formal liquidation since 7 August 2025, still active as of a 25 June 2026 filing, worth weighing before committing new workloads there.
- No node minimum: PowerGPU rents 1× to 8× GPUs at the same per-GPU price; Genesis Cloud's H100 is reported sold by the 8-GPU node even for single-GPU jobs.
- Published, re-checked pricing: every PowerGPU rate sits on the public sheet at market median × 0.70; Genesis Cloud's own pricing pages could not be reached directly during this review.
- Crypto settlement with no KYC and 32 regions on five continents, against two reported Nordic sites.

## Switching from Genesis Cloud: what maps to what

| On Genesis Cloud | On PowerGPU |
| --- | --- |
| 8× H100 node, reported ~$2.19/GPU-hr | 1× to 8× H100 SXM, same per-GPU price, fixed at market median × 0.70 |
| Card or other payment (unconfirmed, site unreachable) | Crypto top-up: USDT, BTC, XMR, LTC, ETH, TRX, SOL; no card, no KYC |
| 2 reported Nordic sites (Norway, Iceland) | 32 regions on five continents |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Genesis Cloud machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Genesis Cloud | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $1,599 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $350 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Not confirmed: none found in the sources checked | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Genesis Cloud alternative: FAQ

**Is Genesis Cloud still operating in 2026?**

Genesis Cloud GmbH (Handelsregister Munich, HRB 250051) entered formal liquidation on 7 August 2025 and is still registered "i.L." as of a continuation filing on 25 June 2026; third-party pricing trackers still listed its capacity in 2026. Its own site could not be reached directly for this fiche, TLS error, checked 2026-09-04, so operating status beyond the registry filing is not independently confirmed.

**Is PowerGPU cheaper than Genesis Cloud?**

On the one price found with reasonable confidence, an 8× H100 SXM node at $17.52/hr, about $2.19/GPU-hr, via third-party pricing trackers rather than genesiscloud.com directly, PowerGPU's H100 SXM at the public market median × 0.70 is in the same range or lower, with no 8-GPU minimum (checked 2026-09-04).

**Where are Genesis Cloud's datacenters?**

Public reporting points to Kristiansand, Norway, built with Bulk Infrastructure Group, and an earlier Iceland site; current status could not be confirmed directly from genesiscloud.com, checked 2026-09-04.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Genesis Cloud, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Genesis Cloud and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Genesis Cloud. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/genesis-cloud · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/hetzner -->

---
title: "Hetzner GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "Hetzner vs PowerGPU: RTX PRO 4000 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/hetzner
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Hetzner GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

Hetzner sells single-tenant dedicated GPU servers rather than a multi-tenant cloud: GEX45 (RTX PRO 4000 Blackwell SFF, Helsinki) and GEX131 (RTX PRO 6000 Blackwell Max-Q, Nuremberg/Falkenstein), each a full physical machine with root access and unlimited traffic on a 1 Gbit/s port. Billing is monthly by default, with GEX131 also offered on direct hourly billing at a premium over the monthly-equivalent rate. The Ada-generation GEX44 (RTX 4000 SFF Ada) and GEX130 (RTX 6000 Ada) this comparison used to target have been discontinued and replaced by these Blackwell models — the GEX44 product page returns a 404 as of the date checked.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Hetzner vs PowerGPU at a glance

Hetzner

Dedicated GPU servers by the month (or hour): RTX PRO 4000/6000 Blackwell, Germany/Finland

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Hetzner | PowerGPU |
| --- | --- | --- |
| Billing | Monthly by default; GEX131 also offered as direct hourly billing at a premium rate | Per second, no minimum, price locked at deploy |
| Payment | Card, PayPal, SEPA direct debit | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account and payment verification; ID checks common for new dedicated-server accounts | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None | None — every hour is ≥30% under the market median instead |
| Cheaper tier | None — flat rate; hourly billing (GEX131) costs more per hour than the monthly-equivalent rate | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 2 — Helsinki (GEX45), Nuremberg and Falkenstein (GEX131) | 32 regions on 5 continents, verified datacenters only |
| Minimums | Whole server, 1 GPU each; monthly billing on GEX131 carries a one-time setup fee, waived on hourly billing | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Fixed NVMe included in the server; no separate GPU-attached volume product | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Unlimited traffic included on the 1 Gbit/s port | $0.01/GB in and out, every region |
| Access | Root access (bare metal), Hetzner Robot panel, rescue system, API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Hetzner column: as published on hetzner.com on 2026-09-04. Details change — verify before you decide.

## Hetzner vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Hetzner's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Hetzner list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [RTX PRO 4000](https://powergpu.ai/gpu/rtx-pro-4000) (GEX45 dedicated, €214/mo ÷ 730h, EUR→USD 1.1627 (2026-09-04); one-time setup fee on monthly billing) | $0.34 | **$0.183** | $0.091 | −46% on PowerGPU |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) (GEX131 dedicated (Max-Q), €889/mo ÷ 730h, EUR→USD 1.1627 (2026-09-04); direct hourly billing listed at €1.4247/hr (≈$1.66 today), no setup fee) | $1.42 | **$1.040** | $0.520 | −27% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Hetzner price is quoted. Negative differences mean Hetzner is cheaper on that card.

## When to stay with Hetzner

- A dedicated single-tenant machine with no hypervisor and full root access is the point, not a side effect.
- Unlimited traffic on a real physical NIC beats a metered per-GB rate for constant, heavy egress.
- A predictable flat monthly bill fits your budgeting better than a per-second meter.

## Why teams switch to PowerGPU

- One GPU model per region: GEX45 is Helsinki-only and GEX131 is Germany-only. PowerGPU runs the RTX PRO 4000 and RTX PRO 6000 across 32 regions on five continents.
- Per-second billing from zero commitment instead of a monthly server, or a premium hourly rate on GEX131 alone — deploy for ten minutes and pay for ten minutes.
- No account/ID verification, no card — top up in crypto and deploy.
- Flagship datacenter GPUs (H100, H200, B200) and consumer cards (RTX 5090, 4090) that Hetzner's GPU line does not carry at all.

## Switching from Hetzner: what maps to what

| On Hetzner | On PowerGPU |
| --- | --- |
| GEX45 / GEX131 dedicated server | Instance on the equivalent card, or a full KVM VM |
| Fixed NVMe (included) | Instance disk plus volumes at $0.08/GB/month if more is needed |
| Hetzner Robot panel | powergpu dashboard / REST API / CLI |
| Monthly commitment | Reserved instance, −35% from 3 months, no server swap needed |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu rtx-pro-4000 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $0.183/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Hetzner machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Hetzner | PowerGPU |
| --- | --- | --- |
| 1× RTX PRO 4000, 730 hours on-demand | $248 | **$134** |
| 1× RTX PRO 4000, 8 h/day × 20 days | $54 | **$29** |
| Same 160 hours, PowerGPU interruptible | — | **$15** |
| 500 GB of storage, one month | Fixed NVMe included in the server; no separate GPU-attached volume product | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Hetzner GPU alternative: FAQ

**Is PowerGPU cheaper than Hetzner's GPU servers?**

On an hourly-equivalent basis, yes: GEX131 (RTX PRO 6000 Blackwell Max-Q) works out to about $1.42/hr from its €889/month price (÷730h, checked 2026-09-04), or about $1.66/hr on Hetzner's own direct hourly billing; GEX45 (RTX PRO 4000 Blackwell) is about $0.34/hr from €214/month. PowerGPU fixes the equivalent cards at the public market median × 0.70 and bills per second from zero commitment.

**What happened to the GEX44 and GEX130 servers?**

Hetzner replaced them with Blackwell-generation models: GEX45 (RTX PRO 4000 Blackwell SFF) and GEX131 (RTX PRO 6000 Blackwell Max-Q). The GEX44 product page returned a 404 when checked 2026-09-04.

**Can I rent a Hetzner-class GPU server by the second on PowerGPU?**

Yes — the same RTX PRO 4000 and RTX PRO 6000 tiers are billed per second with no monthly minimum and no setup fee, across 32 regions rather than one server in one city.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Hetzner, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Hetzner and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Hetzner. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/hetzner · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/hyperbolic -->

---
title: "Hyperbolic Alternative (2026): GPU Prices vs PowerGPU"
description: "Hyperbolic vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/hyperbolic
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Hyperbolic alternative: same NVIDIA GPUs, fixed prices 30% under the market

Hyperbolic aggregates GPU capacity from a network of datacenter partners and resells it on-demand, alongside an OpenAI-compatible inference API for open-weight models. On-demand rates are refreshed weekly from supplier pricing but locked for the lifetime of an instance once deployed; reserved capacity is prepaid for a fixed term (self-serve from one week, longer terms via sales) at a discount off on-demand. Sign-up takes a single form with no sales call, and a cluster reports ready in under a minute.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Hyperbolic vs PowerGPU at a glance

Hyperbolic

GPU marketplace and inference API: weekly-refreshed on-demand rates, reserved and private-cloud tiers

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Hyperbolic | PowerGPU |
| --- | --- | --- |
| Billing | Per GPU-hour, rate locked at deploy time, refreshed weekly for new instances | Per second, no minimum, price locked at deploy |
| Payment | Card, wire/ACH upfront, or crypto — USDC, USDT or DAI on Base | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Email sign-up, no forms or sales call for on-demand; KYC policy not published | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None published | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Prepaid reserved rate, self-serve from 1 week, longer terms via sales | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Datacenter-partner network, no fixed list; availability shown per configuration in the console | 32 regions on 5 continents, verified datacenters only |
| Minimums | None on-demand; reserved has a 1-week self-serve minimum | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not published as a flat per-GB rate | $0.08/GB/month NVMe, volumes survive instances |
| Egress | No egress fees | $0.01/GB in and out, every region |
| Access | SSH, OpenAI-compatible inference API, console, REST API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Hyperbolic column: as published on hyperbolic.xyz on 2026-09-04. Details change — verify before you decide.

## Hyperbolic vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Hyperbolic's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Hyperbolic list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (80 GB, on-demand, weekly-refreshed rate) | $3.19 | **$1.428** | $0.714 | −55% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (141 GB, on-demand, weekly-refreshed rate) | $3.99 | **$2.791** | $1.395 | −30% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (180 GB, on-demand, weekly-refreshed rate) | $5.99 | **$5.425** | $2.712 | −9% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Hyperbolic price is quoted. Negative differences mean Hyperbolic is cheaper on that card.

## When to stay with Hyperbolic

- The OpenAI-compatible inference API for open-weight models is your product, not raw GPU rental.
- A week-long prepaid reserved block at a discount off on-demand fits a short, fixed-length job better than a monthly commitment.
- You need private-cloud capacity carved out of a specific datacenter partner.

## Why teams switch to PowerGPU

- Price does not move week to week: PowerGPU fixes every card at the public market median × 0.70, instead of a weekly refresh that has gone from $1.50 to $3.19 on the H100 SXM since June 2026.
- No egress fee is matched here, and storage is a flat $0.08/GB/month on top of it, published rather than left off the pricing page.
- A flat −50% interruptible tier and −35% reserved from 3 months, both published, against Hyperbolic's sales-quoted longer reservations.
- Consumer cards (RTX 5090, 4090, 3090) and 32 fixed regions, instead of a datacenter-partner network with no published location list.

## Switching from Hyperbolic: what maps to what

| On Hyperbolic | On PowerGPU |
| --- | --- |
| On-demand cluster | Instance — container in 30 s or full KVM VM |
| Reserved (prepaid, 1 week+) | Reserved instance, −35% from 3 months |
| Private cloud | Dedicated capacity, quoted from the public sheet |
| OpenAI-compatible inference API | Serverless endpoint (vLLM / any container, scale to zero) |
| Hyperbolic console / API | powergpu CLI / REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Hyperbolic machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Hyperbolic | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $2,329 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $510 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Not published as a flat per-GB rate | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Hyperbolic alternative: FAQ

**Is PowerGPU cheaper than Hyperbolic?**

On the cards Hyperbolic lists, generally yes: its on-demand H100 SXM is $3.19/hr, H200 is $3.99/hr and B200 is $5.99/hr (checked 2026-09-04), all refreshed weekly and only known at deploy time; PowerGPU fixes the same cards at the public market median × 0.70 and does not move the price on an existing instance, or on a new one week to week.

**Does PowerGPU accept the same crypto as Hyperbolic?**

Hyperbolic takes USDC, USDT or DAI on Base (checked 2026-09-04). PowerGPU is crypto-only and takes more directly — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX and SOL — with no card option and no KYC step.

**Can I still get an OpenAI-compatible inference endpoint like Hyperbolic's?**

Yes — PowerGPU serverless endpoints run vLLM or any container behind an OpenAI-compatible route, scale to zero, and bill per second at the same fixed rates as an instance.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Hyperbolic, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Hyperbolic and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Hyperbolic. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/hyperbolic · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/latitude-sh -->

---
title: "Latitude.sh Alternative (2026): GPU Prices vs PowerGPU"
description: "Latitude.sh vs PowerGPU: H100 PCIE price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/latitude-sh
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Latitude.sh alternative: same NVIDIA GPUs, fixed prices 30% under the market

Latitude.sh (formerly Maxihost, a São Paulo hosting company founded in 2001 that moved to an all-bare-metal model in 2018 and rebranded in August 2022) sells single-tenant bare metal GPU servers with both hourly and monthly billing available on the same plan, across roughly 25 locations spanning North and South America, Europe and Asia-Pacific. The current GPU line-up on the public pricing page is three plans: a single H100 80 GB, an 8-way RTX PRO 6000 (Server Edition) node, and an 8-way HGX B300 node. An L40S plan was announced on Latitude's own changelog in the past but is not listed on the pricing page as of the date checked.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Latitude.sh vs PowerGPU at a glance

Latitude.sh

Bare metal GPU by the hour or the month: H100, RTX PRO 6000, HGX B300 across the Americas, Europe and Asia

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Latitude.sh | PowerGPU |
| --- | --- | --- |
| Billing | Hourly or monthly, chosen per server; monthly is 30% off the hourly-equivalent rate, yearly is 50% off | Per second, no minimum, price locked at deploy |
| Payment | Card (automatic $5 verification hold, refunded in 7 days) or crypto — USDC on Solana or Ethereum | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Not published beyond payment-method verification | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None published | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Reserved billing only — 30% off monthly, 50% off yearly versus the hourly rate; no interruptible/spot tier | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Roughly 25 locations across the Americas, Europe and Asia-Pacific; GPU plan availability by location is not itemised on the pricing page | 32 regions on 5 continents, verified datacenters only |
| Minimums | Whole server per plan — 1× H100, or 8× on the RTX PRO 6000 and B300 nodes | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Fixed NVMe included per plan (2 to 4 drives); no separate GPU-attached volume product listed | $0.08/GB/month NVMe, volumes survive instances |
| Egress | 20 TB outbound free per server per month, then $0.01/GB; inbound always free | $0.01/GB in and out, every region |
| Access | Account dashboard and REST API; bare metal implies root access, not itemised on the pricing page | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Latitude.sh column: as published on latitude.sh on 2026-09-04. Details change — verify before you decide.

## Latitude.sh vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Latitude.sh's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Latitude.sh list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (g3.h100.small, 1x H100 80 GB · $1,230/mo; PCIe inferred from the single-GPU packaging, not labelled by Latitude.sh) | $1.68 | **$1.867** | $0.933 | +11% on PowerGPU |
| [RTX PRO 6000 S](https://powergpu.ai/gpu/rtx-pro-6000-s) (g4.rtx6kpro.large, 8x RTX PRO 6000 Server Edition, $24/hr node ÷ 8 · $17,520/mo node) | $3.00 | **$1.073** | $0.536 | −64% on PowerGPU |
| [B300](https://powergpu.ai/gpu/b300) (g4.b300.large, 8x HGX B300, $64/hr node ÷ 8 · $46,720/mo node) | $8.00 | **$6.737** | $3.368 | −16% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Latitude.sh price is quoted. Negative differences mean Latitude.sh is cheaper on that card.

## When to stay with Latitude.sh

- You want hourly and monthly servers side by side on one account and the option to flip a given server between the two.
- A regional footprint this dense in Latin America specifically (Buenos Aires, Bogotá, Santiago, São Paulo ×2) matters more than raw region count elsewhere.
- A 50%-off yearly commitment on a bare metal node fits your budgeting better than a rolling reserved discount.

## Why teams switch to PowerGPU

- Price: the H100 is fixed at market median × 0.70 here versus $1.68/hr at Latitude.sh's g3.h100.small, and the RTX PRO 6000 gap is wider still ($3.00/GPU-hour on the 8-way node).
- Single GPUs on every model, not just the H100: Latitude.sh's RTX PRO 6000 and B300 are 8-GPU-only nodes with no smaller size published.
- Direct crypto settlement in USDT, BTC, XMR, LTC, ETH, TRX or SOL, instead of a card hold or a USDC-only crypto option.
- 32 regions on five continents, each confirmed to carry the full GPU line-up, versus roughly 25 locations where GPU availability is not itemised per site.

## Switching from Latitude.sh: what maps to what

| On Latitude.sh | On PowerGPU |
| --- | --- |
| g3.h100.small / g4.rtx6kpro.large / g4.b300.large | Instance, 1× to 8× GPUs at the same per-GPU price |
| Hourly billing | Instance, per second, no minimum |
| Monthly / yearly reserved billing (−30% / −50%) | Reserved instance, flat −35% from 3 months |
| Fixed NVMe per plan | Instance disk plus volumes at $0.08/GB/month if more is needed |
| Dashboard / REST API | powergpu dashboard / REST API / CLI |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-pcie --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.867/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Latitude.sh machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Latitude.sh | PowerGPU |
| --- | --- | --- |
| 1× H100 PCIE, 730 hours on-demand | $1,226 | **$1,363** |
| 1× H100 PCIE, 8 h/day × 20 days | $269 | **$299** |
| Same 160 hours, PowerGPU interruptible | — | **$149** |
| 500 GB of storage, one month | Fixed NVMe included per plan (2 to 4 drives); no separate GPU-attached volume product listed | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Latitude.sh alternative: FAQ

**Is PowerGPU cheaper than Latitude.sh?**

Yes on both published cards: Latitude.sh's g3.h100.small (1x H100 80 GB) is $1.68/hr and its g4.rtx6kpro.large (8x RTX PRO 6000) works out to $3.00/GPU-hour from its $24/hr node price (checked 2026-09-04); PowerGPU fixes both cards at the public market median × 0.70.

**Does Latitude.sh still offer the L40S?**

Not on its current pricing page: only the H100, RTX PRO 6000 and HGX B300 plans are listed as of 2026-09-04. Latitude's changelog announced L40S availability previously; treat any L40S price quoted for Latitude.sh elsewhere as unconfirmed.

**Can I rent a single GPU instead of a full node?**

For the H100, yes — g3.h100.small is already a 1-GPU server. For the RTX PRO 6000 and B300, Latitude.sh only lists 8-GPU nodes; PowerGPU rents both cards from 1× at the same per-GPU price, fixed at the market median × 0.70.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Latitude.sh, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Latitude.sh and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Latitude.sh. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/latitude-sh · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/lightning-ai -->

---
title: "Lightning AI Alternative (2026): GPU Prices vs PowerGPU"
description: "Lightning AI vs PowerGPU: Tesla T4 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/lightning-ai
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Lightning AI alternative: same NVIDIA GPUs, fixed prices 30% under the market

Lightning AI Studios is a browser-based ML IDE — JupyterLab/VS Code plus a persistent Linux machine — that attaches a GPU on demand and bills per hour while it runs. Stopping a Studio pauses GPU billing immediately while the disk persists. The Free plan needs no card and includes a monthly credit allowance; concurrent-GPU limits and storage caps then scale with the paid plans.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Lightning AI vs PowerGPU at a glance

Lightning AI

Cloud IDE with built-in GPUs: Studios, per-hour billing, monthly free credits

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Lightning AI | PowerGPU |
| --- | --- | --- |
| Billing | Per hour, credit-based; stopping a Studio pauses GPU billing immediately | Per second, no minimum, price locked at deploy |
| Payment | Not published at time of check; the Free plan needs no card, so a payment method is presumably required only for paid plans | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Email, Google or GitHub signup | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | 15 credits/month on the Free plan (about 80 GPU hours on interruptible machines); phone verification reported to add 7 more | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Interruptible machines discounted up to 80%; multi-GPU Studios (8×) price per GPU lower than a single GPU | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Not published; one tracker reports a single US-East region | 32 regions on 5 continents, verified datacenters only |
| Minimums | None on single GPUs; concurrent-GPU count is capped by plan (2 Free, 6 Pro, 12 Teams) | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Persistent SSD bundled with a Studio, capacity by plan (up to 200 GB Pro, 2 TB Teams); not itemised per GB | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Reported as free by one tracker; not confirmed on lightning.ai itself | $0.01/GB in and out, every region |
| Access | Browser IDE (Studios) with SSH into the same machine, Python SDK, CLI | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Lightning AI column: as published on lightning.ai on 2026-09-04. Details change — verify before you decide.

## Lightning AI vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Lightning AI's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Lightning AI list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) (Studios single-GPU on-demand) | $0.41 | **$0.103** | $0.051 | −75% on PowerGPU |
| [L4](https://powergpu.ai/gpu/l4) (Studios single-GPU on-demand; another tracker cites $0.48 (checked 2026-09-04)) | $0.60 | **$0.225** | $0.112 | −63% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (80GB, form factor unspecified; single-GPU on-demand — 8-GPU Studios reported as low as $1.79/GPU; another tracker cites $2.71 (checked 2026-09-04)) | $2.99 | **$0.560** | $0.280 | −81% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (single-GPU on-demand — 8-GPU Studios reported as low as $1.99/GPU; another tracker cites $3.29 (checked 2026-09-04)) | $3.50 | **$1.428** | $0.714 | −59% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Lightning AI price is quoted. Negative differences mean Lightning AI is cheaper on that card.

## When to stay with Lightning AI

- You want a hosted, opinionated ML IDE — JupyterLab/VS Code, environment snapshots, team sharing — not a bare machine you configure yourself.
- The monthly free credits cover your workload on interruptible GPUs and you never need to pay.
- Your stack is already built on PyTorch Lightning / Fabric and the tight integration is worth the premium.

## Why teams switch to PowerGPU

- Price on the cards we could confirm: PowerGPU fixes the H100 at the public market median × 0.70 and the A100 80GB the same way, against Lightning's single-GPU Studio rates above $3 and $2.70 respectively.
- No plan-tied concurrency cap: Lightning limits concurrent GPUs by subscription tier (2 on Free, up to 12 on Teams); PowerGPU has none.
- Crypto payment with no KYC and no card, instead of a credit-and-plan system.
- Consumer flagships (RTX 5090, 4090, 3090) across 32 regions on five continents — Lightning's GPU and region selection is narrower.

## Switching from Lightning AI: what maps to what

| On Lightning AI | On PowerGPU |
| --- | --- |
| Studio (GPU machine) | Instance |
| Stop a Studio | Stop — billing ends immediately, disk kept at $0.08/GB/month |
| Interruptible machine | Interruptible instance, flat −50% |
| Multi-GPU Studio (up to 8×) | 1× to 8× instance, same per-GPU price at any count |
| Lightning SDK / CLI | powergpu CLI / REST API / Python SDK |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu tesla-t4 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $0.103/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Lightning AI machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Lightning AI | PowerGPU |
| --- | --- | --- |
| 1× Tesla T4, 730 hours on-demand | $299 | **$75** |
| 1× Tesla T4, 8 h/day × 20 days | $66 | **$16** |
| Same 160 hours, PowerGPU interruptible | — | **$8** |
| 500 GB of storage, one month | Persistent SSD bundled with a Studio, capacity by plan (up to 200 GB Pro, 2 TB Teams); not itemised per GB | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Lightning AI alternative: FAQ

**Is PowerGPU cheaper than Lightning AI Studios?**

On the rates we could confirm, yes: Lightning's single-GPU Studio pricing runs from about $0.41/hr (T4) to $3.29–3.50/hr (H100) depending on which tracker you check (checked 2026-09-04) — Lightning's own pricing page did not render a rate table for us. PowerGPU fixes the same cards at the public market median × 0.70, the same price whether you rent one GPU or eight.

**Does PowerGPU offer a hosted IDE like Studios?**

Not a browser IDE specifically, but every instance ships with an optional one-click Jupyter template and SSH, so the same JupyterLab workflow runs on a machine you fully control.

**What happens to my data if I stop a machine?**

On both platforms the disk survives a stop; PowerGPU charges $0.08/GB/month for it whether the instance is running or stopped.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Lightning AI, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Lightning AI and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Lightning AI. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/lightning-ai · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/massed-compute -->

---
title: "Massed Compute Alternative (2026): GPU Prices vs PowerGPU"
description: "Massed Compute vs PowerGPU: B300 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/massed-compute
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Massed Compute alternative: same NVIDIA GPUs, fixed prices 30% under the market

Massed Compute is an NVIDIA Preferred Partner selling on-demand GPU virtual machines and bare metal from its own Tier III datacenters in the United States (US-East, US-West), aimed at AI training and inference without contracts. Its catalogue runs from single RTX A5000s to 8-GPU B200 and B300 nodes, billed per minute with no setup fees and no bandwidth charges. It also publishes SOC 2 Type II, HIPAA and GDPR compliance, a trust signal most GPU-marketplace competitors do not carry.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Massed Compute vs PowerGPU at a glance

Massed Compute

NVIDIA Preferred Partner: on-demand GPU VMs, bare metal, US-only, per-minute billing

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Massed Compute | PowerGPU |
| --- | --- | --- |
| Billing | Per minute, no termination fees | Per second, no minimum, price locked at deploy |
| Payment | Card, prepaid account balance | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account and payment method; SOC 2 Type II / HIPAA / GDPR-compliant infrastructure | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None published | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot instances, roughly 40-70% below on-demand (GPU-dependent) | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | United States only — US-East and US-West, Tier III facilities | 32 regions on 5 continents, verified datacenters only |
| Minimums | None — no setup fees, no minimum spend, no contracts | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not itemised on the pricing page | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free — no bandwidth charges | $0.01/GB in and out, every region |
| Access | Virtual machines (SSH, Jupyter), bare metal and clusters on request, API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Massed Compute column: as published on massedcompute.com on 2026-09-04. Details change — verify before you decide.

