Qwen 3.5 397B-A17B — B200 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB200 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.
B200 posts 5830 tok/s/chip for $0.08 per million tokens at 125 tok/s/user on Qwen 3.5 397B-A17B; GB200 NVL72 posts 8500 tok/s/chip for $0.06. GB200 NVL72 is 35% cheaper per token; GB200 NVL72 delivers 46% more tok/s/chip.
Throughput at 204 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 3746 tok/s/chip, GB200 NVL72 hits 2712. Per-million costs land at $0.13 and $0.19 respectively. B200 is 49% cheaper per token; B200 delivers 38% more tok/s/chip.
B200 / GB200 NVL72 on Qwen 3.5 397B-A17B at 283 tok/s/user: 2500 / 1286 tok/s/chip, $0.19 / $0.39 per million tokens. B200 is 104% cheaper per token; B200 delivers 94% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:5830.5GB200 NVL72:8499.6 | B200:3746.4GB200 NVL72:2712.3 | B200:2500.2GB200 NVL72:1285.8 |
| Cost ($/M tok) | B200:$0.082GB200 NVL72:$0.061 | B200:$0.128GB200 NVL72:$0.191 | B200:$0.191GB200 NVL72:$0.391 |
| tok/s/MW | B200:3409646GB200 NVL72:4545244 | B200:2190873GB200 NVL72:1450449 | B200:1462092GB200 NVL72:687588 |
| Concurrency | B200:~23GB200 NVL72:~558 | B200:~9GB200 NVL72:~30 | B200:~4GB200 NVL72:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.