B200 vs H200.
prices baked 2026-08-24 06:03 UTC · live, always-current version of this comparison →
| vs | B200 | H200 |
|---|---|---|
| cheapest idle /GPU/hr | $3.06 | $0.927 |
| cheapest on-demand /GPU/hr | $4.63 | $3.30 |
| idle discount | −34% | −72% |
| live offers | 54 | 51 |
| vram, max listed | 180 GB | 141 GB |
| fp16 spec TFLOPS | 2250 | 989 |
| mem bandwidth GB/s | 8,000 | 4,800 |
| TFLOPS·h per $ (at floor) | 736 | 1,067 |
| VRAM GB·h per $ (at floor) | 59 | 152 |
At today’s floors the H200 delivers 1,067 FP16 TFLOPS·hours per dollar against 736 — 1.4× the training compute per dollar. Spec ceilings, not benchmarks.
The amber cell wins its row — lowest on price, highest on everything else. Prices are each model’s cheapest live listing across 10 marketplaces at bake time; idle prices are reclaimable bid-market floors. Spec figures are vendor ceilings, unrated cells say so.
Other classics: H100 SXM vs A100 SXM4 H200 vs H100 SXM RTX 4090 vs RTX 3090 RTX 5090 vs RTX 4090 RTX 4090 vs A100 PCIE L40S vs A100 PCIE A100 PCIE vs TESLA V100 RTX A6000 vs RTX 3090 RTX 6000 ADA vs RTX A6000 RTX 4090 vs L40S RTX 5090 vs H100 SXM RTX 4090 vs RTX 4080 RTX 4080 vs RTX 3090 RTX 3090 vs RTX 3080 RTX A6000 vs A100 PCIE RTX 5090 vs L40S L4 vs TESLA T4 H100 SXM vs MI300X RTX 5090 vs A100 PCIE H100 PCIE vs H100 SXM L40S vs RTX 6000 ADA RTX 4070 vs RTX 3080 B200 vs H100 SXM