## Massed Compute vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Massed Compute's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Massed Compute list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [B300](https://powergpu.ai/gpu/b300) (8× SXM6 node only, $52.80/hr ÷ 8) | $6.60 | **$6.737** | $3.368 | +2% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (8× SXM6 node only, $43.46/hr ÷ 8) | $5.43 | **$5.425** | $2.712 | +0% on PowerGPU |
| [H200 NVL](https://powergpu.ai/gpu/h200-nvl) (141 GB NVL, 1×) | $3.62 | **$2.650** | $1.325 | −27% on PowerGPU |
| [H100 NVL](https://powergpu.ai/gpu/h100-nvl) (1×) | $3.11 | **$1.811** | $0.905 | −42% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (SXM5 80 GB, 1×) | $2.89 | **$1.428** | $0.714 | −51% on PowerGPU |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (80 GB, 1×) | $2.73 | **$1.867** | $0.933 | −32% on PowerGPU |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) (Blackwell 96 GB workstation, 1×) | $2.19 | **$1.040** | $0.520 | −53% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (SXM4 80 GB, 1×) | $1.38 | **$0.560** | $0.280 | −59% on PowerGPU |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) (80 GB, 1×) | $1.35 | **$0.374** | $0.187 | −72% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (1×) | $0.88 | **$0.514** | $0.257 | −42% on PowerGPU |
| [RTX 6000Ada](https://powergpu.ai/gpu/rtx-6000ada) (1×) | $0.79 | **$0.467** | $0.233 | −41% on PowerGPU |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) (1×) | $0.57 | **$0.281** | $0.140 | −51% on PowerGPU |
| [RTX A5000](https://powergpu.ai/gpu/rtx-a5000) (1×) | $0.44 | **$0.161** | $0.080 | −63% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Massed Compute price is quoted. Negative differences mean Massed Compute is cheaper on that card.

## When to stay with Massed Compute

- SOC 2 Type II, HIPAA and GDPR compliance out of the box matters if a US-regulated workload is the whole point.
- Zero bandwidth charges and genuinely low raw on-demand rates on H100 and A100 — no separate vCPU/RAM/storage lines on top of the GPU price.
- You specifically need US-East or US-West placement and nothing further afield.

## Why teams switch to PowerGPU

- 32 regions on five continents versus two US regions — useful the moment a workload, a team or a compliance rule sits outside the United States.
- Crypto, no KYC: top up in USDT, BTC, XMR, LTC, ETH, TRX or SOL instead of adding a card to an account balance.
- Consumer flagships (RTX 5090, 4090, 3090) that Massed Compute does not sell alongside its datacenter and workstation line-up.
- A published flat −50% interruptible tier and −35% reserved tier, where Massed Compute's own spot rates are not listed on its public pricing page.

## Switching from Massed Compute: what maps to what

| On Massed Compute | On PowerGPU |
| --- | --- |
| On-demand VM | Instance — container in 30 s or full KVM VM |
| 8-GPU-only B200 / B300 node | 8× machine, or 1×-4× at the same per-GPU price |
| Spot instance | Interruptible instance, flat −50% |
| Manage Billing account balance | Crypto top-up, no card, no KYC |
| VM disk | Volume, $0.08/GB/month |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu b300 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $6.737/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Massed Compute machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Massed Compute | PowerGPU |
| --- | --- | --- |
| 1× B300, 730 hours on-demand | $4,818 | **$4,918** |
| 1× B300, 8 h/day × 20 days | $1,056 | **$1,078** |
| Same 160 hours, PowerGPU interruptible | — | **$539** |
| 500 GB of storage, one month | Not itemised on the pricing page | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Massed Compute alternative: FAQ

**Is PowerGPU cheaper than Massed Compute?**

On most overlapping cards, yes: Massed Compute lists the H100 SXM5 at $2.89/hr and the A100 80 GB at $1.35-1.38/hr on its public pricing page (checked 2026-09-04); PowerGPU fixes both at the public market median × 0.70. Massed Compute's no-bandwidth-charges policy is matched here by a flat, low per-GB egress rate rather than a bundled allowance.

**Does Massed Compute accept crypto?**

Not as far as its public billing documentation shows — payment runs through a card-funded account balance. PowerGPU is crypto-only, with no card option at all.

**Can I deploy outside the United States on Massed Compute?**

No — its published footprint is US-East and US-West only. PowerGPU runs 32 regions on five continents.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Massed Compute, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Massed Compute and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Massed Compute. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/massed-compute · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/nebius -->

---
title: "Nebius AI Cloud Alternative (2026): GPU Prices vs PowerGPU"
description: "Nebius vs PowerGPU: B300 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/nebius
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Nebius AI Cloud alternative: same NVIDIA GPUs, fixed prices 30% under the market

Nebius is the AI-infrastructure business that separated from Yandex and now operates and is funded independently. It owns and builds its own datacenters rather than reselling capacity, running its original site in Maentsaelae, Finland since 2014 (now tripling capacity) alongside a newer deployment in Saint-Denis, France (Equinix) and announced sites in Lappeenranta, Finland, the UK, Iceland and Israel. Its catalogue spans H100 through H200, B200 and B300, with GB200 and GB300 NVL72 rack-scale systems available by sales contact only.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Nebius vs PowerGPU at a glance

Nebius

Ex-Yandex hyperscaler-scale cloud: owned Finland/France datacenters, H100 through B300

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Nebius | PowerGPU |
| --- | --- | --- |
| Billing | Hourly rate shown per GPU on the pricing page; sub-hour increment not stated | Per second, no minimum, price locked at deploy |
| Payment | Card via Stripe, or bank transfer; minimum first payment $25 | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Company signup required to unlock commitment discounts; no KYC process documented on the pricing page | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Not mentioned on the pricing page | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Preemptible at roughly 44–47% below on-demand on every listed GPU; separate multi-month commitments advertised up to −35% | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Owned or leased sites in Finland (Maentsaelae, plus a new 310MW site announced for Lappeenranta) and Paris (Saint-Denis); expansion announced for the UK, Iceland and Israel, no single published region count | 32 regions on 5 continents, verified datacenters only |
| Minimums | Not stated for on-demand; commitment tiers start at one month for the discount | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | $0.08/GiB/month (Shared Filesystem); $0.0147/GiB/month (Object Standard) | $0.08/GB/month NVMe, volumes survive instances |
| Egress | $0.015/GiB (Object Standard); free on the Intelligent storage tier | $0.01/GB in and out, every region |
| Access | Self-serve console and API for the instances above; GB200 and GB300 NVL72 systems are contact-sales only | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Nebius column: as published on nebius.com on 2026-09-04. Details change — verify before you decide.

## Nebius vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Nebius's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Nebius list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [B300](https://powergpu.ai/gpu/b300) (HGX B300, per-GPU slice: 24 vCPU / 346GB RAM) | $7.85 | **$6.737** | $3.368 | −14% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (HGX B200, per-GPU slice: 20 vCPU / 224GB RAM) | $7.15 | **$5.425** | $2.712 | −24% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (HGX H200, per-GPU slice: 16 vCPU / 200GB RAM) | $4.50 | **$2.791** | $1.395 | −38% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (HGX H100, per-GPU slice: 16 vCPU / 200GB RAM) | $3.85 | **$1.428** | $0.714 | −63% on PowerGPU |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) (edition not specified by Nebius; 24 vCPU / 218GB RAM) | $1.80 | **$1.040** | $0.520 | −42% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (from, AMD host; Intel host from $1.82) | $1.55 | **$0.514** | $0.257 | −67% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Nebius price is quoted. Negative differences mean Nebius is cheaper on that card.

## When to stay with Nebius

- The newest NVIDIA generations, B200, B300, and GB200/GB300 NVL72 by contract, from a single hyperscaler-scale European vendor matter more than price.
- Multi-month commitment discounts, up to −35%, plus a separate preemptible tier fit a mixed steady-state-plus-burst workload.
- EU-owned infrastructure at hyperscaler scale, rather than a marketplace of smaller regions, is a specific requirement.

## Why teams switch to PowerGPU

- Price: Nebius lists the H100 HGX slice at $3.85/hr and the H200 at $4.50/hr; PowerGPU fixes both at the public market median × 0.70.
- Access: GB200 and GB300 NVL72 and some large configurations are sales-gated; every PowerGPU card, including the newest Blackwell parts, is on the public sheet and self-serve.
- Geography: Nebius's owned sites cluster in Finland and France with expansion announced elsewhere; PowerGPU runs 32 regions on five continents today.
- Crypto settlement with no KYC, against Nebius's card or bank-transfer billing and company signup for its best discounts.

## Switching from Nebius: what maps to what

| On Nebius | On PowerGPU |
| --- | --- |
| HGX H100 / H200 / B200 / B300 slice | Instance, same per-GPU price from 1× to 8× |
| Preemptible, about 45% below on-demand | Interruptible flat −50% |
| Multi-month commitment, up to −35% | Reserved −35% (3+ months) |
| Shared Filesystem, $0.08/GiB/month | Volume, $0.08/GB/month |
| GB200 / GB300 NVL72, contact sales | Clusters, quoted per GPU from the public sheet |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu b300 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $6.737/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Nebius machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Nebius | PowerGPU |
| --- | --- | --- |
| 1× B300, 730 hours on-demand | $5,731 | **$4,918** |
| 1× B300, 8 h/day × 20 days | $1,256 | **$1,078** |
| Same 160 hours, PowerGPU interruptible | — | **$539** |
| 500 GB of storage, one month | $0.08/GiB/month (Shared Filesystem); $0.0147/GiB/month (Object Standard) | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Nebius AI Cloud alternative: FAQ

**Is PowerGPU cheaper than Nebius?**

On every overlapping card, yes: Nebius prices the H100 HGX slice at $3.85/hr and the B200 at $7.15/hr (checked 2026-09-04); PowerGPU fixes both at the public market median × 0.70. Nebius's preemptible tier, roughly 45% below its own on-demand rate, narrows the gap but still sits on a higher base price.

**Can I get GB200 NVL72 capacity on PowerGPU like Nebius offers?**

Rack-scale NVL72 systems are handled the way multi-node clusters are here: quoted per GPU from the public sheet rather than a private sales negotiation. Nebius lists GB200 and GB300 NVL72 as contact-sales only (checked 2026-09-04).

**Is Nebius the same company as Yandex?**

No longer. Nebius is the international AI-infrastructure business that separated from Yandex and now operates and is funded independently, building its own datacenters rather than reselling cloud capacity.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Nebius, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Nebius and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Nebius. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/nebius · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/novita -->

---
title: "Novita AI Alternative (2026): GPU Prices vs PowerGPU"
description: "Novita AI vs PowerGPU: RTX 4090 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/novita
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Novita AI alternative: same NVIDIA GPUs, fixed prices 30% under the market

Novita AI sells three things side by side: on-demand and spot GPU instances, serverless GPU endpoints, and a large catalog of hosted model APIs (LLM, image, video, audio) billed per token or per output. It markets itself on startup-friendly rates and a strong Asia-Pacific footprint alongside US and EU capacity. Its own pricing page renders its GPU-instance rate table client-side, so the figures below are triangulated from third-party trackers rather than read directly off novita.ai.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Novita AI vs PowerGPU at a glance

Novita AI

GPU instances, serverless endpoints and hosted model APIs, strong APAC reach

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Novita AI | PowerGPU |
| --- | --- | --- |
| Billing | Hourly with per-second granularity on GPU instances; model APIs billed per token or per output separately | Per second, no minimum, price locked at deploy |
| Payment | Not confirmed at time of check — no primary source found | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account signup; specifics not confirmed | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | A small one-time signup voucher (commonly reported around $0.50) for API testing is mentioned by third parties, not confirmed on Novita's own pricing page | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Spot/interruptible instances, roughly 40–70% below on-demand per third-party trackers | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | US, EU and Asia-Pacific per third-party trackers; not broken out on novita.ai/pricing itself at time of check | 32 regions on 5 continents, verified datacenters only |
| Minimums | None reported | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not published at time of check | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not published at time of check | $0.01/GB in and out, every region |
| Access | GPU instances (VMs/containers), serverless endpoints, hosted model APIs, console and API | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Novita AI column: as published on novita.ai on 2026-09-04. Details change — verify before you decide.

## Novita AI vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Novita AI's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Novita AI list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (on-demand, 24GB; one tracker cites "from $0.33/hr", spot reported near $0.25/hr (checked 2026-09-04)) | $0.61 | **$0.327** | $0.163 | −46% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (8-GPU bare-metal node, per GPU; a dedicated-endpoint listing cites $1.99/GPU-hr (checked 2026-09-04)) | $1.70 | **$1.428** | $0.714 | −16% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Novita AI price is quoted. Negative differences mean Novita AI is cheaper on that card.

## When to stay with Novita AI

- You want serverless GPU endpoints and a large catalog of hosted model APIs (LLM, image, video, audio) on one bill — PowerGPU does not host third-party models.
- Your users are concentrated in Asia-Pacific and Novita's regional presence there matters for latency.
- Sub-$0.35/hr RTX 4090 spot pricing, if the lower figure we found holds, suits large-scale fault-tolerant batch jobs.

## Why teams switch to PowerGPU

- One fixed, published price at market median × 0.70, rather than a page where "starting at" headline rates and third-party trackers diverge by 2–3× depending on the source.
- Verified datacenters only with a 99.9% SLA, instead of a mixed on-demand/spot pool where trackers mark some cards as only "Med" availability.
- Crypto payment with no KYC.
- Consumer and workstation flagships (RTX 5090, RTX 6000 Ada, RTX A6000) alongside H100/H200/B200 in one catalog, at the same per-GPU price from 1× to 8×.

## Switching from Novita AI: what maps to what

| On Novita AI | On PowerGPU |
| --- | --- |
| GPU instance (on-demand) | Instance |
| Spot GPU instance | Interruptible instance, flat −50% |
| Serverless GPU endpoint | Serverless endpoint |
| Hosted model API (per token/output) | No direct equivalent — bring the model as a container to a serverless endpoint instead |
| Novita console / API | powergpu CLI / REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu rtx-4090 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $0.327/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Novita AI machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Novita AI | PowerGPU |
| --- | --- | --- |
| 1× RTX 4090, 730 hours on-demand | $445 | **$239** |
| 1× RTX 4090, 8 h/day × 20 days | $98 | **$52** |
| Same 160 hours, PowerGPU interruptible | — | **$26** |
| 500 GB of storage, one month | Not published at time of check | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Novita AI alternative: FAQ

**Is PowerGPU cheaper than Novita for GPU instances?**

On the figures we could confirm, Novita's RTX 4090 on-demand instances are reported between $0.33 and $0.61/hr with spot near $0.25/hr, and H100 SXM bare-metal nodes around $1.70/GPU-hr (checked 2026-09-04, third-party trackers — Novita's own pricing page did not render a rate table for us). PowerGPU fixes both cards at the public market median × 0.70, the same price at 1× or 8×.

**Does PowerGPU host third-party models like Novita's API catalog?**

No — Novita's per-token model APIs are a separate hosted-model product. PowerGPU gives you the GPU, as an instance or a serverless endpoint, to run whatever model or container you choose, including one of 37 ready templates.

**Does PowerGPU serve Asia-Pacific like Novita?**

Yes, as part of 32 regions on five continents, including Asia-Pacific locations.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Novita AI, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Novita AI and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Novita AI. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/novita · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/oracle-cloud -->

---
title: "Oracle Cloud GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "Oracle Cloud vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/oracle-cloud
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Oracle Cloud GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

Oracle Cloud Infrastructure sells GPUs as named bare-metal and VM shapes — BM.GPU.H100.8, BM.GPU.H200.8, BM.GPU.A100.80, BM.GPU.L40S.4, VM.GPU.A10.1 — billed hourly per GPU, with a fixed 50% preemptible discount and standard OCI tooling (Console, CLI, Terraform, OKE). Pricing is published per GPU rather than negotiated, which is unusual among hyperscalers, but the flagship H100 and H200 shapes are 8-GPU-only and new tenancies start with GPU service limits that must be raised before anything deploys.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Oracle Cloud vs PowerGPU at a glance

Oracle Cloud

OCI bare metal and VMs: H100, H200, A100, L40S, A10, preemptible at 50% off

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Oracle Cloud | PowerGPU |
| --- | --- | --- |
| Billing | Per hour, per GPU | Per second, no minimum, price locked at deploy |
| Payment | Card, invoicing for Universal Credits / enterprise agreements | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account and card verification; GPU service-limit increase requests | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | $300 trial credit (GPU limits are typically 0 on trial tenancies) | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Preemptible instances at a fixed 50% off on-demand (2-minute reclaim notice) | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Commercial regions worldwide; OCI does not publish regional price variation | 32 regions on 5 continents, verified datacenters only |
| Minimums | BM.GPU.H100.8 and BM.GPU.H200.8 are 8-GPU shapes only | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Block Volume from $0.0255/GB/month | $0.08/GB/month NVMe, volumes survive instances |
| Egress | First 10 TB/month free per region/SKU, then $0.0085/GB | $0.01/GB in and out, every region |
| Access | Bare metal and VMs, OCI Console, CLI, API, Terraform, OKE (Kubernetes) | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Oracle Cloud column: as published on oracle.com on 2026-09-04. Details change — verify before you decide.

## Oracle Cloud vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Oracle Cloud's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Oracle Cloud list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (BM.GPU.H100.8, 8× GPU bare metal, $80/hr node ÷ 8) | $10.00 | **$1.428** | $0.714 | −86% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (BM.GPU.H200.8, 8× GPU bare metal, $80/hr node ÷ 8) | $10.00 | **$2.791** | $1.395 | −72% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (BM.GPU.A100.80, per GPU) | $4.00 | **$0.560** | $0.280 | −86% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (BM.GPU.L40S.4, 4× GPU bare metal, per GPU) | $3.50 | **$0.514** | $0.257 | −85% on PowerGPU |
| [A10](https://powergpu.ai/gpu/a10) (VM.GPU.A10.1, 1× GPU) | $2.00 | **$0.168** | $0.084 | −92% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Oracle Cloud price is quoted. Negative differences mean Oracle Cloud is cheaper on that card.

## When to stay with Oracle Cloud

- Your workloads already sit on OCI next to Autonomous Database, Fusion Apps or another Oracle-only service.
- A published, non-negotiated per-GPU rate card and enterprise support/SLA structure is what procurement wants.
- 10 TB of free egress per region/SKU each month covers a genuinely bandwidth-heavy pipeline.

## Why teams switch to PowerGPU

- Price: OCI bills the H100 and H200 at $10.00/GPU-hour and the A100 80 GB at $4.00/GPU-hour; PowerGPU fixes each at the public market median × 0.70.
- Single GPUs without a service-limit request: BM.GPU.H100.8 and BM.GPU.H200.8 are 8-GPU-only on OCI, while PowerGPU rents 1× to 8× at the same per-GPU price.
- Consumer and workstation cards (RTX 5090, 4090, RTX PRO 6000, RTX 6000 Ada) that OCI's GPU shapes do not include.
- Crypto payments with no KYC, per-second billing, and a flat −50% interruptible tier instead of a 2-minute-reclaim preemptible shape.

## Switching from Oracle Cloud: what maps to what

| On Oracle Cloud | On PowerGPU |
| --- | --- |
| BM.GPU.H100.8 / BM.GPU.H200.8 (8×) | 8× SXM machine, or 1×–4× at the same per-GPU price |
| BM.GPU.A100.80 / BM.GPU.L40S.4 / VM.GPU.A10.1 | A100 / L40S / A10 instance |
| Preemptible instance | Interruptible instance, flat −50%, auto-requeue |
| Block Volume | Volume, $0.08/GB/month |
| OCI Console / CLI / Terraform / OKE | powergpu dashboard / CLI / REST API / Terraform provider |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Oracle Cloud machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Oracle Cloud | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $7,300 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $1,600 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Block Volume from $0.0255/GB/month | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Oracle Cloud GPU alternative: FAQ

**Is PowerGPU cheaper than Oracle Cloud for GPUs?**

Yes, on every overlapping shape: OCI bills BM.GPU.H100.8 and BM.GPU.H200.8 at $10.00/GPU-hour and BM.GPU.A100.80 at $4.00/GPU-hour on-demand (checked 2026-09-04); PowerGPU fixes the same cards at the public market median × 0.70. OCI's own 50%-off preemptible shapes still land above PowerGPU on-demand on most cards.

**Do I need a GPU service-limit increase on PowerGPU like on OCI?**

No. New accounts have modest default limits that raise automatically with usage history; nothing starts at zero the way OCI GPU limits do on a trial tenancy.

**Can I rent a single H100 instead of an 8-GPU node?**

Yes — PowerGPU rents 1×, 2×, 4× or 8× H100 SXM at the same per-GPU price. OCI's BM.GPU.H100.8 bare-metal shape only comes as a full 8-GPU node.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Oracle Cloud, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Oracle Cloud and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Oracle Cloud. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/oracle-cloud · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/ovhcloud -->

---
title: "OVHcloud GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "OVHcloud vs PowerGPU: H100 PCIE price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/ovhcloud
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# OVHcloud GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

OVHcloud is a European infrastructure provider built around data sovereignty: self-service Public Cloud GPU instances (H100, A100, L40S, L4) billed hourly, plus dedicated Scale and HGR-AI bare-metal GPU servers sold by quote outside the self-service catalog. The current H100/A100/L40S/L4 flavors are concentrated in a single region, Gravelines (GRA11), while OVHcloud's non-GPU Public Cloud spans dozens of datacenters across Europe, North America and Asia-Pacific. Billing is hourly, payment is by card or SEPA, and most European and North American datacenters include unlimited outbound bandwidth.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## OVHcloud vs PowerGPU at a glance

OVHcloud

European sovereign cloud: hourly Public Cloud GPUs, Scale/HGR-AI bare metal by quote

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | OVHcloud | PowerGPU |
| --- | --- | --- |
| Billing | Per hour (Public Cloud GPU instances); Scale/HGR-AI dedicated servers quoted separately | Per second, no minimum, price locked at deploy |
| Payment | Card, SEPA direct debit, PayPal; invoicing for business accounts | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account and card/SEPA verification; KYB for dedicated bare metal | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Occasional promotional credits for new Public Cloud accounts (varies) | None — every hour is ≥30% under the market median instead |
| Cheaper tier | None published for GPU flavors; committed-spend discounts via sales quote | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | GPU flavors: Gravelines (GRA11) only today · non-GPU Public Cloud: 30+ regions across Europe, North America, APAC | 32 regions on 5 continents, verified datacenters only |
| Minimums | None on Public Cloud GPU instances (1×, 2×, 4× shown); dedicated Scale/HGR-AI servers sold whole | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Classic Volume about $0.047/GB/month (€0.04, converted) | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Unlimited and free in most EU and North American datacenters; 1 TB/month free elsewhere then €0.069/GB | $0.01/GB in and out, every region |
| Access | OpenStack-based VMs, Manager console, API, CLI, Terraform provider | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

OVHcloud column: as published on ovhcloud.com on 2026-09-04. Details change — verify before you decide.

## OVHcloud vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on OVHcloud's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | OVHcloud list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (h100-380 flavor, 1× GPU 80 GB, Public Cloud GRA11 (Gravelines) · h100-1 flavor $3.39) | $2.99 | **$1.867** | $0.933 | −38% on PowerGPU |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) (a100-180 flavor, 1× GPU 80 GB, Public Cloud GRA11) | $3.07 | **$0.374** | $0.187 | −88% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (l40s-1-gpu flavor, 1× GPU 48 GB, Public Cloud GRA11) | $1.69 | **$0.514** | $0.257 | −70% on PowerGPU |
| [L4](https://powergpu.ai/gpu/l4) (l4-1-gpu flavor, 1× GPU 24 GB, Public Cloud GRA11) | $0.91 | **$0.225** | $0.112 | −75% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which OVHcloud price is quoted. Negative differences mean OVHcloud is cheaper on that card.

## When to stay with OVHcloud

- Data sovereignty in a French/EU-headquartered cloud is a hard requirement for your workload.
- Free, unlimited outbound bandwidth in EU/North American datacenters matters more than the per-GPU rate for bandwidth-heavy jobs.
- You need a quoted Scale/HGR-AI bare-metal contract rather than a self-service hourly instance.

## Why teams switch to PowerGPU

- Every current OVHcloud GPU flavor sits in one region, Gravelines; PowerGPU fixes the same cards at market median × 0.70 across 32 regions on five continents.
- Consumer and workstation cards (RTX 5090, 4090, RTX PRO 6000) that OVHcloud does not sell in self-service Public Cloud.
- No card, no SEPA mandate, no KYC — top up in crypto and deploy in about 30 seconds instead of provisioning through the Manager console.
- A flat −50% interruptible tier and −35% reserved tier instead of a quote-only discount process.

## Switching from OVHcloud: what maps to what

| On OVHcloud | On PowerGPU |
| --- | --- |
| H100 / A100 / L40S / L4 instance (GRA11) | Instance — the same cards available in all 32 regions |
| Additional Disk (Classic Volume) | Volume, $0.08/GB/month |
| Scale / HGR-AI bare metal (quoted) | 8× SXM machine or a cluster at the same per-GPU price, no quote needed |
| OVHcloud Manager / API / Terraform | powergpu CLI / REST API / Terraform provider |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-pcie --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.867/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your OVHcloud machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | OVHcloud | PowerGPU |
| --- | --- | --- |
| 1× H100 PCIE, 730 hours on-demand | $2,183 | **$1,363** |
| 1× H100 PCIE, 8 h/day × 20 days | $478 | **$299** |
| Same 160 hours, PowerGPU interruptible | — | **$149** |
| 500 GB of storage, one month | Classic Volume about $0.047/GB/month (€0.04, converted) | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## OVHcloud GPU alternative: FAQ

**Is PowerGPU cheaper than OVHcloud for GPUs?**

On the flavors OVHcloud lists for self-service, yes: the H100 (h100-380) is $2.99/hr and the A100 80 GB (a100-180) is $3.07/hr on OVHcloud Public Cloud (checked 2026-09-04); PowerGPU fixes both at the public market median × 0.70. OVHcloud's L40S ($1.69/hr) and L4 ($0.91/hr) are also priced above PowerGPU's on-demand rate for the same cards.

