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.
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.
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.
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.
B200 vs B300: FAQ
Deeper reading: H100 vs H200 vs B200, RTX 4090 vs RTX 5090, how cloud GPU pricing works.
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.