Qwen 3.5 397B-A17B — B200 vs RTX PRO 6000
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and RTX PRO 6000 (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.
Near the low end of the 11–153 tok/s/user interactivity band, at 47 tok/s/user on Qwen 3.5 397B-A17B: B200 runs 13815 tok/s/chip at $0.03/M tokens, RTX PRO 6000 runs 1069 at $0.17/M. B200 is 394% cheaper per token; B200 delivers 1192% more tok/s/chip.
Setting 82 tok/s/user as the target on Qwen 3.5 397B-A17B, B200 produces 10097 tok/s/chip ($0.05 per million tokens) and RTX PRO 6000 produces 796 ($0.23). B200 is 383% cheaper per token; B200 delivers 1168% more tok/s/chip.
At 118 tok/s/user interactivity on Qwen 3.5 397B-A17B, B200 delivers 7667 tok/s/chip at $0.06 per million tokens; RTX PRO 6000 delivers 538 tok/s/chip at $0.38. B200 is 511% cheaper per token; B200 delivers 1325% more tok/s/chip at this point. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:13814.7RTX PRO 6000:1069.0 | B200:10097.1RTX PRO 6000:796.2 | B200:7667.4RTX PRO 6000:537.9 |
| Cost ($/M tok) | B200:$0.035RTX PRO 6000:$0.172 | B200:$0.048RTX PRO 6000:$0.230 | B200:$0.063RTX PRO 6000:$0.383 |
| tok/s/MW | B200:8078776RTX PRO 6000:1096391 | B200:5904717RTX PRO 6000:816652 | B200:4483837RTX PRO 6000:551701 |
| Concurrency | B200:~104RTX PRO 6000:~12 | B200:~73RTX PRO 6000:~5 | B200:~16RTX PRO 6000:~2 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.