**Does OVHcloud offer more GPU regions than Gravelines?**

Not yet for the current H100/A100/L40S/L4 flavors — they remain concentrated in GRA11 as of the date checked. PowerGPU runs the same catalog across 32 regions today.

**Is OVHcloud's free bandwidth better than PowerGPU's?**

OVHcloud's unlimited outbound bandwidth in EU/North American datacenters is real and worth factoring in for very large, repeated transfers. PowerGPU charges a flat $0.01/GB in and out, which most GPU workloads — weights, checkpoints, generated media — do not run up significantly against.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on OVHcloud, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

OVHcloud and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with OVHcloud. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/ovhcloud · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/prime-intellect -->

---
title: "Prime Intellect Alternative (2026): GPU Prices vs PowerGPU"
description: "Prime Intellect vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/prime-intellect
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Prime Intellect alternative: same NVIDIA GPUs, fixed prices 30% under the market

Prime Intellect aggregates on-demand GPU capacity from a growing set of integrated cloud and datacenter providers (12+ at its compute-exchange launch, with a stated roadmap past 20) into one dashboard, plus multi-node training clusters from 8 up to 1,024+ GPUs for distributed runs — the company is also known for coordinating open, community-run distributed training of frontier open-weight models. On-demand instances up to 8 GPUs deploy instantly; larger clusters are reserved by quote. Billing draws down a USD credit balance rather than a fixed catalogue price.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Prime Intellect vs PowerGPU at a glance

Prime Intellect

Multi-provider compute exchange: aggregated on-demand GPUs, multi-node clusters, decentralized training runs

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Prime Intellect | PowerGPU |
| --- | --- | --- |
| Billing | Per second or per hour depending on instance/provider; credits deducted while running | Per second, no minimum, price locked at deploy |
| Payment | USD credit balance, standard payment methods; other rails only "on request" per the FAQ | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Email account sign-up; no KYC documented in public docs | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | None standing — occasional promo codes and one-off programs (e.g. ICLR) only | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Quoted reserved clusters, 30-60% below on-demand for long-running workloads; no spot tier — all instances are on-demand | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 12+ integrated providers spanning the US, Finland, Norway, Canada and India; multi-node clusters hosted in US datacenters | 32 regions on 5 continents, verified datacenters only |
| Minimums | None on 1-8 GPU on-demand; multi-node clusters run 8 to 1,024+ GPUs, quoted | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Configured per instance at creation; no published flat $/GB rate | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not itemised on public pages | $0.01/GB in and out, every region |
| Access | SSH, Jupyter, REST API, CLI, multi-node clusters | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Prime Intellect column: as published on primeintellect.ai on 2026-09-04. Details change — verify before you decide.

## Prime Intellect vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Prime Intellect's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Prime Intellect list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (from; provider-set, up to $4.00 by provider/region) | $1.65 | **$1.428** | $0.714 | −13% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (from; provider-set, varies by provider/region) | $0.87 | **$0.560** | $0.280 | −36% on PowerGPU |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (from; provider-set, varies by provider/region) | $0.30 | **$0.327** | $0.163 | +9% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (from; third-party aggregate, provider-set) | $0.91 | **$0.514** | $0.257 | −44% on PowerGPU |
| [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) (from; third-party aggregate, up to $1.00 by provider) | $0.47 | **$0.281** | $0.140 | −40% on PowerGPU |
| [Tesla V100](https://powergpu.ai/gpu/tesla-v100) (from; third-party aggregate, up to $0.55 by provider) | $0.14 | **$0.130** | $0.065 | −7% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Prime Intellect price is quoted. Negative differences mean Prime Intellect is cheaper on that card.

## When to stay with Prime Intellect

- One dashboard shopping live rates across 12+ integrated providers beats comparing pricing pages yourself.
- Your work is coordinated distributed training on open-weight models — the community runs Prime Intellect organizes are exactly that.
- A 30-60% quoted discount on a long, predictable reserved cluster beats a fixed on-demand rate for your job length.

## Why teams switch to PowerGPU

- One price per card, not a range: Prime Intellect lists the H100 from $1.65/hr up to $4.00/hr depending on provider (checked 2026-09-04); PowerGPU fixes the same class at the public market median × 0.70, the same price regardless of which of 32 regions you land on.
- Crypto settlement Prime Intellect does not publish: USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX and SOL, no card, no KYC, against a USD credit balance billed through standard payment methods.
- A flat −50% interruptible tier and −35% reserved from 3 months, both published prices — Prime Intellect has no spot tier at all, and reserved pricing is quote-only.
- Storage at a flat $0.08/GB/month instead of configured per instance with no published rate.

## Switching from Prime Intellect: what maps to what

| On Prime Intellect | On PowerGPU |
| --- | --- |
| On-demand GPU (1-8x) | Instance — container in 30 s or full KVM VM |
| Multi-node cluster (8-1,024+ GPUs) | Clusters — 16 to 512 GPUs over InfiniBand, quoted from the public sheet |
| Quoted reserved cluster | Reserved instance, −35% from 3 months |
| Prime Intellect credit balance | USDT, BTC, XMR, LTC, ETH, TRX or SOL top-up |
| Prime Intellect dashboard / API | powergpu CLI / REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Prime Intellect machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Prime Intellect | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $1,205 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $264 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Configured per instance at creation; no published flat $/GB rate | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Prime Intellect alternative: FAQ

**Is PowerGPU cheaper than Prime Intellect?**

On the flagship cards, usually: Prime Intellect aggregates 50+ providers and lists the H100 from $1.65/hr up to $4.00/hr depending on which one fills the order, and the A100 from about $0.87/hr (checked 2026-09-04); PowerGPU fixes the same class at the public market median × 0.70, the same price regardless of region. Prime Intellect's cheapest aggregated listings on older cards can undercut PowerGPU in exchange for that provider-to-provider variance.

**Can PowerGPU replace a Prime Intellect distributed training cluster?**

For 16-512 GPU training pods over InfiniBand, yes — quoted per GPU from the public sheet and provisioned in days, on verified Tier-III machines rather than aggregated third-party capacity. Prime Intellect's own community-coordinated runs at extreme scale remain its own use case.

**Does PowerGPU take crypto like a decentralized compute market?**

Yes, more directly than Prime Intellect: USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX and SOL straight to a balance, no KYC. Prime Intellect bills in USD through standard payment methods; its own FAQ notes other rails only "on request".

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Prime Intellect, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Prime Intellect and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Prime Intellect. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/prime-intellect · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/replicate -->

---
title: "Replicate Alternative (2026): GPU Prices vs PowerGPU"
description: "Replicate vs PowerGPU: Tesla T4 price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/replicate
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Replicate alternative: same NVIDIA GPUs, fixed prices 30% under the market

Replicate runs models — thousands of public ones plus anything you package as a Cog container — behind a REST API and per-model versioning, with no server to manage. Billing is per second of GPU time actually used during a prediction, metered separately by hardware tier; some models are billed per output instead. There is no VM, no SSH and no way to keep a machine idle between requests — every run starts a fresh container.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Replicate vs PowerGPU at a glance

Replicate

Hosted models via API, Cog containers, billed per second of GPU time

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Replicate | PowerGPU |
| --- | --- | --- |
| Billing | Per second of GPU time metered per prediction; some models billed per output instead | Per second, no minimum, price locked at deploy |
| Payment | Card, debit card, or bank transfer, per Replicate's billing docs | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account signup; billing must be added once free-run limits are hit | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | No blanket credit; a curated "Try for free" collection allows a limited number of runs on specific models before billing is required | None — every hour is ≥30% under the market median instead |
| Cheaper tier | None published — a flat per-second rate regardless of volume | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Not published — no region selection exposed to the user | 32 regions on 5 continents, verified datacenters only |
| Minimums | None; billed only for the seconds a prediction actually runs | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not itemised — no user-facing volume product | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not itemised | $0.01/GB in and out, every region |
| Access | REST API and client SDKs (Python, Node, etc.) via Cog-packaged models; no SSH, no persistent VM | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Replicate column: as published on replicate.com on 2026-09-04. Details change — verify before you decide.

## Replicate vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Replicate's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Replicate list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) ($0.000225/s) | $0.81 | **$0.103** | $0.051 | −87% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) ($0.000975/s) | $3.51 | **$0.514** | $0.257 | −85% on PowerGPU |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) (80GB, form factor unspecified; $0.001400/s) | $5.04 | **$0.560** | $0.280 | −89% on PowerGPU |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) ($0.001525/s) | $5.49 | **$1.428** | $0.714 | −74% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) ($0.001525/s — same listed rate as H100) | $5.49 | **$2.791** | $1.395 | −49% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Replicate price is quoted. Negative differences mean Replicate is cheaper on that card.

## When to stay with Replicate

- You want to call a hosted model over an API without provisioning anything — thousands of ready-made public models are one request away.
- You publish your own model as a Cog container and want Replicate's hosting, versioning and API wrapper around it.
- Usage is genuinely spiky at the prediction level and you never want a machine sitting idle between calls.

## Why teams switch to PowerGPU

- Per-GPU-hour equivalent: an H100 works out to $5.49/hr on Replicate's meter with no volume discount; PowerGPU fixes the same card at the public market median × 0.70.
- A machine you actually keep: SSH, Jupyter, persistent volumes, full root — Replicate has none of that, every prediction is a fresh container with no shell.
- Crypto payment with no KYC and no card on file, instead of Replicate's card or bank-transfer billing.
- Serverless endpoints here too, scale-to-zero and billed per second, but on a fixed published GPU rate rather than a per-model meter that varies by hardware tier.

## Switching from Replicate: what maps to what

| On Replicate | On PowerGPU |
| --- | --- |
| Public or custom model via API | Serverless endpoint (bring the container — Cog images are OCI-compatible) |
| Prediction (per-second GPU meter) | Serverless endpoint, billed per second on the fixed GPU rate |
| Cog container | Any OCI image, on a serverless endpoint or an instance you keep running |
| Replicate API / client SDKs | powergpu CLI / REST API / Python SDK |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu tesla-t4 --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $0.103/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Replicate machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Replicate | PowerGPU |
| --- | --- | --- |
| 1× Tesla T4, 730 hours on-demand | $591 | **$75** |
| 1× Tesla T4, 8 h/day × 20 days | $130 | **$16** |
| Same 160 hours, PowerGPU interruptible | — | **$8** |
| 500 GB of storage, one month | Not itemised — no user-facing volume product | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Replicate alternative: FAQ

**Is PowerGPU cheaper than Replicate?**

Per GPU-hour equivalent, yes on every card compared: Replicate meters the H100 at $0.001525/s ($5.49/hr) and the A100 80GB at $0.001400/s ($5.04/hr), with no volume or reservation discount (checked 2026-09-04); PowerGPU fixes both at the public market median × 0.70.

**Can I run my Cog model on PowerGPU?**

Yes — a built Cog image is a standard OCI container; deploy it directly as a serverless endpoint, or on an instance if you want to keep the machine between calls.

**Does PowerGPU bill per second like Replicate?**

Serverless endpoints do, on a fixed per-GPU rate. Instances also bill per second, but you keep the machine running rather than paying per prediction.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Replicate, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Replicate and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Replicate. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/replicate · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/scaleway -->

---
title: "Scaleway GPU Alternative (2026): GPU Prices vs PowerGPU"
description: "Scaleway vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/scaleway
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Scaleway GPU alternative: same NVIDIA GPUs, fixed prices 30% under the market

Scaleway is a French cloud provider selling H100 SXM, L40S and L4 GPU instances by the hour across three EU regions — Paris, Amsterdam, Warsaw — plus older RENDER-S instances for rendering workloads. Egress bandwidth is free and intra-region transfers are free, which is unusual among the providers on this sheet. Payment is by card, billing is hourly, and there is no published spot or interruptible tier for GPU instances.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Scaleway vs PowerGPU at a glance

Scaleway

European GPU cloud: H100 SXM, L40S, L4 by the hour, free egress, Paris/Amsterdam/Warsaw

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Scaleway | PowerGPU |
| --- | --- | --- |
| Billing | Per hour, no minimum | Per second, no minimum, price locked at deploy |
| Payment | Card | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Card and account verification | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | Occasional promotional credits for new accounts (varies) | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Reserved pricing on some instance families via console; none published specifically for GPU instances | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | 3 — Paris, Amsterdam, Warsaw | 32 regions on 5 continents, verified datacenters only |
| Minimums | None (1× to 8× GPU configurations) | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Block storage about $0.101/GB/month (€0.087, converted) | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Free — no published egress fee | $0.01/GB in and out, every region |
| Access | VMs, Kubernetes (Kapsule), Scaleway CLI/API, Terraform provider | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Scaleway column: as published on scaleway.com on 2026-09-04. Details change — verify before you decide.

## Scaleway vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Scaleway's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Scaleway list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (H100-1-80G, 1× GPU 80 GB, Paris/Warsaw · €2.73/hr → USD @1.1627 (2026-09-04)) | $3.17 | **$1.428** | $0.714 | −55% on PowerGPU |
| [L40S](https://powergpu.ai/gpu/l40s) (L40S-1-48G, 1× GPU 48 GB, Paris · €1.47/hr → USD @1.1627 (2026-09-04)) | $1.71 | **$0.514** | $0.257 | −70% on PowerGPU |
| [L4](https://powergpu.ai/gpu/l4) (L4-1-24G, 1× GPU 24 GB · €0.79/hr → USD @1.1627 (2026-09-04)) | $0.92 | **$0.225** | $0.112 | −76% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Scaleway price is quoted. Negative differences mean Scaleway is cheaper on that card.

## When to stay with Scaleway

- EU-only data residency (Paris, Amsterdam, Warsaw) with no non-EU region is itself the requirement.
- Free egress matters more than the per-GPU rate for a workload that ships large result sets constantly.
- RENDER-S instances or another Scaleway-specific product are already load-bearing in your pipeline.

## Why teams switch to PowerGPU

- Price: Scaleway's H100 SXM starts at $3.17/hr (€2.73) and the L40S at $1.71/hr (€1.47); PowerGPU fixes both at the public market median × 0.70.
- 32 regions on five continents instead of three EU cities — useful the moment latency to users outside Europe matters.
- Consumer cards (RTX 5090, 4090, 3090) for diffusion and quantized LLM work, which Scaleway does not sell.
- Crypto payments with no KYC, and a flat −50% interruptible tier where Scaleway has none for GPU instances.

## Switching from Scaleway: what maps to what

| On Scaleway | On PowerGPU |
| --- | --- |
| H100-*/L40S-*/L4-* instance | Instance (container or full KVM VM) |
| Block storage volume | Volume, $0.08/GB/month |
| RENDER-S instance | Nearest equivalent consumer/workstation card at a fraction of the price |
| Scaleway CLI / API / Terraform | powergpu CLI / REST API |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Scaleway machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Scaleway | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $2,314 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $507 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Block storage about $0.101/GB/month (€0.087, converted) | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Scaleway GPU alternative: FAQ

**Is PowerGPU cheaper than Scaleway?**

Yes on every overlapping card: Scaleway lists the H100 SXM from €2.73/hr ($3.17) and the L40S from €1.47/hr ($1.71), checked 2026-09-04; PowerGPU fixes both at the public market median × 0.70. Scaleway's free egress narrows the gap for bandwidth-heavy jobs but does not close it on compute.

**Does PowerGPU have EU regions like Scaleway?**

Yes — ten European regions including Paris, Amsterdam and Warsaw, all Tier-III facilities, alongside the Americas, Asia-Pacific and other regions Scaleway does not offer.

**Is Scaleway's free egress better than PowerGPU's flat rate?**

For very large, repeated transfers, Scaleway's zero egress fee is cheaper in isolation. PowerGPU charges $0.01/GB, which most GPU workloads — weights, checkpoints, generated media — do not run up significantly against.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Scaleway, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Scaleway and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Scaleway. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/scaleway · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/alternatives/together-ai -->

---
title: "Together AI Alternative (2026): GPU Prices vs PowerGPU"
description: "Together AI vs PowerGPU: H100 SXM price, billing, payment (crypto, no KYC) and what maps to what when you switch. Public list prices checked 2026-09-04."
url: https://powergpu.ai/alternatives/together-ai
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Alternative · prices checked 2026-09-04

# Together AI alternative: same NVIDIA GPUs, fixed prices 30% under the market

Together AI is primarily an inference and fine-tuning platform: serverless model APIs billed per token, Dedicated Endpoints for a single model on reserved capacity, and GPU Clusters for teams that want the bare metal. GPU Clusters ship as multi-GPU HGX nodes (H100, H200, B200) with preemptible, on-demand and reserved pricing across five commitment tiers. Provisioning beyond a small allocation is sales-assisted rather than instant self-serve.

4 min read Updated 2026-09-14 (PowerGPU prices live from the sheet)

## Together AI vs PowerGPU at a glance

Together AI

Inference and fine-tuning API, plus dedicated GPU Clusters on H100, H200, B200

PowerGPU

Fixed prices at market median × 0.70, crypto only, verified datacenters

|  | Together AI | PowerGPU |
| --- | --- | --- |
| Billing | Hourly per GPU on Clusters; serverless inference billed per token instead | Per second, no minimum, price locked at deploy |
| Payment | Card; platform is prepaid with a $5 minimum credit purchase | Crypto only — USDT (TRC-20/ERC-20), BTC, XMR, LTC, ETH, TRX, SOL |
| Identity | Account signup; a card is required to fund the prepaid balance | Email + password. No KYC, no card, no stored IP addresses |
| Free credit | No general signup credit confirmed at time of check; a startup program grants up to $50,000 for qualifying companies | None — every hour is ≥30% under the market median instead |
| Cheaper tier | Preemptible around −50%; reserved from 7 to 180+ days, progressively cheaper | Interruptible flat −50% (no auction) · reserved −35% (3+ months) |
| Regions | Not broken out on the pricing page | 32 regions on 5 continents, verified datacenters only |
| Minimums | Dedicated Inference priced by 8-GPU HGX node; GPU Clusters quoted and billed per GPU | None — 1× to 8× GPUs at the same per-GPU price |
| Storage | Not published on the pricing page | $0.08/GB/month NVMe, volumes survive instances |
| Egress | Not published on the pricing page | $0.01/GB in and out, every region |
| Access | Inference API, fine-tuning API, GPU Clusters (Slurm/Kubernetes), CLI and SDK | Containers or full KVM VMs, SSH, Jupyter, REST API, CLI, Python SDK |

Together AI column: as published on together.ai on 2026-09-04. Details change — verify before you decide.

## Together AI vs PowerGPU prices, GPU by GPU

Public on-demand list prices per GPU-hour on Together AI's pricing page (2026-09-04) against PowerGPU's fixed rates today — every PowerGPU price is the public marketplace median × 0.70, rounded down, re-checked weekly.

| GPU | Together AI list | PowerGPU on-demand | PowerGPU interruptible | Difference |
| --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) (GPU Clusters on-demand, HGX H100 SXM, promotional rate (list $5.49)) | $3.99 | **$1.428** | $0.714 | −64% on PowerGPU |
| [H200](https://powergpu.ai/gpu/h200) (GPU Clusters on-demand, HGX H200 SXM) | $5.99 | **$2.791** | $1.395 | −53% on PowerGPU |
| [B200](https://powergpu.ai/gpu/b200) (GPU Clusters on-demand, HGX B200) | $8.19 | **$5.425** | $2.712 | −34% on PowerGPU |

Configurations differ (node sizes, tiers, regions); the note beside each card says which Together AI price is quoted. Negative differences mean Together AI is cheaper on that card.

## When to stay with Together AI

- Inference, fine-tuning and serving are your actual product, not raw GPU access — Together's managed model layer is more built-out than anything PowerGPU offers.
- You need hundreds to thousands of H100/H200/B200 GPUs on a reserved, sales-negotiated contract with InfiniBand.
- A 91+ day reservation already prices your H100 at $3.19/hr or below, close to a fixed-price on-demand card.

## Why teams switch to PowerGPU

- Self-serve on-demand price: PowerGPU fixes the H100 SXM at the public market median × 0.70, against Together's promotional $3.99/hr (list $5.49) with no reservation required.
- No prepaid-credit gate: PowerGPU has no $5 minimum purchase or card-on-file requirement — top up in crypto and the balance draws down per second.
- Consumer and workstation cards Together does not sell at all: RTX 5090, 4090, 3090, RTX 6000 Ada, A6000.
- A single GPU in about 30 seconds instead of a cluster allocation; interruptible here is a flat −50%, not a separate preemptible SKU with its own provisioning.

## Switching from Together AI: what maps to what

| On Together AI | On PowerGPU |
| --- | --- |
| GPU Cluster node (on-demand) | Instance — 1× to 8× GPUs at the same per-GPU price |
| Preemptible GPU Cluster | Interruptible instance, flat −50% |
| Reserved GPU Cluster (7–180+ days) | Reserved instance, −35% from 3 months |
| Dedicated Inference endpoint | Serverless endpoint (vLLM / ComfyUI / any container, scale to zero) |
| Together CLI / SDK | powergpu CLI / REST API / Python SDK |

