Qwen 3.5 397B-A17B — GB200 NVL72 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of GB200 NVL72 (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 121 tok/s/user on Qwen 3.5 397B-A17B: GB200 NVL72 hits 5636 tok/s/chip, GB300 NVL72 hits 6462. Per-million costs land at $0.10 and $0.10 respectively. GB200 NVL72 is 7% cheaper per token; GB300 NVL72 delivers 15% more tok/s/chip.
GB200 NVL72 / GB300 NVL72 on Qwen 3.5 397B-A17B at 203 tok/s/user: 1828 / 2070 tok/s/chip, $0.28 / $0.31 per million tokens. GB200 NVL72 is 11% cheaper per token; GB300 NVL72 delivers 13% more tok/s/chip.
Toward the upper edge of the 40–365 tok/s/user interactivity band, at 284 tok/s/user on Qwen 3.5 397B-A17B: GB200 NVL72 runs 811 tok/s/chip at $0.60/M tokens, GB300 NVL72 runs 818 at $0.75/M. GB200 NVL72 is 26% cheaper per token; throughput per chip is essentially tied. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:5636.4GB300 NVL72:6461.6 | GB200 NVL72:1828.1GB300 NVL72:2069.9 | GB200 NVL72:811.4GB300 NVL72:818.5 |
| Cost ($/M tok) | GB200 NVL72:$0.095GB300 NVL72:$0.102 | GB200 NVL72:$0.281GB300 NVL72:$0.312 | GB200 NVL72:$0.596GB300 NVL72:$0.749 |
| tok/s/MW | GB200 NVL72:3014110GB300 NVL72:3047939 | GB200 NVL72:977571GB300 NVL72:976389 | GB200 NVL72:433896GB300 NVL72:386081 |
| Concurrency | GB200 NVL72:~191GB300 NVL72:~241 | GB200 NVL72:~17GB300 NVL72:~23 | GB200 NVL72:~4GB300 NVL72:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.