Qwen 3.5 397B-A17B — B300 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B300 (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.
At 92 tok/s/user interactivity on Qwen 3.5 397B-A17B, B300 delivers 4876 tok/s/chip at $0.13 per million tokens; GB200 NVL72 delivers 10226 tok/s/chip at $0.05. GB200 NVL72 is 156% cheaper per token; GB200 NVL72 delivers 110% more tok/s/chip at this point.
B300 posts 3421 tok/s/chip for $0.18 per million tokens at 138 tok/s/user on Qwen 3.5 397B-A17B; GB200 NVL72 posts 5238 tok/s/chip for $0.10. GB200 NVL72 is 82% cheaper per token; GB200 NVL72 delivers 53% more tok/s/chip.
Throughput at 184 tok/s/user on Qwen 3.5 397B-A17B: B300 hits 2557 tok/s/chip, GB200 NVL72 hits 3397. Per-million costs land at $0.25 and $0.15 respectively. GB200 NVL72 is 61% cheaper per token; GB200 NVL72 delivers 33% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:4876.4GB200 NVL72:10225.8 | B300:3421.2GB200 NVL72:5238.4 | B300:2557.0GB200 NVL72:3396.8 |
| Cost ($/M tok) | B300:$0.129GB200 NVL72:$0.050 | B300:$0.182GB200 NVL72:$0.100 | B300:$0.246GB200 NVL72:$0.153 |
| tok/s/MW | B300:2566548GB200 NVL72:5468366 | B300:1800636GB200 NVL72:2801290 | B300:1345811GB200 NVL72:1816453 |
| Concurrency | B300:~25GB200 NVL72:~1042 | B300:~12GB200 NVL72:~133 | B300:~7GB200 NVL72:~41 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.