*the whole migration, from the shell*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…          # console → API keys
powergpu launch --gpu h100-sxm --template pytorch --disk 100 --volume data:/data
# ✓ instance i-52ab77c1 running (24.1s) · $1.428/hr · per second
powergpu stop i-52ab77c1                      # billing ends this second
```

Data moves the boring way: rsync or rclone from your Together AI machine to a PowerGPU [volume](https://powergpu.ai/products/volumes), which then mounts on every future instance in seconds. Templates for [PyTorch](https://powergpu.ai/templates/pytorch), [vLLM](https://powergpu.ai/templates/vllm), [ComfyUI](https://powergpu.ai/templates/comfyui) and [Ollama](https://powergpu.ai/templates/ollama) are official images; anything else runs from its OCI reference.

## A real month, costed

| Scenario | Together AI | PowerGPU |
| --- | --- | --- |
| 1× H100 SXM, 730 hours on-demand | $2,913 | **$1,042** |
| 1× H100 SXM, 8 h/day × 20 days | $638 | **$228** |
| Same 160 hours, PowerGPU interruptible | — | **$114** |
| 500 GB of storage, one month | Not published on the pricing page | $40 |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares against the marketplace median.

## Together AI alternative: FAQ

**Is PowerGPU cheaper than Together AI for raw GPUs?**

On the GPU Clusters product, yes: Together lists on-demand HGX H100 at $3.99/hr (promotional; list $5.49) and HGX H200 at $5.99/hr (checked 2026-09-04); PowerGPU fixes both at the public market median × 0.70. Together's long-term reservations (91+ days) close much of that gap.

**Can PowerGPU replace Together's inference API?**

Not directly — Together's serverless model endpoints and fine-tuning pipeline are a managed product PowerGPU does not build. PowerGPU's serverless endpoints run a container you bring (vLLM, TGI, your own image) rather than a hosted model catalog.

**Do I need a sales call to get GPUs on PowerGPU like Together's clusters?**

No — single GPUs and small multi-GPU machines deploy self-serve in about 30 seconds. Multi-node InfiniBand clusters (16 to 512 GPUs) go through provisioning but not a prepaid-credit gate.

---

Try the switch for the price of a coffee

Top up $40 in USDT or Monero, deploy the same image you run on Together AI, and benchmark it. No card, no KYC, per-second billing — stop it the minute you are done.

Together AI and its logo are trademarks of their owner, used here only to identify the compared service; PowerGPU is not affiliated with Together AI. NVIDIA GPU names are trademarks of NVIDIA Corporation.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/alternatives/together-ai · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides -->

---
title: "Cloud GPU Guides: VRAM Sizing, Costs, Walkthroughs | PowerGPU"
description: "Practical cloud GPU guides with live prices: LLM VRAM requirements, H100 vs H200 vs B200, QLoRA fine-tuning, vLLM serving, Stable Diffusion costs, Blender rendering."
url: https://powergpu.ai/guides
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guides · 16 articles · numbers included

# Cloud GPU guides where the prices are never stale

Every figure in these articles is rendered live from our price sheet — the same data the console bills from. Written by the team that runs the fleet, updated with every weekly market re-check.

[Start here · Choosing hardware LLM VRAM requirements: how much GPU memory for 7B–405B models Complete sizing tables for running and training LLMs — FP16, INT8 and 4-bit, with KV-cache math and the cheapest GPU that fits each model. 12 min read · updated 2026-09-03 Read the guide](https://powergpu.ai/guides/llm-vram-requirements)

## All guides (16)

Costs, hardware choices and hands-on walkthroughs — every number on these pages is today's sheet.

[Costs & pricing **Cloud GPU pricing explained (2026): on-demand vs spot vs reserved** What actually drives cloud GPU prices in 2026, how marketplaces and hyperscalers differ, and when each billing mode saves you money. 9 min read updated 2026-09-03](https://powergpu.ai/guides/cloud-gpu-pricing-explained) [Choosing hardware **LLM VRAM requirements: how much GPU memory for 7B–405B models** Complete sizing tables for running and training LLMs — FP16, INT8 and 4-bit, with KV-cache math and the cheapest GPU that fits each model. 12 min read updated 2026-09-03](https://powergpu.ai/guides/llm-vram-requirements) [Choosing hardware **H100 vs H200 vs B200 (2026): specs, price per hour, which to rent** Memory, bandwidth and real rental economics of NVIDIA's three datacenter flagships — and when the older card is the better deal. 10 min read updated 2026-09-03](https://powergpu.ai/guides/h100-vs-h200-vs-b200) [Choosing hardware **RTX 5090 vs RTX 4090 for AI (2026): benchmarks, VRAM, rental cost** Specs, VRAM, real throughput differences and cost per run for inference, fine-tuning and image generation on both consumer flagships. 8 min read updated 2026-09-03](https://powergpu.ai/guides/rtx-4090-vs-rtx-5090) [Hands-on walkthrough **How to fine-tune Llama 3.1 8B with QLoRA on a single GPU** A complete, copy-pasteable walkthrough: dataset to merged weights in about an hour on a single 24 GB card, with axolotl. 14 min read updated 2026-09-03](https://powergpu.ai/guides/fine-tune-llm-qlora) [Hands-on walkthrough **How to deploy vLLM on a cloud GPU: an OpenAI-compatible endpoint** Deploy vLLM on a rented GPU, pick the right card for your model size, benchmark tokens per second and put a price on every million tokens. 11 min read updated 2026-09-03](https://powergpu.ai/guides/serve-llm-vllm) [Costs & pricing **Cheapest cloud GPU for Stable Diffusion & Flux in 2026 ($/image)** What SDXL and Flux actually need, images-per-dollar on eight rentable GPUs, and where paying more per hour costs less per image. 9 min read updated 2026-09-03](https://powergpu.ai/guides/cheapest-gpu-for-stable-diffusion) [Hands-on walkthrough **Blender cloud rendering on GPUs: setup, cost per frame, pitfalls** Render Cycles scenes on rented RTX hardware: headless setup, per-frame cost math, and the mistakes that quietly triple a render bill. 10 min read updated 2026-09-03](https://powergpu.ai/guides/blender-cloud-rendering) [Costs & pricing **How much does it cost to rent an H100? Per-hour math for 2026** H100 SXM, PCIe and NVL rental prices per hour and per month, what marketplaces and hyperscalers charge, the rent-vs-buy break-even, and five ways to pay less. 9 min read updated 2026-09-03](https://powergpu.ai/guides/h100-rental-cost-per-hour) [Choosing hardware **Best cloud GPU for LLM inference in 2026, by model size** From 8B to 405B: the cheapest rentable card that holds each model, indicative tokens per second with vLLM, and the cost per million tokens on today's sheet. 10 min read updated 2026-09-03](https://powergpu.ai/guides/best-cloud-gpu-for-llm-inference) [Hands-on walkthrough **How to run ComfyUI on a cloud GPU: setup, models, cost per image** Deploy ComfyUI on a rented RTX 4090 or 5090 in 30 seconds, keep checkpoints on a volume, run workflows headless through the API, and know what each image costs. 9 min read updated 2026-09-03](https://powergpu.ai/guides/run-comfyui-on-a-cloud-gpu) [Hands-on walkthrough **How to rent a GPU with crypto and no KYC (2026)** Renting cloud GPUs with USDT, Bitcoin or Monero and no identity check: how top-ups work, which coin to use, what data is kept, and the step-by-step deploy. 7 min read updated 2026-09-03](https://powergpu.ai/guides/rent-gpu-with-crypto-no-kyc) [Choosing hardware **A100 vs H100 for fine-tuning (2026): cost per run, not per hour** Same 80 GB, 2.7× the hourly price: when an H100 finishes a LoRA, QLoRA or full fine-tune fast enough to beat the A100 on total cost — live prices, break-even rule. 8 min read updated 2026-09-03](https://powergpu.ai/guides/a100-vs-h100-for-fine-tuning) [Costs & pricing **Cheapest cloud GPU for Ollama (2026): 8B to 70B models, by the hour** Which rented card runs each Ollama model size at Q4, indicative tokens per second per card, and what an always-on private assistant costs per month. 8 min read updated 2026-09-03](https://powergpu.ai/guides/cheapest-gpu-for-ollama) [Choosing hardware **Wan 2.x video generation: which GPU, minutes per clip, cost per clip** VRAM needs for Wan 2.1 and 2.2 (1.3B, 5B, 14B), realistic minutes per five-second clip on RTX 4090, 5090, L40S, H100 and H200, and the price of a batch night. 9 min read updated 2026-09-03](https://powergpu.ai/guides/wan-video-generation-gpu) [Costs & pricing **Best GPU for Whisper transcription (2026): speed and cost per audio hour** faster-whisper large-v3 throughput on T4, L4, RTX 3060, RTX 4090 and H100 — and the only number that matters: cents per hour of audio transcribed. 7 min read updated 2026-09-03](https://powergpu.ai/guides/best-gpu-for-whisper)

## Looking for reference material instead?

The [documentation](https://powergpu.ai/docs) covers the platform itself; the [use-case playbooks](https://powergpu.ai/use-cases) pre-match GPUs to workloads; the [API reference](https://powergpu.ai/api) has curl you can run before signing up.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/cloud-gpu-pricing-explained -->

---
title: "Cloud GPU pricing explained (2026): on-demand vs spot vs reserved"
description: "What actually drives cloud GPU prices in 2026, how marketplaces and hyperscalers differ, and when each billing mode saves you money."
url: https://powergpu.ai/guides/cloud-gpu-pricing-explained
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Costs & pricing

# Cloud GPU pricing explained (2026): on-demand vs spot vs reserved

What actually drives cloud GPU prices in 2026, how marketplaces and hyperscalers differ, and when each billing mode saves you money.

9 min read Published 2026-07-14 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Cloud GPU prices split three ways: hyperscalers at $4-7 per H100-hour, marketplaces with moving medians, and fixed-price clouds like PowerGPU at median × 0.70.
- On PowerGPU an H100 SXM is $1.428 per hour on-demand, 30% below the $2.04 public marketplace median.
- Interruptible costs exactly half of on-demand; reserved is 65 percent of it and wins above roughly 65 percent utilisation.
- Hidden costs: egress at $0.05-0.12 per GB on hyperscalers, idle stopped disks, and hourly rounding that bills a 61-minute job as two hours.

## Who sells GPU time, and how

Three kinds of sellers set today's market:

- **Hyperscalers** (AWS, GCP, Azure) — list prices, enterprise contracts, and H100-class instances typically at $4–7 per GPU-hour before committed-use gymnastics.
- **GPU marketplaces** — thousands of independent hosts auctioning capacity. Deep supply and low medians, but the price is a moving target and the host quality is a distribution, not a promise.
- **Fixed-price GPU clouds** — the newer category we belong to: capacity bought in bulk from verified datacenters, resold at a published flat rate. Our rule is mechanical: *marketplace median × 0.70, rounded down, re-checked weekly*. Today that puts the H100 SXM at **$1.428** /hr against a $2.04 median.

## What actually moves prices

GPU-hour prices are supply economics with a silicon accent:

- **Generation launches** — every Blackwell shipment pushes Hopper prices down a step; the 90-day drift on H100 medians is visible in any public feed.
- **VRAM per card** — memory sells the hour. 80 GB parts hold a price floor long after their FLOPS are matched by consumer cards, because model sizes grew faster than compute needs.
- **Electricity and density** — hosts with cheap power and dense racks undercut; that is why medians differ by region and why our fleet skews to power-cheap regions.
- **Bursts of demand** — a hot open-weights release can double marketplace spot prices for a week. Fixed pricing exists precisely to opt out of that volatility.

## On-demand vs interruptible vs reserved

| Mode | Price rule | H100 SXM today | Use when |
| --- | --- | --- | --- |
| [On-demand](https://powergpu.ai/products/on-demand) | median × 0.70 ↓ | $1.428/hr | stateful, interactive, deadline work |
| [Interruptible](https://powergpu.ai/products/interruptible) | od × 0.50 | $0.714/hr | checkpointed training, batch queues |
| [Reserved (3 mo)](https://powergpu.ai/products/reserved) | od × 0.65 | $0.928/hr | utilisation above ~65%, production serving |

The industry uses "spot" for our interruptible tier, but classic spot is an auction: you bid, you win, a higher bid evicts you. A flat −50% with stop-not-destroy semantics behaves very differently in practice — the discount is predictable, so pipelines can be designed around it instead of around bid strategy.

## The break-even math

Two formulas cover 90% of purchasing decisions:

*break-even rules*

```
reserved beats on-demand when:
    utilisation > res_rate / od_rate            (= 65% here)

interruptible beats on-demand when:
    (1 + overhead) x 0.50 < 1
    i.e. restart overhead under 100% of runtime — checkpointing
    every 15 min on a 6 h job is ~4% overhead, not 100%.
```

Concrete: a fine-tune that needs 200 GPU-hours of A100 SXM4 per month —

- on-demand: 200 × $0.560 = **$112**;
- interruptible with 5% restart overhead: 210 × $0.280 = **$59** — the obvious winner;
- reserved only wins here at 475+ monthly hours (65% of 730).

## Where bills quietly grow

- **Egress** — hyperscalers charge $0.05–0.12/GB out; moving a 2 TB dataset off can cost more than the training run. Flat $0.01/GB (ours) or free egress changes which workflows are even viable.
- **Idle storage** — forgotten stopped instances bill their disks forever. Our dashboard surfaces runway and stopped-disk burn; check whatever provider you use for the same view.
- **Hourly rounding** — per-hour billing turns a 61-minute job into 2 hours. Per-second billing is worth 0–49% on short jobs — the shorter the job, the bigger the gap.
- **Multi-GPU premiums** — some sellers price 8× machines above 8× the single price. Check the multiplication; here it is exactly linear.

## A buyer's checklist

1. Compute $/hr *per GPU*, not per instance, and normalise VRAM (a [RTX 5090](https://powergpu.ai/gpu/rtx-5090) hour buys 32 GB; an 80 GB card should justify its multiple).
2. Ask what happens at eviction: auction re-price, destroy, or stop-with-disk?
3. Price the full loop: GPU + storage-month + egress of your artefacts.
4. Prefer sellers who publish their pricing rule — if the rule is secret, it can move against you.
5. Benchmark once: an hour of testing on a $0.163 interruptible [RTX 4090](https://powergpu.ai/gpu/rtx-4090) answers throughput questions no spec sheet can.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/cloud-gpu-pricing-explained · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/llm-vram-requirements -->

---
title: "LLM VRAM requirements: how much GPU memory for 7B–405B models"
description: "Complete sizing tables for running and training LLMs — FP16, INT8 and 4-bit, with KV-cache math and the cheapest GPU that fits each model."
url: https://powergpu.ai/guides/llm-vram-requirements
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Choosing hardware

# LLM VRAM requirements: how much GPU memory for 7B–405B models

Complete sizing tables for running and training LLMs — FP16, INT8 and 4-bit, with KV-cache math and the cheapest GPU that fits each model.

12 min read Published 2026-07-21 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Weights need 2 GB per billion parameters in FP16, 1 GB in INT8 and about 0.55 GB in 4-bit.
- Add 10-20 percent runtime overhead plus KV-cache: Llama 3.1 8B costs about 0.13 GB per 1k tokens per sequence.
- Practical fits: 8B is 18 GB FP16 or 6.5 GB in 4-bit; Llama 70B is 150 GB FP16, 44 GB in 4-bit.
- PowerGPU prices multi-GPU linearly, so two RTX 5090s at $0.439 each give 64 GB for 70B 4-bit serving.

## The 60-second rule of thumb

*the whole method*

```
weights_gb  = params_B x bytes_per_param      # fp16 2.0 · int8 1.0 · 4-bit ~0.55
runtime_gb  = weights_gb x 1.1..1.2           # CUDA ctx, activations, fragmentation
total_gb    = runtime_gb + kv_cache_gb        # see below — grows with context
```

That is genuinely the whole method. The table below applies it to the models people actually deploy, with the cheapest card on our sheet that fits each cell.

## The sizing table (inference)

| Model | FP16 | INT8 | 4-bit | Cheapest card that fits (4-bit) | $/hr |
| --- | --- | --- | --- | --- | --- |
| **Llama 3.2 3B** | 7 GB | 4.5 GB | 2.8 GB | [RTX 3060 · 12 GB](https://powergpu.ai/gpu/rtx-3060) | $0.042 |
| **Llama 3.1 8B** | 18 GB | 10 GB | 6.5 GB | [RTX 3090 · 24 GB](https://powergpu.ai/gpu/rtx-3090) | $0.108 |
| **Qwen 2.5 14B** | 31 GB | 17 GB | 10.5 GB | [RTX 4090 · 24 GB](https://powergpu.ai/gpu/rtx-4090) | $0.327 |
| **Qwen 2.5 32B** | 70 GB | 37 GB | 21 GB | [RTX 5090 · 32 GB](https://powergpu.ai/gpu/rtx-5090) | $0.439 |
| **Llama 3.1 70B** | 150 GB | 78 GB | 44 GB | [H100 PCIE · 80 GB](https://powergpu.ai/gpu/h100-pcie) | $1.867 |
| **Qwen 2.5 72B** | 155 GB | 80 GB | 46 GB | [H100 PCIE · 80 GB](https://powergpu.ai/gpu/h100-pcie) | $1.867 |
| **Mistral Large 123B** | 260 GB | 133 GB | 78 GB | [H200 · 141 GB](https://powergpu.ai/gpu/h200) | $2.791 |
| **Llama 3.1 405B** | 850 GB | 440 GB | 245 GB | [B200 · 192 GB](https://powergpu.ai/gpu/b200) | $5.425 |

Weights + 15% overhead, excluding KV-cache; 405B rows assume multi-GPU sharding (per-card price shown). Figures are engineering estimates for planning, not benchmarks.

## KV-cache: the part everyone forgets

Serving crashes rarely come from weights — they come from context. Every token in flight stores keys and values for every layer:

*why the 24 GB card OOMs at batch 32*

```
kv_gb = 2 x layers x kv_heads x head_dim x bytes x context x batch / 1e9

# Llama 3.1 8B (GQA, fp16): ~0.13 GB per 1k tokens per sequence
# 32 concurrent chats x 8k context ≈ 33 GB of cache — MORE than the weights
```

Levers, in order of cheapness: quantize the cache (FP8 KV halves it), cap concurrent context (vLLM's --max-num-batched-tokens), then buy VRAM. This is why serving pages recommend 32–48 GB cards for "models that fit in 16 GB".

## Training & fine-tuning VRAM

| Method | VRAM ≈ | 8B lands on | 70B lands on |
| --- | --- | --- | --- |
| Full fine-tune, AdamW fp16 | 16 GB / B params | 8× 24 GB or 2× 80 GB | 16× 80 GB (cluster) |
| Full, 8-bit optimizer | ~10 GB / B | 1× 80 GB | 9× 80 GB |
| LoRA (fp16 base) | weights + ~2 GB | 1× 24 GB | 2× 80 GB |
| QLoRA (4-bit base) | ~0.7 GB / B + 2 GB | 1× 12 GB | 1× 48–80 GB |

Gradient checkpointing trades ~20% speed for ~30% memory and is on by default in axolotl configs — the [QLoRA walkthrough](https://powergpu.ai/guides/fine-tune-llm-qlora) shows real consumption at each step.

## When one card is not enough

- **Tensor parallelism** (vLLM --tensor-parallel-size) splits layers across 2–8 GPUs on one machine — NVLink helps but PCIe 4/5 serves fine to 4×.
- **Our pricing is linear**: 2× [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (64 GB total) costs exactly 2× $0.439 — often the cheapest 70B-4bit rig on the sheet.
- **Past 8 GPUs**, you are in [cluster](https://powergpu.ai/products/clusters) territory: pipeline or FSDP sharding over InfiniBand.

Cross-check any plan against the per-card fit lists on the [GPU pages](https://powergpu.ai/gpus) — each fiche computes what it can hold from these same formulas.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/llm-vram-requirements · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/h100-vs-h200-vs-b200 -->

---
title: "H100 vs H200 vs B200 (2026): specs, price per hour, which to rent"
description: "Memory, bandwidth and real rental economics of NVIDIA's three datacenter flagships — and when the older card is the better deal."
url: https://powergpu.ai/guides/h100-vs-h200-vs-b200
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Choosing hardware

# H100 vs H200 vs B200 (2026): specs, price per hour, which to rent

Memory, bandwidth and real rental economics of NVIDIA's three datacenter flagships — and when the older card is the better deal.

10 min read Published 2026-08-04 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- The H100 gives 80 GB at 3.35 TB/s, the H200 141 GB at 4.8 TB/s, the B200 192 GB at 8 TB/s.
- On PowerGPU the H100 SXM rents at $1.428 per hour, the H200 at $2.791 and the B200 at $5.425.
- Pick the H200 for long context and memory-bound work: its 43 percent bandwidth jump shows up almost one-to-one.
- The B200 delivers 2,250 FP16 TFLOPS, roughly 2.3× the H100, so it buys deadlines rather than cheaper FLOPS.

## Specs that matter, side by side

|  | A100 SXM4 | H100 SXM | H200 | B200 |
| --- | --- | --- | --- | --- |
| Architecture | Ampere | Hopper | Hopper | Blackwell |
| VRAM | 80 GB HBM2e | 80 GB HBM3 | **141 GB** HBM3e | **192 GB** HBM3e |
| Memory bandwidth | 2.0 TB/s | 3.35 TB/s | 4.8 TB/s | 8 TB/s |
| FP16 tensor (dense) | 312 TF | 990 TF | 990 TF | 2,250 TF |
| Low-precision extras | — | FP8 | FP8 | FP8 + FP4 |
| NVLink | 600 GB/s | 900 GB/s | 900 GB/s | 1.8 TB/s |

Public NVIDIA figures, dense (non-sparsity) numbers.

## Rental economics today

Specs age; the ratio of spec to price is the actual decision. Live from [the sheet](https://powergpu.ai/pricing) (snapshot 2026-09-14):

| Card | On-demand | Interruptible | $/hr per TB/s | $/hr per 100 FP16 TF |
| --- | --- | --- | --- | --- |
| [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | $0.560 | $0.280 | $0.280 | $0.179 |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | $1.428 | $0.714 | $0.426 | $0.144 |
| [H200](https://powergpu.ai/gpu/h200) | $2.791 | $1.395 | $0.581 | $0.282 |
| [B200](https://powergpu.ai/gpu/b200) | $5.425 | $2.712 | $0.678 | $0.241 |

Read the last two columns as "price of bandwidth" and "price of compute". The H100 usually wins compute per dollar; the H200 wins bandwidth per dollar; the B200 buys time.

## Pick the H100 when…

- your model + optimizer fits in 80 GB per shard (most ≤34B full fine-tunes, all LoRA work);
- throughput per dollar is the metric — FP8 on Hopper remains the efficiency sweet spot;
- you want depth of supply: H100 pools are the deepest of the three, so [interruptible](https://powergpu.ai/products/interruptible) slots are nearly always available.

## Pick the H200 when…

- context length is the product: KV-cache at 128k eats 80 GB cards alive, 141 GB breathes;
- the workload is memory-bound (MoE inference, stencil HPC, giant batch serving) — the 43% bandwidth jump shows up almost 1:1;
- you can trade a $1.363/hr premium for fewer, fatter shards (less inter-GPU traffic, simpler parallelism).

## Pick the B200 when…

- the deadline is the budget: ~2.3× H100 FP16 throughput means Monday's checkpoint on Friday;
- FP4/FP8 inference at scale — Blackwell's low-precision path is where its $/token leadership lives;
- the model shard needs 192 GB — some 100B+ configurations simply do not fit anything else per card.

## The A100 dark horse

Four years old and still the best $/GB of HBM on the sheet: $0.560/hr for 80 GB. For ≤13B training runs, LoRA farms and FP64 simulation it routinely beats the newer cards on total job cost — the software stack is bulletproof and interruptible A100 supply is deep. Run the [break-even math](https://powergpu.ai/guides/cloud-gpu-pricing-explained) before assuming newer is cheaper.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/h100-vs-h200-vs-b200 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/rtx-4090-vs-rtx-5090 -->

---
title: "RTX 5090 vs RTX 4090 for AI (2026): benchmarks, VRAM, rental cost"
description: "Specs, VRAM, real throughput differences and cost per run for inference, fine-tuning and image generation on both consumer flagships."
url: https://powergpu.ai/guides/rtx-4090-vs-rtx-5090
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Choosing hardware

# RTX 5090 vs RTX 4090 for AI (2026): benchmarks, VRAM, rental cost

Specs, VRAM, real throughput differences and cost per run for inference, fine-tuning and image generation on both consumer flagships.

8 min read Published 2026-08-11 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Rent the RTX 4090 for SDXL batches and 8-13B QLoRA; rent the RTX 5090 whenever a job needs 24-32 GB.
- The RTX 5090 brings 32 GB GDDR7, 1.79 TB/s and 419 TFLOPS against the 4090's 24 GB, 1.0 TB/s and 330 TFLOPS.
- On PowerGPU the 4090 is $0.327 per hour and the 5090 $0.439; the 4090 still wins images per dollar on SDXL.
- QLoRA ceilings are about 13B on 24 GB and 34B on 32 GB; video models need 32 GB or more.

## Side by side

|  | RTX 4090 | RTX 5090 |
| --- | --- | --- |
| Architecture | Ada Lovelace | Blackwell |
| VRAM | 24 GB GDDR6X | **32 GB** GDDR7 |
| Memory bandwidth | 1.0 TB/s | 1.79 TB/s |
| FP16 tensor (dense) | 330 TF | 419 TF |
| Rent, on-demand | $0.327/hr | $0.439/hr |
| Rent, interruptible | $0.163/hr | $0.219/hr |

The premium is $0.112/hr (34%) today. The question is always: does the job use the extra 8 GB or the extra bandwidth? If not, the 4090's price wins.

## LLM inference math

- **8B class (fits both)** — the 5090's bandwidth pushes ~1.6–1.8× the tokens/s; per token that beats its 1.34× price. Win: 5090, narrowly.
- **14B class** — FP16 needs ~31 GB: 5090 serves it native, 4090 must quantize. Win: 5090.
- **32B 4-bit (~21 GB weights)** — loads on both, but KV-cache drowns the 4090 at real concurrency. Win: 5090, decisively.
- **Batch embeddings / Whisper** — compute-bound small models: rent whichever is cheaper per hour that day; it is usually the [4090](https://powergpu.ai/gpu/rtx-4090).

## Image & video generation

SDXL at 1024px is not VRAM-bound: the 4090's ~25–30% speed deficit is smaller than its 26% price advantage — **more images per dollar on the 4090**. Flux dev flips it: FP16 weights + text encoders brush against 24 GB, and every offload event stalls the 4090 while the 5090 keeps everything resident. Video models (Wan, LTX) are 5090-or-bigger territory — see the [video playbook](https://powergpu.ai/use-cases/video-generation).

## Fine-tuning

QLoRA ceilings: ~13B on 24 GB, ~34B on 32 GB. If your target model is Qwen 32B, the 5090 is the cheapest single-card trainer on the sheet; at 8B, the 4090 (or even a [3090 at $0.108/hr](https://powergpu.ai/gpu/rtx-3090)) does the same epochs for less. Walkthrough with live consumption numbers: [QLoRA guide](https://powergpu.ai/guides/fine-tune-llm-qlora).

## The verdict table

| Job | Rent | Why |
| --- | --- | --- |
| SDXL batches | **4090** | images/$ wins |
| Flux heavy workflows | **5090** | stays resident in 32 GB |
| 7–8B serving | **5090** | bandwidth → tokens/$ |
| 32B 4-bit serving | **5090** | only one with cache headroom |
| 8–13B QLoRA | **4090** | same result, lower rate |
| 34B QLoRA | **5090** | single-card ceiling |
| Video generation | **5090+** | 24 GB is below entry |

Undecided? Rent both for an hour — $0.766 total on-demand — and benchmark your actual workload. That experiment costs less than this article took to read.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/rtx-4090-vs-rtx-5090 · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/fine-tune-llm-qlora -->

---
title: "How to fine-tune Llama 3.1 8B with QLoRA on a single GPU | PowerGPU"
description: "A complete, copy-pasteable walkthrough: dataset to merged weights in about an hour on a single 24 GB card, with axolotl."
url: https://powergpu.ai/guides/fine-tune-llm-qlora
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Hands-on walkthrough

# How to fine-tune Llama 3.1 8B with QLoRA on a single GPU

A complete, copy-pasteable walkthrough: dataset to merged weights in about an hour on a single 24 GB card, with axolotl.

14 min read Published 2026-08-18 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- A QLoRA fine-tune of Llama 3.1 8B on 10,000 instruction pairs, three epochs, takes about 95 minutes on one RTX 4090.
- Cost on PowerGPU is roughly 1.6 hours at the interruptible rate of $0.163 per hour, plus a 20 GB volume.
- 24 GB is enough: the 4-bit 8B base takes about 5.5 GB and peak VRAM is near 14 GB at batch 4, sequence 2048.
- Key config: lora_r 32, alpha 64, all linear layers, gradient checkpointing, and save_steps 100 so an interruption costs one resume.

## Setup: one instance, one volume

Everything durable goes on a volume; the GPU instance stays disposable:

*infrastructure, 2 commands*

```
powergpu volume create --name ft --size 20 --region eu-west-1
powergpu launch --gpu rtx-4090 --type interruptible \
  --template axolotl --volume ft:/ft --disk 40
# ✓ i-c41b9a02 running · $0.163/hr
```

Why interruptible: axolotl checkpoints to /ft, so an interruption costs one resume, not the run. Why 24 GB is enough: QLoRA holds the 8B base in 4-bit (~5.5 GB) + adapters + optimizer (~3 GB) + activations — peaks near 14 GB at batch 4, sequence 2048. A [RTX 3090](https://powergpu.ai/gpu/rtx-3090) at $0.054/hr does it too, ~35% slower.

## The dataset format

*data.jsonl*

```
// /ft/data.jsonl — one instruction pair per line
{"instruction": "Summarize this ticket for an engineer.",
 "input": "Customer reports…",
 "output": "P1 — checkout API 500s when…"}
```

Quality beats volume decisively at this scale: 2,000 clean, consistent pairs outperform 50k scraped ones. Dedupe, strip formatting noise, keep outputs in exactly the voice you want back.

## The axolotl config, annotated

*qlora.yml — the whole config*

```
# /ft/qlora.yml
base_model: meta-llama/Llama-3.1-8B-Instruct
load_in_4bit: true            # the Q in QLoRA
adapter: qlora
lora_r: 32                    # capacity of the adapter; 16-64 is the sane band
lora_alpha: 64                # 2x r is the boring, correct default
lora_target_linear: true      # all linear layers — better than picking modules

datasets:
  - path: /ft/data.jsonl
    type: alpaca
val_set_size: 0.05

sequence_len: 2048
micro_batch_size: 4
gradient_accumulation_steps: 4   # effective batch 16
num_epochs: 3
learning_rate: 2e-4
lr_scheduler: cosine
warmup_ratio: 0.03

gradient_checkpointing: true  # ~30% VRAM back for ~20% speed
flash_attention: true
bf16: true

output_dir: /ft/out
save_steps: 100               # ← interruption insurance
logging_steps: 10
```

## Run and watch

*the run*

