Llama 3.3 70B — B200 vs H100
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and H100 (NVIDIA Hopper) on Llama 3.3 70B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
Throughput at 53 tok/s/user on Llama 3.3 70B: B200 hits 6496 tok/s/GPU, H100 hits 1465. Per-million costs land at $0.08 and $0.25 respectively. B200 is 199% cheaper per token; B200 delivers 343% more tok/s/GPU.
B200 / H100 on Llama 3.3 70B at 72 tok/s/user: 4738 / 773 tok/s/GPU, $0.11 / $0.45 per million tokens. B200 is 296% cheaper per token; B200 delivers 513% more tok/s/GPU.
Toward the upper edge of the 35–109 tok/s/user interactivity band, at 91 tok/s/user on Llama 3.3 70B: B200 runs 3574 tok/s/GPU at $0.15/M tokens, H100 runs 304 at $1.19/M. B200 is 704% cheaper per token; B200 delivers 1077% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:6496.2H100:1465.2 | B200:4738.0H100:772.9 | B200:3574.5H100:303.7 |
| Cost ($/M tok) | B200:$0.083H100:$0.249 | B200:$0.115H100:$0.454 | B200:$0.149H100:$1.194 |
| tok/s/MW | B200:2993645H100:846932 | B200:2183424H100:446737 | B200:1647221H100:175535 |
| Concurrency | B200:~128H100:~64 | B200:~128H100:~50 | B200:~93H100:~14 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.