Qwen 3.5 397B-A17B — B300 vs H200
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on Qwen 3.5 397B-A17B. 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 55 tok/s/user on Qwen 3.5 397B-A17B: B300 hits 3643 tok/s/GPU, H200 hits 463. Per-million costs land at $0.18 and $0.85 respectively. B300 is 373% cheaper per token; B300 delivers 687% more tok/s/GPU.
B300 / H200 on Qwen 3.5 397B-A17B at 81 tok/s/user: 2395 / 319 tok/s/GPU, $0.27 / $1.23 per million tokens. B300 is 353% cheaper per token; B300 delivers 651% more tok/s/GPU.
Toward the upper edge of the 30–132 tok/s/user interactivity band, at 107 tok/s/user on Qwen 3.5 397B-A17B: B300 runs 1535 tok/s/GPU at $0.42/M tokens, H200 runs 271 at $1.45/M. B300 is 243% cheaper per token; B300 delivers 466% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:3642.8H200:462.9 | B300:2394.8H200:318.8 | B300:1534.9H200:271.4 |
| Cost ($/M tok) | B300:$0.179H200:$0.846 | B300:$0.272H200:$1.229 | B300:$0.421H200:$1.446 |
| tok/s/MW | B300:1678708H200:267580 | B300:1103604H200:184266 | B300:707349H200:156874 |
| Concurrency | B300:~140H200:~34 | B300:~61H200:~16 | B300:~30H200:~12 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.