```
axolotl train /ft/qlora.yml
# step 10/561 · loss 1.842 · 14.1 GB VRAM · 3.4 it/s
# …
# step 561/561 · loss 0.914 · eval_loss 0.987 · 94 min
```

Read the curves like this: train loss should fall fast then flatten; eval loss following it down means learning, eval loss rising while train falls means memorising — stop at the divergence (or raise val_set_size and lower epochs). For a first run, 3 epochs on 10k pairs almost never overfits an 8B.

## Merge, test, serve

*from adapters to endpoint*

```
# merge adapters into standalone weights (on /ft, survives the instance)
axolotl merge-lora /ft/qlora.yml --lora-model-dir /ft/out

# smoke-test locally
python -m vllm.entrypoints.openai.api_server \
  --model /ft/out/merged --max-model-len 4096 &
curl localhost:8000/v1/chat/completions -d '{…}'

# then destroy the trainer and serve properly
powergpu destroy i-c41b9a02 --yes
powergpu launch --gpu rtx-5090 --template vllm \
  --volume ft:/ft:ro --env MODEL=/ft/out/merged
```

Serving card choice is a different optimisation than training — the [vLLM guide](https://powergpu.ai/guides/serve-llm-vllm) picks it properly.

## The final bill

- **GPU, ~96 min interruptible**: $0.26
- **Volume 20 GB, 2 days**: $0.107
- **Total**: **≈ $0.37**

A custom-behaviour 8B for the price of a sandwich. The same recipe scales: 14B on the same card, ~34B on a [32 GB card](https://powergpu.ai/gpu/rtx-5090), 70B on an [80 GB card](https://powergpu.ai/gpu/h100-pcie) overnight — the [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) has the ceilings.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/fine-tune-llm-qlora · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/serve-llm-vllm -->

---
title: "How to deploy vLLM on a cloud GPU: an OpenAI-compatible endpoint"
description: "Deploy vLLM on a rented GPU, pick the right card for your model size, benchmark tokens per second and put a price on every million tokens."
url: https://powergpu.ai/guides/serve-llm-vllm
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Hands-on walkthrough

# How to deploy vLLM on a cloud GPU: an OpenAI-compatible endpoint

Deploy vLLM on a rented GPU, pick the right card for your model size, benchmark tokens per second and put a price on every million tokens.

11 min read Published 2026-08-25 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- On PowerGPU one vLLM template instance gives an OpenAI-compatible endpoint; 8B on an RTX 5090 costs $0.439 per hour.
- Expect roughly 2,800 aggregate tokens per second for 8B on an RTX 5090, 900 for 32B 4-bit, 550 for 70B 4-bit.
- Cost per million output tokens equals hourly rate divided by tokens per second times 3,600, times one million; reserved cuts 35 percent.
- The commonest OOM is leaving max-model-len at the model's 128k maximum; cap context and use an fp8 KV-cache.

## Deploy in one command

*zero to endpoint*

```
powergpu launch --gpu rtx-5090 --template vllm --disk 60 \
  --env MODEL=meta-llama/Llama-3.1-8B-Instruct
# ✓ i-52ab77c1 · https://i-52ab77c1.powergpu.ai:8000/v1

curl https://i-52ab77c1.powergpu.ai:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"meta-llama/Llama-3.1-8B-Instruct",
       "messages":[{"role":"user","content":"ping"}]}'
```

The template downloads weights on first boot (billed bandwidth: a 16 GB model costs $0.16). Put HF_HOME on a [volume](https://powergpu.ai/docs/volumes) and every future instance skips the download.

## Pick the card for the model

| Model class | Card | $/hr | Aggregate tok/s ≈ | $ / M output tokens ≈ |
| --- | --- | --- | --- | --- |
| 7–8B FP16 | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | $0.439 | 2,800 | $0.044 |
| 14B FP16 / 32B-4bit | [L40S](https://powergpu.ai/gpu/l40s) | $0.514 | 900 | $0.159 |
| 70B-4bit | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | $1.867 | 550 | $0.94 |

Throughputs are typical continuous-batching aggregates at moderate context; your prompt mix will move them ±40%. That is why the benchmark section exists.

## The flags that matter

*the 5 flags that do 95% of the work*

```
--max-model-len 8192          # cap context = cap KV-cache = predictable memory
--gpu-memory-utilization 0.92 # default 0.90; raise carefully once stable
--max-num-batched-tokens 8192 # throughput/latency dial — higher = more tok/s
--kv-cache-dtype fp8          # halves cache on Hopper/Blackwell — free VRAM
--tensor-parallel-size 2      # split across 2 GPUs when one is short
```

Everything else can stay default until the benchmark says otherwise. The single most common OOM cause is leaving --max-model-len at a model's 128k maximum "just in case" — the [KV-cache section](https://powergpu.ai/guides/llm-vram-requirements) shows what that costs.

## Benchmark before you believe

*10 minutes, real numbers*

```
vllm bench serve \
  --base-url https://i-52ab77c1.powergpu.ai:8000 \
  --model meta-llama/Llama-3.1-8B-Instruct \
  --dataset-name random --random-input-len 512 --random-output-len 256 \
  --request-rate 16 --num-prompts 512
# → throughput, TTFT p50/p99, ITL p50/p99
```

Judge two numbers: aggregate tok/s (your cost) and p99 time-to-first-token (your users). Raise --request-rate until p99 TTFT crosses your budget — that rate is the card's honest capacity for your traffic shape.

## Price your tokens

*the only pricing formula you need*

```
$/M output tokens = price_per_hour / (tokens_per_sec x 3600) x 1e6

# benchmark said 2,910 tok/s on the 5090:
0.439 / (2910 x 3600) x 1e6 = $0.042 per million output tokens
# reserved (-35%) and cache-hit traffic push it lower still
```

## Production posture

- **Baseline on [reserved](https://powergpu.ai/products/reserved)** (−35%), burst on [serverless](https://powergpu.ai/products/serverless) — one balance, two curves.
- **Health-check** GET /health; restart-on-unhealthy is a template toggle.
- **Weights on a read-only volume** — replacement workers boot in seconds, and a bad deploy can never corrupt the library.
- **Never serve on interruptible** — that mode is for the [training side](https://powergpu.ai/use-cases/fine-tuning) of your pipeline.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/serve-llm-vllm · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/cheapest-gpu-for-stable-diffusion -->

---
title: "Cheapest cloud GPU for Stable Diffusion & Flux in 2026 ($/image)"
description: "What SDXL and Flux actually need, images-per-dollar on eight rentable GPUs, and where paying more per hour costs less per image."
url: https://powergpu.ai/guides/cheapest-gpu-for-stable-diffusion
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Costs & pricing

# Cheapest cloud GPU for Stable Diffusion & Flux in 2026 ($/image)

What SDXL and Flux actually need, images-per-dollar on eight rentable GPUs, and where paying more per hour costs less per image.

9 min read Published 2026-08-27 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- For SDXL the RTX 4090 wins images per dollar at $0.163 interruptible, roughly 2.9 seconds per image; for Flux the RTX 5090 leads.
- Rank cards by images per dollar, not hourly rate: 3,600 divided by seconds per image divided by hourly price.
- Flux dev FP8 needs about 17 GB, so 24 GB cards are the practical floor and 32 GB removes the ceiling.
- On PowerGPU cheap hourly cards like the RTX 3060 lose on throughput; keep checkpoints on a volume and stop idle instances.

## The answer table

Eight rentable cards, live prices, images per dollar (interruptible rate — image queues pause gracefully):

| GPU | $/hr int. | SDXL s/img | SDXL img/$ | Flux s/img | Flux img/$ |
| --- | --- | --- | --- | --- | --- |
| [RTX 3060](https://powergpu.ai/gpu/rtx-3060) (12 GB) | $0.021 | 11.0 s | **15,584** | 38 s | **4,511** |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) (24 GB) | $0.054 | 5.6 s | **11,905** | 16 s | **4,167** |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) (12 GB) | $0.037 | 6.5 s | **14,969** | 21 s | **4,633** |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (24 GB) | $0.163 | 2.9 s | **7,616** | 7.5 s | **2,945** |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (32 GB) | $0.219 | 2.1 s | **7,828** | 5.2 s | **3,161** |
| [RTX A4000](https://powergpu.ai/gpu/rtx-a4000) (16 GB) | $0.035 | 8.9 s | **11,557** | 29 s | **3,547** |
| [L40S](https://powergpu.ai/gpu/l40s) (48 GB) | $0.257 | 3.4 s | **4,120** | 9.0 s | **1,556** |
| [RTX 5070 Ti](https://powergpu.ai/gpu/rtx-5070-ti) (16 GB) | $0.065 | 4.8 s | **11,538** | 15 s | **3,692** |

Timings: community-typical SDXL 1024² @30 steps and Flux dev @20 steps; treat as ±20% and re-run on your workflow. Prices re-render live with every weekly sheet update.

## Method: $/image, not $/hour

*the whole method*

```
images_per_dollar = 3600 / seconds_per_image / price_per_hour
```

This single division reorders the whole market. The "cheap" hourly cards at the top of the table lose to the [RTX 4090](https://powergpu.ai/gpu/rtx-4090) the moment throughput enters the equation — and the gap widens on Flux, where small cards spill to system RAM.

## SDXL economics

- **Interactive sessions** — an evening of prompting (~3 h) on a 4090: $0.98 on-demand. Use on-demand for sessions: an interruption mid-flow is worth more than the 50%.
- **Batches** — queue overnight on interruptible; per the table, roughly 7,616 images per dollar.
- **Budget floor** — the 3090 remains the best "always cheap" card: 24 GB means no workflow ever refuses to load.

## Flux economics

- **24 GB is the entry ticket** (FP8 + offloaded text encoders). Below that, generation works but throughput collapses — the img/$ column shows the 3060's honest number.
- **32 GB removes the ceiling** — full-precision weights, LoRA stacks and upscalers resident: the [5090](https://powergpu.ai/gpu/rtx-5090) leads img/$ despite the highest hourly rate on the table.
- **Serving users?** Concurrency changes the criteria — see [serverless endpoints](https://powergpu.ai/products/serverless) with the ComfyUI template.

## Three traps that triple bills

1. **Re-downloading models every session.** 30 GB of checkpoints at every boot is slow and silly. A 120 GB [volume](https://powergpu.ai/docs/volumes) costs $9.60/mo and mounts in seconds.
2. **Idle instances "kept for later".** Stop them — a stopped instance bills only disk. The pause button is the biggest discount on this page.
3. **Benchmarking with someone else's workflow.** Your LoRA count, resolution and sampler move s/image by 3×. Rent two candidates for one hour ($0.766 total) and measure your own pipeline.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/cheapest-gpu-for-stable-diffusion · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/blender-cloud-rendering -->

---
title: "Blender cloud rendering on GPUs: setup, cost per frame, pitfalls"
description: "Render Cycles scenes on rented RTX hardware: headless setup, per-frame cost math, and the mistakes that quietly triple a render bill."
url: https://powergpu.ai/guides/blender-cloud-rendering
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Hands-on walkthrough

# Blender cloud rendering on GPUs: setup, cost per frame, pitfalls

Render Cycles scenes on rented RTX hardware: headless setup, per-frame cost math, and the mistakes that quietly triple a render bill.

10 min read Published 2026-09-01 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Render cost is seconds per frame times frames divided by 3,600, times the rate: 250 frames at 3m41s equals 15.4 GPU-hours.
- On PowerGPU that shot runs at $0.163 per hour on interruptible RTX 4090s, and eight workers cost the same total as one.
- Time one frame before sizing the farm, split frames by stride across workers, and always render EXR rather than PNG.
- Interruptible suits rendering because frames are idempotent: an interruption loses only the frame in flight, at half price.

## Why per-second billing fits rendering

Render load is the least steady workload in computing — nothing for days, then 400 GPU-hours before a deadline. Owning hardware for that peak means paying for idle; classic farms price the peak in. Per-second rental inverts it: 8 instances × 1 hour costs exactly 1 instance × 8 hours ($1.304 on interruptible 4090s), so parallelism is free and idle costs zero.

## Setup: template + volumes

*stage & smoke-test*

```
powergpu volume create --name scene --size 60 --region eu-west-1
powergpu volume create --name frames --size 100 --region eu-west-1

# stage the packed project once (any instance, or the console uploader)
powergpu launch --gpu rtx-4090 --template blender --volume scene:/scene
scp shot.blend root@i-…:/scene/

# test ONE frame before the farm — always
blender -b /scene/shot.blend -o /frames/f#### -F OPEN_EXR -f 40
# Fra:40 … Time: 03:41.20 (Saved: /frames/f0040.exr)
```

That single timed frame is your whole cost model: *seconds-per-frame × frames × rate ÷ 3600*. Never size a farm without it.

## Turning instances into a farm

*an 8-worker farm in one loop*

```
# 8 workers, frame ranges split by stride — no scheduler needed
for i in $(seq 0 7); do
  powergpu launch --gpu rtx-4090 --type interruptible \
    --template blender --volume scene:/scene:ro --volume frames:/frames \
    --env CMD="blender -b /scene/shot.blend -o /frames/f#### \
               -F OPEN_EXR -s $((1+i)) -e 250 -j 8 -a"
done
# each worker renders frames i, i+8, i+16, … — an interruption loses
# at most the frame in flight; -j strides make re-runs idempotent
```

Interruptible is the right mode *because* frames are idempotent queue items: a paused worker resumes its stride and re-renders one frame at half price. The [Python SDK](https://powergpu.ai/sdk) version of this loop (with auto-retry) is twelve lines.

## Costing a shot honestly

- **250 frames × 3m41s**: ≈ 15.4 GPU-hours on RTX 4090 · $2.51 interruptible
- **Wall-clock with 8 workers**: ~1 h 55 m · same total
- **Volumes 160 GB, one week**: scene + frames · $2.99
- **EXR download 30 GB**: once, on approval · $0.30

Scenes over 24 GB (volumes, fur, 8K textures) move to the 48–96 GB workstation cards — the [rendering playbook](https://powergpu.ai/use-cases/rendering) ranks them; the [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) at $1.040/hr holds production scenes whole.

## The pitfalls that triple render bills

1. **Rendering PNG instead of EXR to "save space"** — then re-rendering for the grade. Storage is $0.08/GB/mo; re-renders are GPU-hours. Always EXR.
2. **Unpacked assets** — workers render magenta placeholders for an hour before anyone looks. Pack resources, and eyeball frame 1 from every worker before walking away.
3. **Per-frame instance churn** — booting an instance per frame pays the 30-second boot 250 times. Workers take strides; boot once per worker.
4. **Forgetting the farm** — -a exits when the stride finishes, but the instance keeps billing its disk until destroyed. End every farm script with destroy (or use the SDK context manager that cannot forget).

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/blender-cloud-rendering · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/h100-rental-cost-per-hour -->

---
title: "How much does it cost to rent an H100? Per-hour math for 2026"
description: "H100 SXM, PCIe and NVL rental prices per hour and per month, what marketplaces and hyperscalers charge, the rent-vs-buy break-even, and five ways to pay less."
url: https://powergpu.ai/guides/h100-rental-cost-per-hour
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Costs & pricing

# How much does it cost to rent an H100? Per-hour math for 2026

H100 SXM, PCIe and NVL rental prices per hour and per month, what marketplaces and hyperscalers charge, the rent-vs-buy break-even, and five ways to pay less.

9 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- An H100 SXM rents for $1.428 per hour on-demand on PowerGPU, 30% below the $2.04 public marketplace median.
- The same card is $0.714 interruptible and $0.928 reserved; a full month of on-demand use is about $1,042.
- Elsewhere the same GPU-hour lists at $4-8 on hyperscalers and $2-4 on specialist clouds, a 3-5× spread for identical silicon.
- A bare H100 costs roughly $25,000-30,000, so buying only wins after years of near-100 percent utilisation, hosting and power excluded.

## Today's H100 prices, all three modes

There are three H100 packages on the sheet and one rule behind every number: the public marketplace median for that package × 0.70, rounded down, re-checked weekly (snapshot 2026-09-14). Live figures:

| Package | Memory | On-demand | Interruptible | Reserved (3 mo) | Market median |
| --- | --- | --- | --- | --- | --- |
| [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 80 GB HBM3 | **$1.428** | $0.714 | $0.928 | $2.04 |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB HBM2e | **$1.867** | $0.933 | $1.213 | $2.67 |
| [H100 NVL](https://powergpu.ai/gpu/h100-nvl) | 80 GB HBM3 | **$1.811** | $0.905 | $1.177 | $2.59 |

Two things surprise people. First, the **SXM** part — the one with NVLink and the fastest memory — is the cheapest of the three here, because its marketplace pool is by far the deepest and the median follows supply. Second, **interruptible** is not a fluctuating spot auction: it is a flat 50% of on-demand, on every H100, all the time.

## What the rest of the market charges

Public list prices for a single H100 SXM GPU-hour, September 3, 2026, grouped by provider type. Ranges, because every provider packages the card differently (regions, node sizes, minimum commitments):

| Provider type | Typical H100 $/GPU-hr | What moves the number |
| --- | --- | --- |
| Hyperscalers (list) | $4 – $8 | 8-GPU node minimums, region, committed-use discounts |
| Specialist GPU clouds | $2 – $4 | Secure vs community tiers, PCIe vs SXM, reservations |
| Public GPU marketplaces | $1.5 – $3 (median $2.04) | Per-host auctions; changes hourly, reliability varies by host |
| **PowerGPU** | **$1.428** fixed | Median × 0.70, verified datacenters, re-checked weekly |

The gap between the top and bottom rows is 3–5× for the same silicon. Most of it is not margin: it is the cost of sales teams, card fraud, free tiers and idle capacity that a fixed-price, crypto-settled operator does not carry.

## Per day, per month, per training run

Per-second billing makes the hourly rate the only number you need, but here are the shapes people actually budget:

| Scenario | Math | On-demand | Interruptible |
| --- | --- | --- | --- |
| One H100 SXM, one day | 24 h | $34.27 | $17.14 |
| One H100 SXM, one month | 730 h | $1,042 | $521 |
| 8× H100 node, one month | 8 × 730 h | $8,340 | $4,170 |
| A 72-hour fine-tune on 8× H100 | 8 × 72 h | $823 | $411 |
| Reserved H100 SXM, 3-month term | 3 × 730 h × $0.928 | $2,032 total |  |

Run your own schedule through the [cost calculator](https://powergpu.ai/calculator) — it adds storage and bandwidth and compares the result with the marketplace median.

## The costs that are not the GPU

- **Storage** — $0.08/GB/month here, per second, while the disk or volume exists. A 1 TB checkpoint volume is $80/month; on hyperscalers, $80–170.
- **Egress** — $0.01/GB flat, both directions. Hyperscalers charge $0.05–0.12/GB out: pulling 10 TB of results costs $100 here versus $500–1,200 there.
- **Idle time** — the biggest hidden line everywhere. Per-second billing and volumes let you destroy the GPU between runs and keep the data; hourly rounding and "minimum 1 hour" policies do not.
- **Node minimums** — many H100 offers are 8-GPU nodes only. If you need one card, you pay for eight. Here 1×, 2×, 4× and 8× carry the same per-GPU price.

## Rent or buy: the break-even

A bare H100 SXM sells for roughly $27,000 in 2026, and it needs an HGX host, 700 W of power and cooling, a network port and someone to run it. Ignoring all of that, renting at $1.428/hr buys **18,908 GPU-hours** for the sticker price — about **26 months of 24/7 use**. Add hosting and a realistic 40–60% utilisation and the break-even moves past the card's useful life. Buying makes sense for permanent, saturated fleets with their own datacenter; for everyone else, a [reserved](https://powergpu.ai/products/reserved) H100 at $0.928/hr is the closest thing to owning one without the depreciation.

## Five ways to pay less for H100 time

1. **Go interruptible for anything that checkpoints.** Training does; it is half price and restarts from your volume. Keep on-demand for the deadline run and for serving.
2. **Reserve what runs every day.** Above ~65% utilisation, $0.928/hr beats every other mode, and capacity is held for you during launch-week droughts.
3. **Check whether an A100 is enough.** At $0.560/hr the A100 SXM4 still wins total job cost for ≤13B models and LoRA work — see [H100 vs A100](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4).
4. **Use FP8.** The Hopper Transformer Engine is the discount most teams never enable: roughly 1.5–2× throughput on the same hourly rate.
5. **Destroy, keep the volume.** Datasets and weights stay warm at $0.08/GB/mo; the GPU bills nothing until the next run.

Bigger than an H100? The [H100 vs H200 vs B200 guide](https://powergpu.ai/guides/h100-vs-h200-vs-b200) covers when the $2.791/hr H200 or a Blackwell card ends up cheaper per token.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/h100-rental-cost-per-hour · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/best-cloud-gpu-for-llm-inference -->

---
title: "Best cloud GPU for LLM inference in 2026, by model size | PowerGPU"
description: "From 8B to 405B: the cheapest rentable card that holds each model, indicative tokens per second with vLLM, and the cost per million tokens on today's sheet."
url: https://powergpu.ai/guides/best-cloud-gpu-for-llm-inference
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Choosing hardware

# Best cloud GPU for LLM inference in 2026, by model size

From 8B to 405B: the cheapest rentable card that holds each model, indicative tokens per second with vLLM, and the cost per million tokens on today's sheet.

10 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Match VRAM to the model, then buy bandwidth: RTX 5090 for 7-14B, L40S or A6000 for 32B, H100 or H200 for 70B and up.
- Budget about 2.4 GB per billion parameters in FP16 and 0.62 GB in 4-bit, then add KV-cache for context times concurrency.
- On PowerGPU an RTX 5090 at $0.439 sustains roughly 2,800 tokens per second on 8B; interruptible capacity halves the cost per token.
- Cost per million tokens is hourly rate divided by tokens per second times 3,600, times one million; measure it at real concurrency.

## Three rules before any benchmark

1. **VRAM decides what runs.** Weights plus KV-cache must fit, with headroom: ~2.4 GB per billion parameters in FP16, ~0.62 GB in 4-bit, then add the cache for your context × concurrency. The [VRAM guide](https://powergpu.ai/guides/llm-vram-requirements) has the full tables.
2. **Bandwidth decides how fast.** Decoding is memory-bound: tokens per second track GB/s more than TFLOPS. That is why a 1.79 TB/s RTX 5090 beats a 1 TB/s RTX 4090 by more than its FLOPS suggest, and why HBM cards dominate 70B serving.
3. **Batch decides the bill.** Continuous batching (vLLM, SGLang, TGI) turns one card into a dozen concurrent streams. Cost per million tokens is only meaningful at the concurrency you will actually run.

## The table: model size → card → $ per million tokens

Indicative aggregate throughput under vLLM continuous batching (roughly 1k tokens in, 1k out, batch 32–64) and the resulting cost at today's on-demand rates — snapshot 2026-09-14. Treat the tokens/s column as order-of-magnitude community numbers and benchmark your own model; the price column is exact.

| Model | Precision / weights | Card | On-demand | ~tokens/s | $ / M tokens | Note |
| --- | --- | --- | --- | --- | --- | --- |
| Llama 3.1 8B | FP16, 16 GB | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (24 GB) | $0.327 | ~2,200 | **$0.041** | The default: fits with KV-cache room on 24 GB. |
| Llama 3.1 8B | FP16, 16 GB | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (32 GB) | $0.439 | ~2,800 | **$0.044** | 78% more bandwidth than the 4090 shows up directly in tokens/s. |
| Llama 3.1 8B | 4-bit, ~6 GB | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) (12 GB) | $0.042 | ~900 | **$0.013** | The cheapest working endpoint on the sheet. |
| Qwen 2.5 14B | FP16, 28 GB | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (32 GB) | $0.439 | ~1,500 | **$0.081** | 32 GB holds FP16 14B with modest context. |
| Qwen 2.5 32B | 4-bit, ~20 GB | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (24 GB) | $0.327 | ~700 | **$0.130** | AWQ/GPTQ 4-bit on 24 GB; quality within a point of FP16. |
| Qwen 2.5 32B | FP16, 64 GB | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (80 GB) | $1.867 | ~2,400 | **$0.216** | FP8 on Hopper roughly doubles this again. |
| Llama 3.1 70B | 4-bit, ~40 GB | [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) (48 GB) | $0.281 | ~350 | **$0.223** | Cheapest single card that holds a 70B. |
| Llama 3.1 70B | 4-bit, ~40 GB | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (80 GB) | $1.867 | ~1,200 | **$0.432** | One 80 GB card, real batch sizes, FP8 KV-cache. |
| Llama 3.1 70B | FP8, 70 GB | [H200](https://powergpu.ai/gpu/h200) (141 GB) | $2.791 | ~2,000 | **$0.388** | 141 GB leaves room for long contexts and big batches. |
| Llama 3.1 405B | FP8, ~405 GB | [H200](https://powergpu.ai/gpu/h200) (141 GB) | $2.791 | ~600 | **$1.292** | Tensor-parallel across 4× H200 — figure is per GPU. |

Interruptible capacity halves every number in the price column. Formula, if you want to redo it with your own measurement: $/M = $/hr ÷ (tok/s × 3600) × 1,000,000.

## 7B–14B: consumer cards win

Below 16 GB of weights nothing beats GeForce silicon per dollar. The [RTX 5090](https://powergpu.ai/gpu/rtx-5090) ($0.439/hr, 32 GB GDDR7) is the current king of tokens per dollar for 8B FP16 and 14B models; the [RTX 4090](https://powergpu.ai/gpu/rtx-4090) ($0.327/hr) is a hair slower and cheaper; the [RTX 3090](https://powergpu.ai/gpu/rtx-3090) ($0.108/hr) is the budget pick when throughput matters less than the hourly rate. For always-on endpoints that must survive months of 24/7 duty, the datacenter [L4](https://powergpu.ai/gpu/l4) ($0.225/hr, 72 W) and [L40S](https://powergpu.ai/gpu/l40s) trade a little speed for passive cooling and ECC.

## 24B–32B: the 48 GB middle

A 32B model is ~64 GB in FP16 and ~20 GB in 4-bit. Quantized, it runs on a 24 GB card with short contexts; for serious concurrency you want 48 GB. The [L40S](https://powergpu.ai/gpu/l40s) ($0.514/hr) is the datacenter answer — FP8, 864 GB/s, built for 24/7 — and the [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) ($0.281/hr) the cheapest 48 GB on the sheet. Compare them in [RTX 5090 vs L40S](https://powergpu.ai/compare/rtx-5090-vs-l40s) and [RTX 6000 Ada vs A6000](https://powergpu.ai/compare/rtx-6000ada-vs-rtx-a6000).

## 70B and up: HBM or nothing

Seventy billion parameters is ~40 GB in 4-bit and ~140 GB in FP16. One 80 GB card serves the quantized model with real batch sizes: the [H100 PCIe](https://powergpu.ai/gpu/h100-pcie) ($1.867/hr) with FP8 KV-cache, or the [A100 PCIe](https://powergpu.ai/gpu/a100-pcie) ($0.374/hr) when budget wins. Full-precision 70B, 100B+ models and long-context serving belong on the [H200](https://powergpu.ai/gpu/h200) ($2.791/hr, 141 GB, 4.8 TB/s) or tensor-parallel across 2–8× H100 SXM on one machine — the per-GPU price does not change with count. Frontier-scale inference with FP4 is where the [B200](https://powergpu.ai/gpu/b200) ($5.425/hr) earns its rate: several times the H100's tokens per second at under 3× the price.

## Serving patterns that change the bill

- **Quantize first.** AWQ/GPTQ 4-bit halves the card class you need and rarely hurts chat quality. FP8 on Hopper/Ada/Blackwell is nearly free quality-wise and 1.5–2× faster.
- **Reserve the baseline, burst serverless.** A [reserved](https://powergpu.ai/products/reserved) card at −35% carries steady traffic; [serverless](https://powergpu.ai/products/serverless) workers absorb spikes and scale to zero.
- **Mount the model library read-only.** New workers skip the 40 GB download; cold starts become load-to-VRAM.
- **Prefix caching for agents.** If prompts share a long system prefix, [SGLang](https://powergpu.ai/templates/sglang) can double effective throughput over plain vLLM.
- **Measure, then commit.** Ten minutes on two candidate cards costs cents; the [vLLM guide](https://powergpu.ai/guides/serve-llm-vllm) ships the exact benchmark command.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/best-cloud-gpu-for-llm-inference · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/run-comfyui-on-a-cloud-gpu -->

---
title: "How to run ComfyUI on a cloud GPU: setup, models, cost per image"
description: "Deploy ComfyUI on a rented RTX 4090 or 5090 in 30 seconds, keep checkpoints on a volume, run workflows headless through the API, and know what each image costs."
url: https://powergpu.ai/guides/run-comfyui-on-a-cloud-gpu
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Hands-on walkthrough

# How to run ComfyUI on a cloud GPU: setup, models, cost per image

Deploy ComfyUI on a rented RTX 4090 or 5090 in 30 seconds, keep checkpoints on a volume, run workflows headless through the API, and know what each image costs.

9 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Rent a 24 GB RTX 4090 at $0.327 per hour on PowerGPU: it runs SDXL, Flux dev FP8 and most video nodes.
- Rule of thumb: SDXL wants about 8 GB, Flux dev FP8 about 17 GB, video models 24 GB with offloading and 48 GB without.
- An SDXL image takes about 1.1 seconds and a Flux dev image 2.2 seconds on a 4090, a fraction of a cent each.
- Put checkpoints and custom_nodes on a volume so they survive between sessions; re-downloading models is the main waste of GPU-hours.

## Why rent instead of buying a 4090

ComfyUI is bursty by nature: an evening of prompting, a batch of a thousand images, then nothing for a week. A rented card bills only the seconds the queue is running — $0.327/hr for an [RTX 4090](https://powergpu.ai/gpu/rtx-4090) on-demand, $0.163/hr interruptible — so a heavy month rarely reaches a fraction of the card's retail price, and you can jump to a 32 GB or 96 GB card the day a workflow needs it. No driver upgrades, no 450 W heater under the desk.

## Pick the card: 24, 32 or 48 GB

| Card | VRAM | On-demand | Interruptible | Best for |
| --- | --- | --- | --- | --- |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) | 24 GB | $0.108 | $0.054 | SDXL batch farms, SD 1.5, LoRA training on a budget |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | 24 GB | $0.327 | $0.163 | The default: Flux dev FP8, SDXL + ControlNet stacks, most video nodes |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) | 32 GB | $0.439 | $0.219 | Full-precision Flux, Wan/Hunyuan video, heavy upscale chains |
| [L40S](https://powergpu.ai/gpu/l40s) | 48 GB | $0.514 | $0.257 | Serving images to users 24/7, several workflows loaded at once |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) | 96 GB | $1.040 | $0.520 | Video diffusion without offloading, giant batches |

Rule of thumb: Flux dev in FP8 wants ~17 GB, SDXL ~8 GB, current video models 24 GB with offloading and 48 GB without. The [cheapest-GPU guide](https://powergpu.ai/guides/cheapest-gpu-for-stable-diffusion) ranks the cards by images per dollar.

## Deploy the template (30 seconds)

1. Open the [console](https://cloud.powergpu.ai/?gpu=rtx-4090), filter on the card, pick a machine.
2. Choose [ComfyUI](https://powergpu.ai/templates/comfyui) in the template picker. It ships ComfyUI-Manager, exposes the UI on a TLS-terminated port and mounts /workspace/ComfyUI/models.
3. Set the disk (60 GB is plenty for the OS and cache) and attach a volume for models — next section.
4. Deploy. The instance is running in about 30 seconds; click the port link and the node graph is there.

*same thing from the CLI*

