H100 SXM vs B200
The H100 SXM rents for 47% less per hour today — $3.50 against $6.59 — with 80 GB against 180 GB of memory.
Prices collected 2026-09-18 · 25 providers price both
Index median per GPU-hour, on-demand, per snapshot.
Side by side
| H100 SXM | B200 | |
|---|---|---|
| Median rental, per GPU-hour | $3.50 | $6.59 |
| Cheapest listing | $1.68 | $3.75 |
| Per month at 720 h | $2,516 | $4,745 |
| Providers pricing it | 46 | 26 |
| Memory | 80 GB HBM3 | 180 GB HBM3e |
| Memory bandwidth | 3,350 GB/s | 8,000 GB/s |
| GB of memory per $/hr | 22.9 | 27.3 |
| Architecture | Hopper | Blackwell |
| Class | Datacenter | Datacenter |
| Released | 2022-09 | 2024-12 |
Same provider, both GPUs
The cleanest comparison: 25 providers price both cards, so the difference is the GPU and nothing else. The H100 SXM is cheaper at 25 of them.
| Provider | H100 SXM | B200 | B200 premium |
|---|---|---|---|
| $2.00 | $4.00 | +100% | |
 Koyeb | $2.50 | $5.50 | +120% |
 Hyperstack | $2.60 | $6.00 | +131% |
| $2.64 | $9.36 | +255% | |
 RunPod Community Cloud | $2.69 | $5.98 | +122% |
| $2.69 | $6.99 | +160% | |
| $2.71 | $5.00 | +85% | |
 Lyceum | $2.79 | $6.39 | +129% |
| $2.81 | $5.43 | +93% | |
 Cerebrium | $3.40 | $6.01 | +77% |
 Vast.ai | $3.45 | $8.75 | +154% |
 RunPod Secure Cloud | $3.49 | $6.79 | +95% |
 Beam | $3.50 | $4.11 | +17% |
 DataCrunch (Verda) | $3.52 | $6.81 | +94% |
 GPU.ai | $3.69 | $7.78 | +111% |
 io.net | $3.76 | $4.89 | +30% |
 Nebius | $3.85 | $7.15 | +86% |
 Modal | $3.95 | $6.25 | +58% |
 Together AI | $3.99 | $8.19 | +105% |
 Lambda | $4.14 | $6.84 | +65% |
 fal.ai | $4.50 | $6.25 | +39% |
 CoreWeave | $6.16 | $8.60 | +40% |
 Baseten | $6.50 | $9.98 | +54% |
 Amazon Web Services | $6.88 | $14.24 | +107% |
 Oracle Cloud | $10.00 | $14.00 | +40% |
What each one runs, and what a million tokens costs
Weights + KV cache at 8K context + overhead against 90% of each card's memory. Where a model needs more than one card the cell says how many and prices the set at today's median. $/M tokens is the hourly median divided by estimated single-stream output, from memory bandwidth — an estimate for ranking, not a benchmark.
| Model | Bits | H100 SXM | B200 |
|---|---|---|---|
| Kimi K3 | 4-bit | 22× GPU$76.89/hr | 10× GPU$65.90/hr |
| Kimi K3 | 16-bit | 78× GPU$272.61/hr | 35× GPU$230.65/hr |
| DeepSeek V4.1-Flash | 4-bit | 5× GPU$17.48/hr | 2× GPU$13.18/hr |
| DeepSeek V4.1-Flash | 16-bit | 16× GPU$55.92/hr | 7× GPU$46.13/hr |
| Llama 4 Maverick | 4-bit | 4× GPU$13.98/hr | 2× GPU$13.18/hr |
| Llama 4 Maverick | 16-bit | 12× GPU$41.94/hr | 5× GPU$32.95/hr |
| gpt-oss-120b | 4-bit | ✓ 67.9 GB~934 tok/s · $1.04/M | ✓ 67.9 GB~2231 tok/s · $0.82/M |
| gpt-oss-120b | 16-bit | 4× GPU$13.98/hr | 2× GPU$13.18/hr |
| Llama 3.1 70B | 4-bit | ✓ 44.5 GB~68 tok/s · $14.28/M | ✓ 44.5 GB~163 tok/s · $11.23/M |
| Llama 3.1 70B | 16-bit | 3× GPU$10.49/hr | ✓ 145.1 GB~46 tok/s · $39.79/M |
| Gemma 4 31B | 4-bit | ✓ 21.2 GB~154 tok/s · $6.3/M | ✓ 21.2 GB~367 tok/s · $4.99/M |
| Gemma 4 31B | 16-bit | ✓ 65.7 GB~43 tok/s · $22.58/M | ✓ 65.7 GB~103 tok/s · $17.77/M |
| Qwen3 Coder 30B-A3B | 4-bit | ✓ 19 GB~1444 tok/s · $0.67/M | ✓ 19 GB~3448 tok/s · $0.53/M |
| Qwen3 Coder 30B-A3B | 16-bit | ✓ 62.9 GB~406 tok/s · $2.39/M | ✓ 62.9 GB~970 tok/s · $1.89/M |
| Qwen3.8 27B | 4-bit | ✓ 18.7 GB~176 tok/s · $5.52/M | ✓ 18.7 GB~421 tok/s · $4.35/M |
| Qwen3.8 27B | 16-bit | ✓ 57.6 GB~50 tok/s · $19.42/M | ✓ 57.6 GB~119 tok/s · $15.38/M |
| gpt-oss-20b | 4-bit | ✓ 13.7 GB~1323 tok/s · $0.73/M | ✓ 13.7 GB~3160 tok/s · $0.58/M |
| gpt-oss-20b | 16-bit | ✓ 43.9 GB~372 tok/s · $2.61/M | ✓ 43.9 GB~889 tok/s · $2.06/M |
| Llama 3.1 8B | 4-bit | ✓ 6.9 GB~596 tok/s · $1.63/M | ✓ 6.9 GB~1422 tok/s · $1.29/M |
| Llama 3.1 8B | 16-bit | ✓ 18.4 GB~168 tok/s · $5.78/M | ✓ 18.4 GB~400 tok/s · $4.58/M |
Buy or rent?
Hours of renting at today's median that add up to the card's approximate list price. List prices are typical quoted figures, not tracked, and do not include the server, power or the person to run it — so the break-even is a floor, not a verdict.
Questions
Is the H100 SXM or the B200 cheaper to rent?
On 2026-09-18 the H100 SXM has a market median of $3.50 per GPU-hour across 46 providers and the B200 $6.59 across 26, so the H100 SXM is 47% cheaper per hour. Among the 25 providers that price both, the H100 SXM is cheaper at 25 and the B200 at 0. Per hour is only half the answer — the table below prices a whole model on each.
Which should I rent for a 70B model?
At 4-bit, Llama 3.1 70B needs about 44.5 GB. On the H100 SXM that is one card at $3.50/hr; on the B200 one card at $6.59/hr. Estimated single-stream decode is ~68 vs ~163 tokens per second, from memory bandwidth.
What is the real difference between them?
80 GB HBM3 at 3,350 GB/s (Hopper) against 180 GB HBM3e at 8,000 GB/s (Blackwell). Memory decides what fits; bandwidth decides how fast it generates, because each token streams the active weights once. Compute matters for prompt processing and training, not for how fast text appears.