Qwen 3.5 397B-A17B — B200 vs B300
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and B300 (NVIDIA Blackwell) 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 87 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 2247 tok/s/GPU, B300 hits 2162. Per-million costs land at $0.24 and $0.30 respectively. B200 is 26% cheaper per token; B200 delivers 4% more tok/s/GPU.
B200 / B300 on Qwen 3.5 397B-A17B at 143 tok/s/user: 1024 / 1046 tok/s/GPU, $0.53 / $0.62 per million tokens. B200 is 17% cheaper per token; B300 delivers 2% more tok/s/GPU.
Toward the upper edge of the 31–255 tok/s/user interactivity band, at 199 tok/s/user on Qwen 3.5 397B-A17B: B200 runs 642 tok/s/GPU at $0.84/M tokens, B300 runs 676 at $0.96/M. B200 is 14% cheaper per token; B300 delivers 5% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:2247.4B300:2161.9 | B200:1023.5B300:1046.3 | B200:642.4B300:676.4 |
| Cost ($/M tok) | B200:$0.241B300:$0.303 | B200:$0.529B300:$0.620 | B200:$0.838B300:$0.959 |
| tok/s/MW | B200:1035649B300:996248 | B200:471665B300:482152 | B200:296040B300:311711 |
| Concurrency | B200:~58B300:~51 | B200:~16B300:~15 | B200:~7B300:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.