```
powergpu launch --gpu rtx-4090 --template comfyui \
    --disk 60 --volume sdmodels:/workspace/ComfyUI/models
# ✓ instance i-3c91ab04 running (28.6s)
# ✓ https://i-3c91ab04.powergpu.ai:8188  (ComfyUI)
```

## Models on a volume, once

Checkpoints are 2–12 GB each and downloading them every session is the most common way to waste GPU-hours. Create a [volume](https://powergpu.ai/products/volumes) the first time, mount it on the models folder, fill it once — Hugging Face and Civitai downloads run at multi-Gbps from the datacenter — and every future instance in the region starts with the library present. A 120 GB library costs $9.60/month. Put custom_nodes on the same volume so installed nodes persist too.

## Headless: workflows as an API

Enable dev mode in ComfyUI settings, then "Save (API format)" on any workflow. The JSON you get is a request body: POST it to /prompt, poll /history/<id>, fetch the images from /view. That loop is how batch pipelines run a thousand prompts overnight on [interruptible](https://powergpu.ai/products/interruptible) capacity, and how a [serverless endpoint](https://powergpu.ai/products/serverless) serves the same workflow behind autoscaling workers.

*queue a workflow, read the result*

```
curl -X POST https://i-3c91ab04.powergpu.ai:8188/prompt \
  -H 'Content-Type: application/json' \
  -d @workflow_api.json
# {"prompt_id": "a1c2…", "number": 12}
curl https://i-3c91ab04.powergpu.ai:8188/history/a1c2…
```

## What an image costs, honestly

| Job | Card · mode | Time | Cost |
| --- | --- | --- | --- |
| One SDXL image, 1024², 25 steps | RTX 4090 on-demand | ~1.1 s | $0.00010 |
| One Flux dev image, 1024², 20 steps | RTX 4090 on-demand | ~2.2 s | $0.00020 |
| An evening of prompting | RTX 4090 on-demand | 3 h | $0.98 |
| 10,000 Flux images, batch | RTX 4090 interruptible | ~6.1 h | $1.00 |
| Model library, always warm | 120 GB volume | 1 month | $9.60 |

Roughly **5,004 Flux images per dollar** on-demand, 10,008 interruptible. Speeds are typical for a tuned 4090; your workflow, resolution and step count move them.

## Five traps that waste GPU-hours

- **Downloading models every session** — the volume above fixes it permanently.
- **Leaving the instance running overnight** — stop it; per-second billing means an idle GPU is pure waste. The disk survives stop/start.
- **Running batches on-demand** — queues are interruptible by nature: each prompt is a restartable item. Pay half.
- **Picking a card by hourly price alone** — a 5090 finishes Flux batches faster than its price gap over a 4090; images per dollar is the metric.
- **Forgetting the outputs** — write results to the volume or download them before destroy; the instance disk dies with the instance.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/run-comfyui-on-a-cloud-gpu · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/rent-gpu-with-crypto-no-kyc -->

---
title: "How to rent a GPU with crypto and no KYC (2026) | PowerGPU"
description: "Renting cloud GPUs with USDT, Bitcoin or Monero and no identity check: how top-ups work, which coin to use, what data is kept, and the step-by-step deploy."
url: https://powergpu.ai/guides/rent-gpu-with-crypto-no-kyc
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Hands-on walkthrough

# How to rent a GPU with crypto and no KYC (2026)

Renting cloud GPUs with USDT, Bitcoin or Monero and no identity check: how top-ups work, which coin to use, what data is kept, and the step-by-step deploy.

7 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- PowerGPU rents GPUs with no KYC: an email address and a password create the account, and payment is crypto only.
- Accepted coins are USDT on TRC-20 and ERC-20, Bitcoin, Monero, Litecoin, Ethereum, TRON and Solana; TRC-20 credits in seconds.
- The minimum top-up is $40 and credits never expire; interruptible RTX 4090 time runs at $0.163 per hour.
- No name, ID document, phone or IP address is stored; the acceptable use policy still bans attacks, illegal content and mining.

## Why crypto-only changes what a GPU cloud can be

Most GPU clouds take cards, which means a payment processor, chargebacks, fraud scoring and — increasingly — identity checks before you can deploy anything. Settling in crypto removes that whole layer: the deposit is final, there is no card data to protect and no reason to collect a name or an address. It is why PowerGPU can run with an email and a password only, keep no IP addresses, and still price every GPU [30% under the market median](https://powergpu.ai/pricing): card fraud and KYC tooling are costs the other providers pass on to you.

## Which coin to use

| Coin | Network | Credited after | Typical fee | Pick it when |
| --- | --- | --- | --- | --- |
| **USDT** | TRC-20 (TRON) | seconds | < $1 | Everyday top-ups; the fastest and cheapest path |
| **USDT** | ERC-20 (Ethereum) | ~1–3 min | $1–5 | Your funds already sit on Ethereum; browser wallets (EIP-6963) supported |
| **Monero** | XMR | ~2–20 min | cents | Privacy is the point: amounts and parties hidden on-chain |
| **Solana / TRON** | SOL, TRX | seconds | cents | Fast native-coin payments without a stablecoin |
| **Bitcoin / Litecoin** | BTC, LTC | ~10–30 min | varies | You hold BTC/LTC and are not in a hurry |
| **Ethereum** | ETH | ~1–3 min | $1–5 | Native ETH, wallet deep links (EIP-681) supported |

Whatever you send, the balance is kept in **USD**: usage draws it down per second at the fixed sheet prices, so a top-up made in Monero buys exactly the same GPU-hours as one made in USDT.

## From wallet to running GPU in five minutes

1. **Create the account** — email + password in the [console](https://cloud.powergpu.ai/?auth=signup). No verification email is even sent; the address is only used for password resets and ticket replies.
2. **Top up** — Billing → choose an amount ($30 minimum) and a coin. You get a unique address and a QR code; send the exact amount. Underpaid? The page shows the remainder. Overpaid? The excess is credited.
3. **Wait for confirmation** — seconds on TRON/Solana, minutes on Ethereum, longer on Bitcoin. The page polls itself; credits appear the moment the network confirms.
4. **Deploy** — pick a GPU (an [RTX 4090](https://powergpu.ai/gpu/rtx-4090) is $0.327/hr, an [H100 SXM](https://powergpu.ai/gpu/h100-sxm) $1.428/hr), a template, deploy. Running in about 30 seconds.
5. **Stop when done** — billing ends that second. Credits never expire; unused balances are refundable in the original coin on request, minus network fees.

*the same flow from the CLI, once you hold an API key*

```
$ powergpu whoami
# account ok · balance $41.22 · 0 instances running
$ powergpu launch --gpu rtx-4090 --template ollama
# ✓ instance i-52ab77c1 running (24.1s) · $0.327/hr
```

## What is stored, what is not

| Kept | Never collected |
| --- | --- |
| Email address (for resets and ticket replies) | Name, postal address, phone, ID documents |
| Argon2id password hash | Card numbers or bank details — there are none |
| USD balance and a per-second usage ledger | IP addresses — not in logs, not in the database |
| Instance metadata (model, region, timestamps) and support tickets | Third-party trackers, analytics scripts, ad pixels |
| Deposit references (tx hash, coin, amount) for your own receipts | Snapshots of your disks taken on our initiative |

The [privacy policy](https://powergpu.ai/legal/privacy) fits on one screen because the data does; the [security page](https://powergpu.ai/security) lists the isolation and encryption behind it. Instance disks are encrypted with per-instance keys, and destroy means the key is discarded — cryptographic erasure.

## The honest limits

- **No KYC is not no rules.** The [acceptable use policy](https://powergpu.ai/legal/aup) bans attacks, illegal content, spam infrastructure and mining. Privacy-first hosting is not anything-goes hosting.
- **Crypto is final.** Send to the right network — USDT on the wrong chain is recoverable only with a ticket and a tx hash, and not always.
- **New accounts have modest quotas** (instances, GPUs, top-ups per hour) that raise automatically with history — see [limits](https://powergpu.ai/docs/limits).
- **Enterprise is crypto too.** Fleets and clusters settle the same way, with scheduled instalments and per-team sub-accounts — [details](https://powergpu.ai/enterprise).

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/rent-gpu-with-crypto-no-kyc · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/a100-vs-h100-for-fine-tuning -->

---
title: "A100 vs H100 for fine-tuning (2026): cost per run, not per hour"
description: "Same 80 GB, 2.7× the hourly price: when an H100 finishes a LoRA, QLoRA or full fine-tune fast enough to beat the A100 on total cost — live prices, break-even rule."
url: https://powergpu.ai/guides/a100-vs-h100-for-fine-tuning
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Choosing hardware

# A100 vs H100 for fine-tuning (2026): cost per run, not per hour

Same 80 GB, 2.7× the hourly price: when an H100 finishes a LoRA, QLoRA or full fine-tune fast enough to beat the A100 on total cost — live prices, break-even rule.

8 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Rent the H100 only when it finishes the job faster than the price ratio between the two cards; otherwise the A100 is cheaper per run.
- On PowerGPU that ratio is $1.428 against $0.560 per hour, while BF16 LoRA and QLoRA runs see only 1.7-2.2×.
- Both cards hold 80 GB, so fit never decides; the H100 adds 3.35 versus 2.0 TB/s of bandwidth and FP8.
- FP8 full fine-tunes see 2.5-3× and tip the decision to the H100; without FP8 enabled you pay for bandwidth alone.

## The break-even rule

Both cards hold 80 GB, so the choice is never about fit — it is about dollars per finished run. Today the [H100 SXM](https://powergpu.ai/gpu/h100-sxm) costs **$1.428/hr** and the [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) **$0.560/hr**: a ratio of **2.55×**. The rule that settles every job:

rent the H100 if speedup(H100 / A100) > 2.55, otherwise rent the A100

Speedups below that line make the A100 cheaper per run even though the H100 is faster; above it the H100 is both faster and cheaper. The ratio moves with the weekly price re-check — the number above is live.

## What actually differs

|  | A100 SXM4 80 GB | H100 SXM 80 GB | Why it matters for fine-tuning |
| --- | --- | --- | --- |
| Memory | 80 GB HBM2e | 80 GB HBM3 | Same capacity — identical model fit |
| Bandwidth | 2.0 TB/s | 3.35 TB/s | Optimizer steps and attention are bandwidth-bound: ~1.6× from this alone |
| BF16 tensor (dense) | 312 TFLOPS | 990 TFLOPS | Compute-bound layers scale; small batches rarely saturate either |
| FP8 | — | Transformer Engine | The H100's real weapon: 2.5–3× on full fine-tunes that use it |
| NVLink | 600 GB/s | 900 GB/s | Multi-GPU FSDP all-gathers; matters past 2× cards |
| Price today | $0.560/hr | $1.428/hr | Ratio 2.55× — the break-even line |

## Five fine-tuning jobs, costed

Indicative wall-clock on the A100 and the speedup most teams measure on an H100 for the same recipe. The cost columns use today's on-demand prices; interruptible capacity halves both — fine-tunes checkpoint, so use it.

| Job | Setup | A100 hours | H100 speedup | A100 cost | H100 cost | Cheaper |
| --- | --- | --- | --- | --- | --- | --- |
| QLoRA, Llama 3.1 8B, 10k pairs | BF16 compute, 4-bit base | 1.5 h | 1.8× | $0.84 | $1.19 | **A100** ((29% less)) |
| LoRA, Llama 3.1 70B (4-bit base) | single card, 48 GB used | 3 h | 2.0× | $1.68 | $2.14 | **A100** ((22% less)) |
| Full fine-tune, 8B, BF16 | 8× node, FSDP | 6 h | 2.2× | $26.88 | $31.16 | **A100** ((14% less)) |
| Full fine-tune, 8B, FP8 (Transformer Engine) | 8× node, FSDP | 6 h | 3.0× | $26.88 | $22.85 | **H100** ((15% less)) |
| DPO / RLHF pass, 8B | policy + reference | 2.5 h | 2.4× | $1.40 | $1.49 | **A100** ((6% less)) |

The pattern is consistent: adapter methods in BF16 favour the A100; anything that engages FP8 or runs long enough for bandwidth to dominate favours the H100. Wall-clock is the tie-breaker — an H100 run that costs 10% more but finishes a day earlier is usually the right call for a deadline.

## Decision table

| You are doing | Rent | Because |
| --- | --- | --- |
| QLoRA / LoRA on ≤13B, BF16 | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) or [A100 PCIe](https://powergpu.ai/gpu/a100-pcie) | Fits in 24–80 GB; speedups under the break-even line |
| LoRA on 70B (4-bit base), single card | [A100 SXM4](https://powergpu.ai/gpu/a100-sxm4) | Bandwidth-bound but 2× speedup stays under 2.6× |
| Full fine-tune, BF16, multi-GPU | A100 SXM4 unless deadline-bound | ≈2.2× speedup is close to the line — pick by wall-clock |
| Full fine-tune with FP8 (TE, NeMo, TorchAO) | [H100 SXM](https://powergpu.ai/gpu/h100-sxm) | 2.5–3× speedup beats the price ratio outright |
| Sequence lengths ≥32k or big micro-batches | H100 SXM (or [H200](https://powergpu.ai/gpu/h200)) | Memory bandwidth and, on the H200, 141 GB |

## Making either card cheaper

- **Go interruptible.** Every job above checkpoints; the flat −50% turns an A100 into $0.280/hr and an H100 into $0.714/hr.
- **Enable FP8 before renting an H100.** Without it you are paying for bandwidth only; with it the H100 earns its price.
- **Tune on 1×, train on 8×.** Debug the config on a single card at one-eighth the burn rate, then scale.
- **Datasets and checkpoints on a volume** at $0.08/GB/month — the trainer instance stays disposable.
- **Compare the neighbours.** [H100 SXM vs A100 SXM4](https://powergpu.ai/compare/h100-sxm-vs-a100-sxm4) and [the PCIe pair](https://powergpu.ai/compare/h100-pcie-vs-a100-pcie) keep live numbers; the [QLoRA walkthrough](https://powergpu.ai/guides/fine-tune-llm-qlora) has the exact axolotl config.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/a100-vs-h100-for-fine-tuning · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/cheapest-gpu-for-ollama -->

---
title: "Cheapest cloud GPU for Ollama (2026): 8B to 70B models, by the hour"
description: "Which rented card runs each Ollama model size at Q4, indicative tokens per second per card, and what an always-on private assistant costs per month."
url: https://powergpu.ai/guides/cheapest-gpu-for-ollama
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Costs & pricing

# Cheapest cloud GPU for Ollama (2026): 8B to 70B models, by the hour

Which rented card runs each Ollama model size at Q4, indicative tokens per second per card, and what an always-on private assistant costs per month.

8 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- An RTX 3060 12 GB is the cheapest card that runs Ollama well, covering every 7B-8B model and most 14B models at Q4.
- Ollama pulls 4-bit weights by default, so budget about 0.6 GB per billion parameters plus 1-3 GB for context.
- That puts 8B on 12 GB cards, 27B-32B on 24 GB, and 70B on a 48 GB RTX A6000 at $0.281 per hour on PowerGPU.
- Ollama is single-stream: expect about 120 tokens per second for 8B on an RTX 4090, and use vLLM when many users share the card.

## How much VRAM each Ollama model needs

Ollama pulls 4-bit (Q4_K_M) weights by default, so the rule of thumb is **about 0.6 GB per billion parameters, plus 1–3 GB for context**. That puts 8B models on 12 GB cards, 14B on 12–16 GB, 27B–32B on 24 GB, and 70B on 48 GB. Go up a class if you want 32k+ context or FP16 weights (:fp16 tags need 2 GB per billion).

## The table: model → cheapest card → speed → cost

Tokens per second are indicative single-stream numbers for Ollama with the whole model on the GPU (llama.cpp CUDA backend); your prompt length and quantization move them. Prices are today's on-demand rates; interruptible halves them.

| Ollama model | Weights | Card | On-demand | ~tok/s | $ / M tokens | Note |
| --- | --- | --- | --- | --- | --- | --- |
| llama3.1:8b | 4.9 GB | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) (12 GB) | $0.042 | ~35 | $0.33 | The cheapest card that runs 8B comfortably |
| llama3.1:8b | 4.9 GB | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (24 GB) | $0.327 | ~120 | $0.76 | Snappy chat; room for 32k context |
| qwen2.5:14b | 9.0 GB | [RTX 4070](https://powergpu.ai/gpu/rtx-4070) (12 GB) | $0.075 | ~45 | $0.46 | 12 GB is enough at Q4 |
| qwen2.5:14b | 9.0 GB | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (32 GB) | $0.439 | ~110 | $1.11 | 1.79 TB/s GDDR7 shows in tokens/s |
| gemma3:27b | 17 GB | [RTX 3090](https://powergpu.ai/gpu/rtx-3090) (24 GB) | $0.108 | ~25 | $1.20 | 24 GB with context headroom at the lowest rate |
| qwen2.5:32b | 20 GB | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (24 GB) | $0.327 | ~40 | $2.27 | Tight but fine on 24 GB with 8k context |
| qwen2.5:32b | 20 GB | [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (32 GB) | $0.439 | ~62 | $1.97 | 32 GB gives long-context breathing room |
| llama3.3:70b | 43 GB | [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) (48 GB) | $0.281 | ~15 | $5.20 | Cheapest single card that holds 70B |
| llama3.3:70b | 43 GB | [L40S](https://powergpu.ai/gpu/l40s) (48 GB) | $0.514 | ~20 | $7.14 | Datacenter card for an always-on assistant |
| llama3.3:70b | 43 GB | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (80 GB) | $1.867 | ~38 | $13.65 | HBM bandwidth — the fastest single-card 70B |

Single-stream $/M tokens looks expensive next to API providers because one chat at a time leaves most of the card idle — Ollama is a private-assistant tool, not a serving engine. For many users, [vLLM](https://powergpu.ai/templates/vllm) batches requests and drops the cost per token by 10–30×.

## An always-on private assistant, per month

| Assistant | Card | 24/7 on-demand | Reserved (3 mo) | 8 h/day, workdays |
| --- | --- | --- | --- | --- |
| 8B chat (llama3.1:8b) | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | $31 | $20 | $7 |
| 14B–32B (qwen2.5:32b) | [RTX 4090](https://powergpu.ai/gpu/rtx-4090) | $239 | $155 | $58 |
| 70B (llama3.3:70b) | [RTX A6000](https://powergpu.ai/gpu/rtx-a6000) | $205 | $133 | $49 |
| 70B, fast (llama3.3:70b) | [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | $1,363 | $885 | $329 |

Add a volume for the model library (/root/.ollama, 30–100 GB at $0.08/GB/month) so pulled models survive stop/start and instance swaps.

## Setup in one command

*Ollama on the cheapest 12 GB card*

