Qwen 3.5 397B-A17B — B200 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB300 NVL72 (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 98 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 9203 tok/s/chip, GB300 NVL72 hits 35213. Per-million costs land at $0.05 and $0.02 respectively. GB300 NVL72 is 189% cheaper per token; GB300 NVL72 delivers 283% more tok/s/chip.
B200 / GB300 NVL72 on Qwen 3.5 397B-A17B at 172 tok/s/user: 5421 / 16829 tok/s/chip, $0.09 / $0.04 per million tokens. GB300 NVL72 is 133% cheaper per token; GB300 NVL72 delivers 210% more tok/s/chip.
Toward the upper edge of the 24–320 tok/s/user interactivity band, at 246 tok/s/user on Qwen 3.5 397B-A17B: B200 runs 3629 tok/s/chip at $0.13/M tokens, GB300 NVL72 runs 11105 at $0.06/M. GB300 NVL72 is 135% cheaper per token; GB300 NVL72 delivers 206% more tok/s/chip. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:9203.1GB300 NVL72:35212.7 | B200:5420.8GB300 NVL72:16828.9 | B200:3629.5GB300 NVL72:11105.0 |
| Cost ($/M tok) | B200:$0.053GB300 NVL72:$0.018 | B200:$0.089GB300 NVL72:$0.038 | B200:$0.133GB300 NVL72:$0.056 |
| tok/s/MW | B200:5381926GB300 NVL72:16609765 | B200:3170083GB300 NVL72:7938161 | B200:2122508GB300 NVL72:5238209 |
| Concurrency | B200:~96GB300 NVL72:~1181 | B200:~8GB300 NVL72:~393 | B200:~4GB300 NVL72:~145 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.