```
powergpu launch --gpu rtx-3060 --template ollama --disk 40 --volume ollama:/root/.ollama
# ✓ instance i-2b7e11c0 running (26.4s) · $0.042/hr
# https://i-2b7e11c0.powergpu.ai:11434  (Ollama API, TLS)
curl https://i-2b7e11c0.powergpu.ai:11434/api/pull -d '{"name":"llama3.1:8b"}'
```

Want a chat UI on top? The [Open WebUI template](https://powergpu.ai/templates/open-webui-ollama) bundles Ollama with a multi-user interface, RAG and model management behind TLS.

## Five ways to spend less

- **Match the card to the model, not the ambition.** An 8B assistant on an H100 is 40× the price of the same assistant on a 3060 and does not answer better.
- **Stop it when nobody is talking.** Per-second billing means a 9-to-5 assistant costs a third of a 24/7 one; the disk keeps the models.
- **Reserve if it never sleeps.** −35% from three months — $0.027/hr for the 3060, $0.182/hr for the A6000.
- **Q4 over FP16.** Half the VRAM class, nearly the same answers for chat; keep FP16 for evaluation runs.
- **Two cheap cards can beat one expensive one.** 2× RTX 4090 (48 GB) at $0.654/hr runs 70B for less than one L40S — slower per token, cheaper per month.

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/cheapest-gpu-for-ollama · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/wan-video-generation-gpu -->

---
title: "Wan 2.x video generation: which GPU, minutes per clip, cost per clip"
description: "VRAM needs for Wan 2.1 and 2.2 (1.3B, 5B, 14B), realistic minutes per five-second clip on RTX 4090, 5090, L40S, H100 and H200, and the price of a batch night."
url: https://powergpu.ai/guides/wan-video-generation-gpu
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Choosing hardware

# Wan 2.x video generation: which GPU, minutes per clip, cost per clip

VRAM needs for Wan 2.1 and 2.2 (1.3B, 5B, 14B), realistic minutes per five-second clip on RTX 4090, 5090, L40S, H100 and H200, and the price of a batch night.

9 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- Wan 14B needs 24 GB with FP8 and block swapping, is comfortable at 32 GB, and runs without tricks from 48 GB.
- A 5-second 81-frame clip takes about 7 minutes at 480p and 14 at 720p on an RTX 4090, 5 minutes on an H100 PCIe.
- On PowerGPU the H100 PCIe at $1.867 per hour is cheaper per 720p clip than the RTX 4090, because the clock runs shorter.
- Wan 2.1 1.3B and Wan 2.2 5B run on 12-24 GB; watch VRAM rather than utilisation, since offloading is a speed tax.

## Wan 2.1 and 2.2: which model needs what

| Model | Weights (BF16) | Minimum VRAM (with offload) | Comfortable | Best for |
| --- | --- | --- | --- | --- |
| Wan 2.1 T2V 1.3B | ~2.6 GB | 8 GB | 12 GB | 480p tests, fast iteration |
| Wan 2.2 TI2V 5B | ~10 GB | 12 GB | 24 GB | 720p on consumer cards, image-to-video |
| Wan 2.1 / 2.2 14B (T2V, I2V) | ~28 GB | 24 GB (FP8 + block swap) | 48–80 GB | Production quality, 720p, LoRAs |
| Wan 2.2 14B MoE (high + low noise) | ~56 GB total | 24 GB (FP8, swapped) | 80–141 GB | Best motion quality; two experts loaded in turn |

The text encoder (umT5, ~11 GB in BF16) and the VAE add to those numbers unless offloaded to CPU — which is exactly what ComfyUI's Wan workflows and Wan2GP do on 24 GB cards, at a speed cost.

## Minutes and dollars per clip, card by card

Indicative times for a 5-second, 81-frame clip with the 14B model at default steps (community benchmarks, single clip, no batching). Cost = today's on-demand rate × minutes; interruptible is half.

| Card | Setup | 480p | 720p | $ / 480p clip | $ / 720p clip | Note |
| --- | --- | --- | --- | --- | --- | --- |
| [RTX 3090](https://powergpu.ai/gpu/rtx-3090) (24 GB · $0.108/hr) | Wan 2.1 1.3B / Wan 2.2 5B only | 4 min | — | $0.007 | — | Too little VRAM for 14B without heavy offload |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (24 GB · $0.327/hr) | FP8 + offloading | 7 min | 14 min | $0.038 | $0.076 | The community default; 24 GB with block swapping |
| [RTX 5090](https://powergpu.ai/gpu/rtx-5090) (32 GB · $0.439/hr) | FP8, light offload | 4.5 min | 9 min | $0.033 | $0.066 | 32 GB and 1.79 TB/s — noticeably faster |
| [L40S](https://powergpu.ai/gpu/l40s) (48 GB · $0.514/hr) | BF16, no offload at 480p | 5 min | 9.5 min | $0.043 | $0.081 | 48 GB; datacenter card for batch queues |
| [RTX PRO 6000 WS](https://powergpu.ai/gpu/rtx-pro-6000-ws) (96 GB · $1.040/hr) | BF16, no offload | 3.5 min | 7 min | $0.061 | $0.121 | 96 GB — 720p without any offload tricks |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (80 GB · $1.867/hr) | BF16, no offload | 2.5 min | 5 min | $0.078 | $0.156 | 80 GB HBM; the throughput pick |
| [H200](https://powergpu.ai/gpu/h200) (141 GB · $2.791/hr) | BF16, big batches | 2.2 min | 4.2 min | $0.102 | $0.195 | 141 GB — longest clips, highest resolutions |

Read it as two tiers. Consumer cards make each clip cheap but slow; 80 GB+ cards make each clip fast and, because the clock runs shorter, often *cheaper too* once you count wall-clock. The H100 PCIE at $0.156 per 720p clip beats the RTX 4090 at $0.076 outright.

## A batch night, budgeted

Sixty 720p clips for a storyboard, queued overnight on interruptible capacity:

| Card | Wall-clock | GPU cost (interruptible) | Model volume 200 GB | Total |
| --- | --- | --- | --- | --- |
| RTX 4090 | 14.0 h | $2.28 | $16.00/mo | **$18.28** |
| RTX 5090 | 9.0 h | $1.97 | $16.00/mo | **$17.97** |
| H100 PCIE | 5.0 h | $4.67 | $16.00/mo | **$20.67** |
| H200 | 4.2 h | $5.86 | $16.00/mo | **$21.86** |

Each clip is an independent queue item, so an interruption re-runs one clip from the volume, not the night.

## Setup: ComfyUI or Wan2GP

*ComfyUI with the model library on a volume*

```
powergpu launch --gpu rtx-5090 --template comfyui --disk 80 --volume video-models:/workspace/ComfyUI/models
# ✓ instance i-77e0c9d3 running (29.1s) · $0.439/hr
# https://i-77e0c9d3.powergpu.ai:8188  (ComfyUI — load the Wan 2.2 template workflow)
```

[ComfyUI](https://powergpu.ai/templates/comfyui) carries the official Wan workflows, LoRA support and the video-helper nodes; [Wan2GP](https://powergpu.ai/templates/wan2gp) is the simpler UI with the most aggressive memory optimisations for 12–24 GB cards. Download the 14B weights (20–60 GB) once onto the volume and they mount on every future instance in seconds.

## What makes video cheap

- **Prototype at 480p on a 5090, final-render at 720p on 80 GB** — same workflow JSON, different card.
- **Queue clips, not sessions.** Per-clip jobs on interruptible capacity pay half and survive interruptions.
- **Watch VRAM, not GPU utilisation.** Video pipelines run out of memory long before they saturate compute; offloading is a speed tax you can buy out with a bigger card.
- **FP8 weights on Ada/Blackwell** halve memory with little visible loss; keep BF16 for the final pass if you can afford the card.
- **Destroy the instance, keep the volume.** Models stay warm at $0.08/GB/month while the GPU bills nothing.

Adjacent reading: the [video generation playbook](https://powergpu.ai/use-cases/video-generation) and [RTX 5090 vs RTX 4090](https://powergpu.ai/compare/rtx-5090-vs-rtx-4090).

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/wan-video-generation-gpu · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/guides/best-gpu-for-whisper -->

---
title: "Best GPU for Whisper transcription (2026): speed and cost per audio hour"
description: "faster-whisper large-v3 throughput on T4, L4, RTX 3060, RTX 4090 and H100 — and the only number that matters: cents per hour of audio transcribed."
url: https://powergpu.ai/guides/best-gpu-for-whisper
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Guide · Costs & pricing

# Best GPU for Whisper transcription (2026): speed and cost per audio hour

faster-whisper large-v3 throughput on T4, L4, RTX 3060, RTX 4090 and H100 — and the only number that matters: cents per hour of audio transcribed.

7 min read Published 2026-09-03 Updated 2026-09-03 (prices live from the sheet)

TL;DR

- The RTX 4090 gives the lowest cost per audio hour for batch transcription, running faster-whisper large-v3 at about 95× real time.
- For an always-on API rent the 72 W L4 at $0.225 per hour on PowerGPU, near 40× real time and 35 percent cheaper reserved.
- Whisper large-v3 needs only about 3 GB in float16 and 1.5 GB in int8, so VRAM never decides; throughput times price does.
- Batching is the difference between 8× and 90× real time on one 4090; VAD skips silence for 20-40 percent less compute.

## The only metric: cents per audio hour

Whisper fits on anything — large-v3 is about 3 GB in float16 — so VRAM never decides. What decides is **throughput × price**: how many hours of audio a card transcribes per hour of rent. With faster-whisper's batched pipeline the spread between a T4 and an H100 is roughly 10× in speed but only about 2× in cost per audio hour, because the fast cards cost more per hour. Pick by the last column below, then by whether the job is a batch or a service.

## Cards ranked by cost per audio hour

Indicative speeds for faster-whisper large-v3, float16, batched pipeline (batch 16, VAD on), expressed as multiples of real time; your audio mix, batch size and int8 settings move them. Prices are today's on-demand rates.

| Card | On-demand | Speed (× real time) | Audio hours per GPU-hour | $ per audio hour | Note |
| --- | --- | --- | --- | --- | --- |
| [Tesla T4](https://powergpu.ai/gpu/tesla-t4) (16 GB) | $0.103 | ~18× | 18 h | **$0.0057** | Cheapest datacenter card; int8 helps |
| [RTX 3060](https://powergpu.ai/gpu/rtx-3060) (12 GB) | $0.042 | ~28× | 28 h | **$0.0015** | Cheapest consumer card that fits large-v3 easily |
| [L4](https://powergpu.ai/gpu/l4) (24 GB) | $0.225 | ~40× | 40 h | **$0.0056** | 72 W, single-slot — the always-on API card |
| [RTX 4070](https://powergpu.ai/gpu/rtx-4070) (12 GB) | $0.075 | ~45× | 45 h | **$0.0017** | Ada speed at a low rate for batch jobs |
| [RTX 4090](https://powergpu.ai/gpu/rtx-4090) (24 GB) | $0.327 | ~95× | 95 h | **$0.0034** | Batch transcription workhorse |
| [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) (80 GB) | $0.374 | ~110× | 110 h | **$0.0034** | Many parallel streams, 80 GB for big batches |
| [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) (80 GB) | $1.867 | ~170× | 170 h | **$0.0110** | Highest throughput per card |

Interruptible capacity halves every number in the price column — transcription queues are the ideal interruptible workload, since every file is an independent, restartable item.

## Which card for which job

| Job | Rent | Because |
| --- | --- | --- |
| A podcast archive, thousands of hours, once | [RTX 4090](https://powergpu.ai/gpu/rtx-4090), interruptible | Lowest $/audio-hour of the consumer cards; batch tolerates pauses |
| An always-on transcription API | [L4](https://powergpu.ai/gpu/l4), reserved | 72 W datacenter card, −35% reserved, ~40× real time is plenty for streaming |
| Meetings for a small team, evenings only | [RTX 3060](https://powergpu.ai/gpu/rtx-3060) | Cheapest card that runs large-v3 comfortably; stop it when idle |
| Thousands of concurrent streams | [A100 PCIE](https://powergpu.ai/gpu/a100-pcie) or [H100 PCIE](https://powergpu.ai/gpu/h100-pcie) | 80 GB for big batches, highest aggregate throughput |
| Diarization + transcription pipelines | RTX 4090 or L4 | Pyannote models add ~2 GB; both cards have the room |

## Setup: WebUI and API in 30 seconds

*the Whisper template on an L4*

```
powergpu launch --gpu l4 --template whisper-webui-api --disk 40 --volume audio:/data
# ✓ instance i-5d0a91f2 running (27.3s) · $0.225/hr
# https://i-5d0a91f2.powergpu.ai:7860  (Web UI)   ·   :8000/transcribe  (REST)
curl -F 'file=@call.mp3' -F 'model=large-v3' https://i-5d0a91f2.powergpu.ai:8000/transcribe
```

The [Whisper WebUI & API template](https://powergpu.ai/templates/whisper-webui-api) runs faster-whisper behind a browser UI and a REST endpoint with word timestamps, language detection and subtitle export. Mount a volume for input files and transcripts so instances stay disposable.

## Squeezing more out of any card

- **Batch.** The batched pipeline is the difference between 8× and 90× real time on the same 4090; single-file sequential decoding wastes the card.
- **int8 on small cards.** On the T4 and 3060, int8 weights run faster with negligible accuracy loss; keep float16 on Ada and newer.
- **VAD first.** Voice activity detection skips silence — 20–40% less compute on meeting audio.
- **Right-size the model.** turbo and distil-large-v3 run 4–6× faster than large-v3 for English at a small accuracy cost.
- **Interruptible for queues, reserved for APIs.** Batch jobs pay half; a 24/7 endpoint pays −35% and never sleeps.

Related: [serving LLMs on the same cards](https://powergpu.ai/use-cases/llm-inference) and the [L4 vs T4 comparison](https://powergpu.ai/compare/l4-vs-tesla-t4).

---

Put the numbers to work

Every price in this guide is our live rate — fixed, ≥30% under the market median, billed per second. Deploy the exact setup above from the [console](https://cloud.powergpu.ai/) in about 30 seconds, paid in crypto, no card and no KYC.

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/guides/best-gpu-for-whisper · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs -->

---
title: "PowerGPU Documentation — Deploy, Manage & Bill Cloud GPUs"
description: "PowerGPU documentation: quickstart, instance lifecycle, templates, volumes, networking, billing, API keys and limits — everything needed to run GPUs at fixed prices."
url: https://powergpu.ai/docs
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Documentation

# PowerGPU documentation: everything the platform does, written down

Short pages, real commands, no marketing. If something here does not match what the console does, that is a bug — [tell us](https://cloud.powergpu.ai/app/support) and we fix both.

- [**Quickstart** Account to running GPU in five minutes — deploy, connect, stop, destroy.](https://powergpu.ai/docs/quickstart)
- [**Billing & credits** Per-second billing mechanics, crypto top-ups, runway and auto-stop.](https://powergpu.ai/docs/billing)
- [**Instance lifecycle** States, actions, what bills when, interruptions and auto-requeue.](https://powergpu.ai/docs/instances)
- [**Templates & images** One-click stacks, environment variables, ports, private registries.](https://powergpu.ai/docs/templates)
- [**Volumes & storage** Instance disks vs volumes, mounts, snapshots, read-only sharing.](https://powergpu.ai/docs/volumes)
- [**Networking & ports** Public endpoints, port mapping, TLS, bandwidth billing.](https://powergpu.ai/docs/networking)
- [**API keys** Create and scope pg_live_ keys for the REST API, CLI and SDK.](https://powergpu.ai/docs/api-keys)
- [**Quotas & limits** Per-account defaults, instance caps and how to raise them.](https://powergpu.ai/docs/limits)
- [**FAQ** The questions support answers every day, written down.](https://powergpu.ai/docs/faq)

## Building against the platform?

- [**REST API** Search offers, launch and manage instances, read balances — try the public endpoints with one curl.](https://powergpu.ai/api)
- [**CLI** powergpu launch / stop / logs from any shell — everything the console does, scriptable.](https://powergpu.ai/cli)
- [**Python SDK** A typed client for pipelines and notebooks — pip install powergpu.](https://powergpu.ai/sdk)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/quickstart -->

---
title: "Quickstart — deploy your first cloud GPU in 5 minutes | PowerGPU Docs"
description: "From account to a running cloud GPU in five minutes: crypto top-up, template deploy, connect over SSH or browser, stop and destroy — exact clicks and commands."
url: https://powergpu.ai/docs/quickstart
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Getting started 2 min read updated 2026-09-03

# Quickstart

Five minutes from nothing to a GPU shell prompt. No card, no KYC — an email, a password and a few dollars of crypto.

## 1 · Create an account

Open the [console](https://cloud.powergpu.ai/?auth=signup), choose email + password, pass the anti-bot check. There is no confirmation email and no trial paperwork — the account exists immediately. New accounts start with a $0.00 balance.

## 2 · Top up credits

**Billing → Add credits.** Pick an amount ($30 minimum — plenty for a first session) and a coin — USDT (TRC-20 or ERC-20), BTC, XMR, LTC, ETH, TRX or SOL. Send the *exact* amount shown to the generated address; the payment page tracks confirmation live and credits land automatically. TRON and Solana confirm in seconds, Bitcoin takes 10–30 minutes.

Send on the network shown on the payment page. USDT on the wrong network is the one mistake support cannot always undo.

## 3 · Deploy a GPU

The console home is the offer search. Filter by GPU (say, [RTX 4090](https://powergpu.ai/gpu/rtx-4090) at $0.327/hr), pick a machine, hit **Deploy**:

- **Template** — PyTorch for a first run (Jupyter + SSH included);
- **Disk** — 50 GB is enough to start (billed $0.08/GB/mo, per second);
- **Mode** — on-demand for interactive work; interruptible is −50% for jobs that can pause.

Deploys need a balance of at least $0.10. The instance reaches *running* in about 30 seconds for cached templates.

## 4 · Connect

The instance page lists every mapped port with copyable endpoints:

*connect*

```
# SSH (key added at deploy, or web terminal in the console)
ssh root@i-9f2c41ab.powergpu.ai -p 2201

# Jupyter — open the mapped 8888 link, token shown on the instance page
nvidia-smi   # your GPU, your prompt
```

## 5 · Stop or destroy

Two different verbs, two different bills:

| Action | GPU billing | Disk billing | Data |
| --- | --- | --- | --- |
| **Stop** | ends that second | continues | kept — restart anytime |
| **Destroy** | ends | ends | gone (volumes survive) |

Balance hits zero? Running instances stop automatically and you get an email — nothing is destroyed. Details in [billing](https://powergpu.ai/docs/billing).

## Where next

- [Instance lifecycle](https://powergpu.ai/docs/instances) — states, actions, interruptions;
- [Volumes](https://powergpu.ai/docs/volumes) — storage that outlives instances;
- [REST API](https://powergpu.ai/api) & [CLI](https://powergpu.ai/cli) — script everything you just clicked.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/quickstart · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/billing -->

---
title: "GPU billing & credits — per-second, crypto top-ups | PowerGPU Docs"
description: "How PowerGPU billing works: USD credit balance, crypto top-ups and confirmations, per-second draw-down, what bills in each instance state, auto-stop at zero."
url: https://powergpu.ai/docs/billing
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Getting started 2 min read updated 2026-09-03

# Billing & credits

One balance in USD, drawn down per second. Everything on this page is the complete billing model — there are no other fees.

## The credit balance

Your account holds a USD balance. Usage — GPU seconds, storage seconds, bandwidth bytes — draws it down continuously; top-ups push it up. The dashboard shows the balance, the current *burn rate* ($/hr of everything running) and the *runway* (balance ÷ burn).

## What bills, when

| Resource | Rate | Bills while… |
| --- | --- | --- |
| GPU instance, running | sheet price × GPUs, per second | state = running |
| GPU instance, stopped | $0 GPU + storage only | disk exists |
| Instance disk / volumes | $0.08/GB/mo ÷ seconds | allocated |
| Bandwidth in/out | $0.01/GB | transferred |

Worked example: an H100 SXM ($1.428/hr) with a 100 GB disk, running 2 h 17 m 08 s then stopped for the rest of the day:

*a day, itemized*

```
GPU     8228 s x $1.428/3600        = $3.2638
disk    24 h x 100 GB x $0.08/730 h  = $0.2630
                                   total = $3.53
```

## Crypto top-ups

**Billing → Add credits**, pick amount and coin, send the exact amount to the generated address before the countdown expires. States you will see on the payment page:

| State | Meaning |
| --- | --- |
| pending | waiting for your transaction to appear on the network |
| confirming | seen on-chain, waiting for confirmations |
| underpaid | partial amount received — the page shows exactly what is left to send |
| completed | credits added (you also get an email receipt) |
| expired / cancelled | window closed or cancelled — funds sent after expiry are handled by support |

Confirmation targets: TRON & Solana seconds, LTC ~2–5 min, ETH/USDT-ERC20 ~1–3 min, XMR ~20 min, BTC 10–30 min. Top-ups are throttled to 8 open payments per hour per account.

## Hitting zero

When the balance reaches $0.00, running instances are **stopped automatically** (never destroyed) and you receive an email. Disks and volumes keep their data; storage billing can push the balance slightly negative until you top up. Deploys and starts require a balance of at least $0.10.

## History & statements

Every transaction — debits per settlement window, top-ups with their on-chain reference — is listed under **Billing** with monthly subtotals. Export needs? The [API](https://powergpu.ai/api) exposes the same ledger as JSON.

Prices are fixed at deploy time: weekly market re-checks never change the rate of an instance that is already running.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/billing · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/instances -->

---
title: "GPU instance lifecycle — states, actions, interruptions | PowerGPU Docs"
description: "Instance state machine: pending, running, stopped, destroyed — what each action does, what bills in each state, how interruptible instances pause and requeue."
url: https://powergpu.ai/docs/instances
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compute 2 min read updated 2026-09-03

# Instance lifecycle

An instance is a GPU allocation + a disk + a template. This page is the complete state machine — what changes state, and what each state costs.

## States

| State | Meaning | GPU bills | Disk bills |
| --- | --- | --- | --- |
| pending | capacity reserved, image pulling / VM booting | no | yes |
| running | workload live, ports reachable | **yes** | yes |
| stopped | processes halted, disk intact, GPU released | no | yes |
| destroyed | gone — disk erased (volumes survive) | no | no |

Transitions are per-second precise: the billing row for a stop written at 14:02:37 ends at 14:02:37.

## Actions

- **Deploy** — needs ≥ $0.10 balance; price is locked to the sheet at that moment.
- **Stop** — graceful signal, then halt. Use it liberally: stopped instances cost only their disk.
- **Start** — re-acquires a GPU on the same machine class and resumes from disk. If the exact machine is busy, the start queues until capacity frees (you can destroy instead at any time).
- **Destroy** — immediate, irreversible for the instance disk. Attached volumes detach unharmed.

*CLI*

```
# the same four verbs everywhere
powergpu launch --gpu rtx-4090 --template pytorch --disk 50
powergpu stop    i-9f2c41ab
powergpu start   i-9f2c41ab
powergpu destroy i-9f2c41ab
```

## Interruptible instances

Interruptible capacity (−50%) can be *paused* when an on-demand deploy needs the slot:

1. The instance receives SIGTERM and a 30-second grace window;
2. it moves to stopped — disk intact, GPU billing over;
3. with auto-requeue on (default), it rejoins the queue and restarts when capacity frees;
4. a console event (and webhook, if configured) records both edges.

Design rule: write checkpoints to a [volume](https://powergpu.ai/docs/volumes) and make your entrypoint resume from the latest one. Then interruptions cost seconds, not work.

## Access

Every instance gets a hostname (i-xxxxxxxx.powergpu.ai) and its template's ports mapped to high ports on it. SSH is available on templates that ship it (a key you provide at deploy, or the in-console web terminal); VMs add full console access. See [networking](https://powergpu.ai/docs/networking) for the port model.

## Events & monitoring

The instance page streams state changes, GPU/RAM utilisation and the live cost counter. The same data is available from GET /v1/instances/{id} for dashboards and schedulers.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/instances · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/templates -->

---
title: "Templates & images — env vars, ports, custom Docker | PowerGPU Docs"
description: "How PowerGPU templates work: pinned images, environment variables, ports, private registries, and bringing your own Docker image with GPU drivers injected."
url: https://powergpu.ai/docs/templates
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compute 2 min read updated 2026-09-03

# Templates & images

A template = a pinned OCI image + launch defaults (ports, mounts, env). The [catalogue](https://powergpu.ai/templates) lists what is maintained; this page is how they behave.

## Anatomy of a template

*template manifest (shown on each card)*

```
image:  powergpu/vllm:0.8            # pinned tag, rebuilt on upstream releases
ports:  8000/http                    # exposed + TLS-terminated for you
mounts: /root/.cache/huggingface     # survives restarts on the instance disk
env:    MODEL=…                      # anything here is overridable at deploy
```

At deploy you can override every field — add env vars, open more ports, change the disk size. Overrides are saved with the instance, so start/stop keeps them.

## Environment variables

Set at deploy (console form, --env flags, or API env map). Secrets note: env values are stored encrypted and never appear in logs, but for long-lived credentials prefer fetching them at boot from your own store.

## Your own Docker image

The *Docker* template accepts any image reference:

*bring your own image*

```
powergpu launch --gpu l40s \
  --image ghcr.io/acme/trainer:v14 \
  --ports 8080 --disk 100 \
  --env WANDB_API_KEY=…
```

- The NVIDIA runtime is injected — do not bundle drivers; CUDA userspace in the image is fine.
- Private registries: add credentials once under **Settings**; they are stored encrypted per-account and used only to pull.
- Entry behaviour: your ENTRYPOINT/CMD runs as-is. No wrapper, no agent inside your container.

## Versioning

Template tags are immutable: a redeploy of powergpu/pytorch:2.6-cuda12.8 is byte-identical next month. New upstream releases arrive as new tags; the console highlights when a newer tag exists but never switches you silently.

## Image caching & cold starts

Maintained templates are pre-pulled on most hosts — that is the ~30-second deploy. Custom images pull on first use on a given machine (network-speed dependent), then cache. Very large images benefit from slimming: each GB is roughly 1–3 s of extra first-boot.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/templates · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/volumes -->

---
title: "Volumes & storage — mounts, snapshots, sharing | PowerGPU Docs"
description: "Instance disks vs network volumes on PowerGPU: creating and mounting volumes, read-only sharing across instances, snapshots, and the flat $0.08/GB/mo billing."
url: https://powergpu.ai/docs/volumes
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compute 2 min read updated 2026-09-03

# Volumes & storage

Two kinds of storage, one price: the disk born with an instance, and volumes that outlive every instance. Both bill $0.08/GB/month, per second.

## Instance disk

Sized at deploy, local NVMe, fastest possible I/O. It survives stop/start and dies with destroy. Use it for scratch, caches and the OS — anything you can regenerate.

## Network volumes

Created independently, attached to instances in the same region, NVMe-backed over the fabric:

*volumes in practice*

```
powergpu volume create --name datasets --size 500 --region eu-west-1
powergpu launch --gpu a100-sxm4 --template pytorch \
  --volume datasets:/data          # read-write
powergpu launch --gpu l40s --template vllm \
  --volume models:/models:ro       # read-only, unlimited attachments
```

- **Attachment model** — one read-write attachment at a time; unlimited read-only attachments. The classic split: one trainer writes checkpoints, a serving fleet reads the model library.
- **Region-bound** — a volume lives in one region; attach from any machine there. Cross-region moves are a snapshot + restore (bandwidth billed at $0.01/GB).
- **Resize** — grow online anytime; shrink = create smaller + copy.

## Snapshots

Point-in-time copies of an instance disk or a volume:

*snapshot & restore*

```
powergpu snapshot create i-9f2c41ab --name golden-env
powergpu volume create --from-snapshot golden-env --name env2 --size 100
```

Snapshots bill as allocated GB at the same flat rate while they exist. The common pattern: snapshot a configured environment, destroy the instance, restore next week for pennies of storage.

## Cost intuition

| Thing | Per month | Per hour |
| --- | --- | --- |
| 50 GB instance disk | $4.00 | $0.0055 |
| 500 GB model library | $40.00 | $0.0548 |
| 2 TB dataset volume | $160 | $0.219 |

Storage is the cheap line — GPU time is the expensive one. Any pattern that lets you destroy GPUs sooner (volumes, snapshots) pays for itself immediately.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/volumes · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/networking -->

---
title: "Networking & ports — endpoints, TLS, bandwidth | PowerGPU Docs"
description: "The PowerGPU network model: per-instance hostnames, port mapping, automatic TLS on HTTP ports, machine port budgets and flat $0.01/GB bandwidth billing."
url: https://powergpu.ai/docs/networking
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Compute 2 min read updated 2026-09-03

# Networking & ports

Every instance gets a hostname and a set of mapped ports. HTTP ports get TLS for free; raw TCP is raw TCP. Bandwidth is one flat number.

## The model

*what a deploy gives you*

```
instance  i-52ab77c1
hostname  i-52ab77c1.powergpu.ai
mapping   container 8000  ->  i-52ab77c1.powergpu.ai:8000   (https, TLS terminated)
          container 22    ->  i-52ab77c1.powergpu.ai:2201   (tcp, direct)
```

- **Declared HTTP ports** are fronted by our edge: TLS certificate, HTTP/2, WebSocket pass-through. Your container speaks plain HTTP inside.
- **TCP ports** (SSH, databases, game servers) map to high ports on the same hostname, direct to the machine.
- **Port budget** — each machine advertises how many ports it can map (8–100, shown on the offer card). Filter by it in the console (portsMin).

## Bandwidth

$0.01/GB, in and out, worldwide — no tiers, no inter-region matrix. The instance page shows a live transfer counter; the ledger itemises it per instance. A 40 GB model download costs $0.40; serving a million 2 KB responses costs about $0.00.

## Network security

- Only declared ports are reachable — everything else is dropped at the edge.
- Instance-to-instance traffic inside a region stays on the private fabric (and is free).
- Outbound is open (package installs, model downloads) with standard abuse controls — see the [acceptable use policy](https://powergpu.ai/legal/aup).

## Practical notes

- **WebSockets & SSE** work through the HTTP edge unmodified (vLLM streaming, ComfyUI live previews).
- **Long-lived connections** idle-timeout at 4 hours on the edge; raw TCP ports have no such timeout.
- **Custom domains** — CNAME your domain to the instance hostname and we mint the certificate; per-instance, from the instance page.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/networking · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/api-keys -->

---
title: "API keys — create, scope, rotate | PowerGPU Docs"
description: "Create pg_live_ API keys in the PowerGPU console, authenticate the REST API, CLI and SDK, rotate and revoke safely. Keys are shown once and stored hashed."
url: https://powergpu.ai/docs/api-keys
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Platform 2 min read updated 2026-09-03

# API keys

One key authenticates the [REST API](https://powergpu.ai/api), the [CLI](https://powergpu.ai/cli) and the [Python SDK](https://powergpu.ai/sdk). Keys act with your account's full permissions — treat them like the balance they can spend.

## Create a key

**Console → API keys → New key.** Give it a label you will recognise in an incident ("laptop-cli", "ci-runner"). The key — format pg_live_ + 40 hex characters — is shown *once*: we store only a SHA-256 hash, so it cannot be re-displayed. Lost key = revoke + new key.

## Use it

*the same key everywhere*

```
# REST
curl -H "Authorization: Bearer pg_live_…" https://powergpu.ai/v1/instances

# CLI
export POWERGPU_API_KEY=pg_live_…
powergpu list

# Python
client = powergpu.Client()          # reads POWERGPU_API_KEY
```

## Key hygiene

- **One key per consumer** — a leaked CI key then revokes without breaking your laptop.
- **Environment, not code** — POWERGPU_API_KEY in your secret store; never commit a key. If a key lands in a public repo, revoke it first, clean git history second.
- **Last-used tracking** — the keys page shows when each key last authenticated; prune the stale ones.
- **Rotation** — create the new key, roll consumers, revoke the old. Both stay valid during the overlap, so rotation is zero-downtime.

## Revocation

Deleting a key invalidates it within seconds platform-wide. Running instances are unaffected — keys gate the management plane, not the workloads.

Changing your account password signs out browser sessions but does *not* touch API keys — revoke those explicitly if the account may be compromised.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/api-keys · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/limits -->

---
title: "Quotas & limits — defaults and raises | PowerGPU Docs"
description: "Per-account defaults on PowerGPU: instance and GPU caps, top-up throttles, API rate limits, port budgets — and how limits raise automatically with account history."
url: https://powergpu.ai/docs/limits
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Platform 1 min read updated 2026-09-03

# Quotas & limits

Defaults exist to keep the platform healthy against abuse, not to upsell you a tier. Most raise automatically as an account builds history; anything else is one support ticket.

## Account defaults

| Limit | New account | Established* |
| --- | --- | --- |
| Concurrent instances | 8 | 32 |
| GPUs per instance | 8 (hardware max) | 8 |
| Concurrent GPUs, total | 16 | 64 |
| Volumes | 10 · 4 TB total | 50 · 32 TB |
| Open top-ups per hour | 8 | 8 |
| API requests | 120/min | 600/min |
| Support tickets open | 10 | unlimited |

*Established = roughly two weeks of normal usage and settled top-ups; raises apply automatically, no request needed.

## Hardware-shaped limits

- **GPUs per machine** — 1× to 8× depending on the chassis; the offer card is the truth.
- **Ports per machine** — 8–100 mapped ports, shown per offer, filterable (portsMin).
- **Disk per machine** — up to the free NVMe on that host, shown at deploy.

## API rate behaviour

Past the per-minute budget the API answers 429 with a Retry-After header — back off and retry; nothing is queued or dropped silently. Polling advice: instance state changes also land as webhooks, which do not count against the budget.

## Need more, sooner?

Fleet-scale needs (64+ GPUs, multi-node reservations) skip the automatic ladder — open a [ticket](https://cloud.powergpu.ai/app/support) with the shape of the workload, or read the [enterprise page](https://powergpu.ai/enterprise). Capacity plans answer within a business day.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/limits · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/faq -->

---
title: "Cloud GPU FAQ — 16 common questions, answered | PowerGPU Docs"
description: "The complete PowerGPU FAQ: pricing mechanics, dedicated GPUs, crypto payments, privacy, account security and support — 16 questions, straight answers."
url: https://powergpu.ai/docs/faq
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Platform 3 min read updated 2026-09-03

# Cloud GPU rental: frequently asked questions

Everything support answers weekly, written down. If your question is missing, [ask it](https://cloud.powergpu.ai/app/support) — good questions get added here.

## Pricing & billing

**Why are you 30% cheaper — what is the catch?**

The model is boring on purpose: we commit capacity in verified datacenters at wholesale, price from the public marketplace median minus 30%, and skip the expensive parts of a cloud business (sales teams, card-fraud losses, free-tier subsidies). The price is the marketing.

**Do you round up billing in any way?**

No. GPU seconds at rate/3600, storage at allocated GB-seconds, bandwidth at bytes transferred. The ledger shows raw numbers; totals round only at display time.

**Can I get an invoice?**

Every top-up generates a receipt with its on-chain reference, and the billing page exports a monthly usage statement (CSV/JSON) suitable for accounting.

**What happens to credits if I stop using the platform?**

They stay on the account — credits do not expire. See the terms for the (narrow) refund policy: unused credits are refundable in the original coin on request, minus network fees.

## Instances & hardware

**Are the GPUs shared or dedicated?**

Dedicated. A deployed GPU is passed through to your instance alone — no MIG slicing, no time-sharing, no noisy neighbours on the silicon. CPU cores and RAM are likewise reserved.

**Can I choose the exact machine?**

Yes — the console lists individual machines with CPU, RAM, disk bandwidth, network and port budget. Deploy targets that machine, and start/stop returns to the same machine class.

**Do you support Windows?**

Containers are Linux. Full VMs can boot a Windows image you licence yourself (BYOL) — most rendering shops use the Linux builds of their tools instead.

**What driver/CUDA versions are installed?**

Hosts run the current NVIDIA production branch; templates carry CUDA 12.4–13.0 userspace. The offer card shows the max CUDA per machine — filter if you need a floor.

## Payments

**Which coins do you accept?**

USDT (TRC-20 and ERC-20), Bitcoin, Monero, Litecoin, Ethereum, TRON and Solana. No cards, no bank transfers, no KYC.

**I sent the wrong amount — now what?**

Underpaid: the payment page shows the exact remainder to send to the same address. Overpaid: the excess credits automatically. Sent after expiry or on the wrong network: open a ticket with the tx hash — recoverable cases are recovered.

**Why crypto only?**

Card processing means chargebacks, fraud-scoring users, and identity paperwork. Crypto settlement is final and lets us keep zero payment PII — which is also why we can afford not to store IP addresses. It is a privacy feature and a margin feature at once.

## Account & security

**What personal data do you hold?**

Email, Argon2id password hash, balances, instance metadata and tickets. No names, no addresses, no cards, no stored IPs. The privacy policy is short because the data is.

**Can I have multiple users on one account?**

Sub-accounts with separate balances and API keys ship with enterprise reservations; for self-serve, one account per human plus per-consumer API keys covers most teams.

**I forgot my password.**

Reset link from the sign-in modal to your account email. Note that a password change signs out other sessions but leaves API keys valid — revoke keys separately if compromised.

## Support

**How do I reach support?**

Tickets in the console, 24/7, median first answer under two hours. Every answer also lands in your email inbox. There is deliberately no public support email address — tickets authenticate you and keep account context attached.

**Do you have a status page?**

Yes — /status shows live per-region state, component health and 90-day uptime, no login needed.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/faq · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/docs/mcp -->

---
title: "MCP server — connect an AI agent to PowerGPU | Docs"
description: "Connect any MCP client to PowerGPU: a public read-only Model Context Protocol server with live GPU prices, comparisons and cost estimates. No key, no account."
url: https://powergpu.ai/docs/mcp
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Platform 5 min read updated 2026-09-03

# MCP server

A public, read-only **Model Context Protocol** server lives at https://powergpu.ai/mcp. Point any MCP client at it and your assistant can read the live price sheet, compare cards, size a workload and cost a job — no API key, no account, no write access.

## What it is for

A language model asked "what does an H100 cost per hour" answers from whatever it memorised months ago. Connected to this server, it calls get_gpu_price and answers with today's number, the market median behind it and the page it came from. The same applies to sizing questions ("which GPU runs a 70B model") and to budgets ("what does a fine-tuning weekend cost").

The server is deliberately narrow: it reads the public catalogue and the public inventory. It cannot see an account, cannot deploy, cannot spend. Deployment stays in the [console](https://cloud.powergpu.ai/) and the [authenticated REST API](https://powergpu.ai/api).

## Connect a client

Transport is **Streamable HTTP**: JSON-RPC 2.0 over POST to a single URL. Most clients only need the URL.

*claude_desktop_config.json, .mcp.json, or your client's equivalent*

```
{
  "mcpServers": {
    "powergpu": {
      "type": "http",
      "url": "https://powergpu.ai/mcp"
    }
  }
}
```

Claude Code takes it in one command; other CLIs are similar.

*Claude Code*

```
claude mcp add --transport http powergpu https://powergpu.ai/mcp
```

For a client that only speaks stdio, bridge it:

*stdio bridge to a remote HTTP MCP server*

```
npx -y mcp-remote https://powergpu.ai/mcp
```

## Tools

| Tool | What it answers |
| --- | --- |
| about_powergpu | What PowerGPU is: catalogue, price rule, billing, payment, identity, regions, SLA, interfaces |
| pricing_methodology | The formula behind every price, a worked example, what it does not promise, how to verify it |
| list_gpus | Rentable models with hourly rates, filterable by VRAM, price and tier |
| get_gpu_price | One model in full: three rates, monthly estimate, specs, what fits in VRAM, availability |
| compare_gpus | Two models side by side, with value per TFLOP-hour and per GB of VRAM |
| recommend_gpu | Cheapest suitable card for a workload or a model size, with the VRAM arithmetic |
| estimate_cost | Cost of N GPUs for H hours, including storage and egress, with the arithmetic shown |
| search_offers | Machines available to deploy right now: GPU count, region, CPU, RAM, disk, price |

## Try it without a client

It is an ordinary HTTP endpoint, so curl works:

*the server, from a shell*

```
# what tools exist
curl -s -X POST https://powergpu.ai/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | jq '.result.tools[].name'

# today's H100 SXM price
curl -s -X POST https://powergpu.ai/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"get_gpu_price","arguments":{"gpu":"h100-sxm"}}}' \
  | jq -r '.result.content[0].text'

# cost of 8x H100 for a 40-hour training run, with 2 TB of storage
curl -s -X POST https://powergpu.ai/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call",
       "params":{"name":"estimate_cost","arguments":{"gpu":"h100-sxm","hours":40,
                  "num_gpus":8,"storage_gb":2000}}}' | jq -r '.result.content[0].text'
```

A GET on the same URL returns a plain description of the server, which is handy when you are wiring things up.

## Response shape

Every tool returns human-readable text plus a structuredContent object, so a model can quote the prose and a program can read the fields.

*tools/call → get_gpu_price*

```
{
  "result": {
    "content": [{ "type": "text", "text": "NVIDIA H100 SXM (80 GB): $1.428/hr on-demand …" }],
    "structuredContent": {
      "slug": "h100-sxm",
      "vram_gb": 80,
      "usd_per_gpu_hour": { "on_demand": 1.428, "interruptible": 0.714, "reserved": 0.928 },
      "market_median_usd_hr": 2.0414,
      "percent_below_median": 30,
      "url": "https://powergpu.ai/gpu/h100-sxm"
    }
  }
}
```

## Limits and guarantees

- **Read-only.** No tool writes anything. There is nothing to authorise and nothing to revoke.
- **No account data.** The server has no notion of users, balances or instances; it sees the public catalogue only.
- **No key, no rate limit worth worrying about.** Ordinary abuse protection applies at the edge, nothing more.
- **Live prices.** Every figure comes from the same sheet the website renders, at the market snapshot in force (2026-09-14).
- **Protocol.** MCP 2025-06-18 over Streamable HTTP. Batched JSON-RPC requests are accepted.

## Other machine-readable surfaces

- [/llms.txt](https://powergpu.ai/llms.txt) — the site index with the key facts, for assistants that read it.
- [/llms-full.txt](https://powergpu.ai/llms-full.txt) — every page of the site converted to Markdown, one file.
- **Markdown for any page** — append.md to a page URL ([/pricing.md](https://powergpu.ai/pricing.md)) or send Accept: text/markdown.
- [/openapi.json](https://powergpu.ai/openapi.json) — OpenAPI 3.1 description of the [REST API](https://powergpu.ai/api), for function calling and client generation.
- [/gpu-price-index.json](https://powergpu.ai/gpu-price-index.json) and [.csv](https://powergpu.ai/gpu-price-index.csv) — the dated price dataset, CC BY 4.0.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/docs/mcp · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/api -->

---
title: "GPU Cloud REST API — Search, Launch & Manage GPUs | PowerGPU"
description: "The PowerGPU REST API: public pricing and offer search with no key needed, plus authenticated instance lifecycle and balance endpoints with Bearer API keys."
url: https://powergpu.ai/api
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Developers · REST API

# A GPU cloud API you can try before signing up

Pricing and offer search are public endpoints — paste the curl below into any terminal right now. Everything the console does (deploy, stop, destroy, balance) is the same API with a Bearer key.

## Basics

- **Base URL**: https://powergpu.ai/v1
- **Auth**: Authorization: Bearer pg_live_…
- **Format**: JSON in, JSON out, UTF-8. Errors: {"error":{"code","message"}}
- **Rate limits**: 120 req/min (auto-raises with account history) — 429 + Retry-After past it

## Public endpoints — try them now

### GET /v1/gpus/pricing

The complete price sheet as data: all 80 models, three modes each, market medians and the pricing rule itself.

*no key needed*

```
curl https://powergpu.ai/v1/gpus/pricing
```

*response (truncated)*

```
{
  "object": "gpu_pricing",
  "currency": "USD",
  "billing": "per_second",
  "price_rule": "market_median * 0.70, rounded down; …",
  "market_snapshot": "2026-09-14",
  "data": [
    {
      "slug": "h100-sxm",
      "name": "NVIDIA H100 SXM",
      "vram_gb": 80,
      "price_per_gpu_hour": {
        "on_demand": 1.428,
        "interruptible": 0.714,
        "reserved": 0.928
      },
      "market_median": 2.0414,
      "available_gpus": 54
    }, …
  ]
}
```

### GET /v1/offers

Live machine inventory. Filters: gpu (slug), region, num_min, price_max ($/hr for the whole machine), limit (≤100).

*find 2× RTX 4090 machines*

```
curl "https://powergpu.ai/v1/offers?gpu=rtx-4090&num_min=2&limit=3"
```

## Authenticated endpoints

Create a key under [Console → API keys](https://cloud.powergpu.ai/app/keys) ([docs](https://powergpu.ai/docs/api-keys)), then:

### GET /v1/balance

```
curl -H "Authorization: Bearer " https://powergpu.ai/v1/balance

{ "object": "balance", "currency": "USD", "balance": 41.2183, "month_spend": 12.8402 }
```

### POST /v1/instances

Deploy on a machine from /v1/offers. Body: machine_id (required), template, disk_gb, type (od | interruptible | res).

*deploy*

```
curl -X POST -H "Authorization: Bearer " \
  -H "Content-Type: application/json" \
  -d '{"machine_id":"m-1d29c04a","template":"vllm","disk_gb":80,"type":"od"}' \
  https://powergpu.ai/v1/instances

{ "object": "instance", "id": "i-52ab77c1", "status": "running",
  "hostname": "i-52ab77c1.powergpu.ai", "price_hr": 1.5983, … }
```

### GET /v1/instances · GET /v1/instances/{id}

List active instances / fetch one (state, price, hostname, timestamps). Billing is settled at read time, so the numbers are per-second fresh.

### POST /v1/instances/{id}/actions

```
-d '{"action":"stop"}'     # or start | destroy
```

### DELETE /v1/instances/{id}

Alias for the destroy action. Instance disk is erased; attached volumes survive.

## Errors

| Status | code | Meaning |
| --- | --- | --- |
| 401 | unauthorized | Missing or unknown Bearer key |
| 404 | not_found / no_route | Unknown instance or endpoint |
| 409 | deploy_failed / action_failed | Business rule said no — the message says why (balance, capacity, state) |
| 422 | missing_machine / bad_action | Request shape is wrong |
| 429 | rate_limited | Slow down; honour Retry-After |

Prefer a wrapper? The [CLI](https://powergpu.ai/cli) and [Python SDK](https://powergpu.ai/sdk) cover this API one-to-one.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/api · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/cli -->

---
title: "GPU Cloud CLI — Launch & Manage GPUs from Your Shell | PowerGPU"
description: "The powergpu CLI: pip install powergpu, then search offers, launch templates, stream logs, stop and destroy — every console action, scriptable."
url: https://powergpu.ai/cli
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Developers · CLI

# The powergpu CLI: the console, without the browser

One package installs the CLI and the Python SDK. Every command maps 1:1 onto the [REST API](https://powergpu.ai/api), so anything you can click, you can script — and anything you script behaves exactly like the console did.

## Install & authenticate

*setup, once*

```
pip install powergpu
export POWERGPU_API_KEY=pg_live_…   # console → API keys
powergpu whoami
# account ok · balance $41.22 · 2 instances running
```

The key can also live in ~/.config/powergpu/config.toml — the CLI reads the environment first, then the file. Create keys in the [console](https://cloud.powergpu.ai/app/keys) ([key docs](https://powergpu.ai/docs/api-keys)).

## Find capacity

*search*

```
# the price sheet, in your terminal
powergpu gpus --sort price

# machines matching a shape
powergpu offers --gpu rtx-4090 --num 2 --region eu-west-1
# ID          GPU        ×  REGION      vCPU  RAM    DISK   $/HR
# m-8b02c1f3  RTX 4090   2  eu-west-1   16    64GB   1TB    0.654
```

## Launch & manage

*lifecycle*

```
powergpu launch --gpu rtx-4090 --template comfyui --disk 60
# ✓ instance i-b81f02aa running (28.4s)
# ✓ https://i-b81f02aa.powergpu.ai:8188  (comfyui)

powergpu list                     # your fleet, states, burn/hr
powergpu logs i-b81f02aa -f       # stream container logs
powergpu stop i-b81f02aa          # GPU billing ends this second
powergpu start i-b81f02aa
powergpu destroy i-b81f02aa --yes
```

## Volumes & snapshots

```
powergpu volume create --name models --size 500 --region eu-west-1
powergpu launch --gpu l40s --template vllm --volume models:/models:ro
powergpu snapshot create i-b81f02aa --name golden
```

## Scripting patterns

Every command takes --json for machine-readable output, and exit codes follow convention (0 ok, 1 API error, 2 usage):

*ephemeral GPU jobs*

```
# deploy, wait, run, destroy — a disposable GPU in a Makefile
ID=$(powergpu launch --gpu rtx-4090 --template pytorch --json | jq -r .id)
powergpu wait "" --state running
powergpu exec "" -- python train.py
powergpu destroy "" --yes
```

## Shell completion

```
powergpu completion bash >> ~/.bashrc     # zsh & fish too
```

Building in Python instead? The [SDK page](https://powergpu.ai/sdk) shows the same flows as typed calls.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/cli · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- page: https://powergpu.ai/sdk -->

---
title: "GPU Cloud Python SDK — Typed Client for the PowerGPU API"
description: "The powergpu Python SDK: a typed client over the REST API — search offers, deploy, wait, execute, destroy. Context managers for disposable GPUs."
url: https://powergpu.ai/sdk
last_modified: 2026-09-14T11:04:20+00:00
prices_as_of: 2026-09-14
site: PowerGPU (powergpu.ai)
---

Developers · Python SDK

# Python SDK: GPUs as a context manager

The same package as the CLI, importable: typed dataclasses over the [REST API](https://powergpu.ai/api), sensible retries, and a context manager that guarantees the meter stops when your block exits — even on exceptions.

## Install

```
pip install powergpu            # Python ≥3.10, zero heavy deps
export POWERGPU_API_KEY=pg_live_…
```

## Sixty seconds of SDK

*core flow*

```
import powergpu

client = powergpu.Client()                     # reads POWERGPU_API_KEY

# the public price sheet — no key needed for this call
for g in client.gpus():
    print(g.slug, g.price.on_demand)

# find capacity and deploy
offer = client.offers(gpu="rtx-4090", num_min=1)[0]
inst  = client.instances.create(
    machine_id=offer.machine_id,
    template="pytorch",
    disk_gb=60,
    type="interruptible",
)
inst.wait("running")
print(inst.hostname, inst.price_hr)

inst.stop()                                    # billing ends this second
```

## Disposable GPUs, guaranteed

The pattern that saves real money: the context manager destroys the instance on exit, *including* when your code raises.

*context manager*

```
with client.ephemeral(gpu="a100-sxm4", template="axolotl",
                      volume="ckpts:/ckpts") as gpu:
    gpu.exec("axolotl train qlora.yml")        # streams output
    gpu.download("/ckpts/adapter", "./out")    # billed at $0.01/GB
# ← destroyed here, success or crash — no forgotten $2/hr instances
```

## Async & fleets

*parallel render farm in 12 lines*

```
import asyncio, powergpu

async def frame(n: int):
    async with powergpu.AsyncClient().ephemeral(
            gpu="rtx-4090", template="blender",
            volume="scene:/scene:ro") as gpu:
        await gpu.exec(f"blender -b /scene/shot.blend -f {n}")

# 8 frames in parallel — per-second billing makes this cost the same
asyncio.run(asyncio.gather(*[frame(n) for n in range(1, 9)]))
```

## Errors & retries

- HTTP 429/5xx retry with exponential backoff (configurable, off for POST by default);
- API errors raise powergpu.APIError carrying.code and.message straight from the [error table](https://powergpu.ai/api);
- business rules (low balance, machine gone) raise DeployError — catch it, top up, retry.

## Typing

Fully typed (py.typed marker): editors autocomplete offers, instances and prices; mypy passes on strict. Dataclasses mirror the JSON of the REST API field for field, so the [API reference](https://powergpu.ai/api) doubles as the SDK reference.

Stuck?

Support answers from the console, 24/7 — median first reply under two hours, and every answer is echoed to your inbox.

[Open a ticket](https://cloud.powergpu.ai/app/support)

Build against it

- [REST API](https://powergpu.ai/api)
- [CLI](https://powergpu.ai/cli)
- [Python SDK](https://powergpu.ai/sdk)
- [Guides](https://powergpu.ai/guides)

---

*About PowerGPU:* PowerGPU (powergpu.ai) is a cloud GPU rental service offering 80 NVIDIA GPU models — from the RTX A2000 at $0.024/hr to the B300 — at fixed prices set at least 30% below the public GPU marketplace median and re-checked weekly (H100 SXM: $1.428/hr on-demand). Billing is per second with no minimums; payment is crypto only (USDT, BTC, XMR, ETH, SOL, LTC, TRX) with no KYC. Instances run in Tier-III datacenters across 32 regions with a 99.9% uptime SLA and deploy in about 30 seconds from the web console (cloud.powergpu.ai) or the REST API.

Source: https://powergpu.ai/sdk · Site index for AI assistants: https://powergpu.ai/llms.txt · Full content: https://powergpu.ai/llms-full.txt


<!-- 243 pages -->