Qwen 3.5 397B-A17B — B300 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B300 (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.
B300 / GB300 NVL72 on Qwen 3.5 397B-A17B at 112 tok/s/user: 10448 / 32371 tok/s/chip, $0.06 / $0.02 per million tokens. GB300 NVL72 is 202% cheaper per token; GB300 NVL72 delivers 210% more tok/s/chip.
Around the middle of the 28–364 tok/s/user interactivity band, at 196 tok/s/user on Qwen 3.5 397B-A17B: B300 runs 6077 tok/s/chip at $0.10/M tokens, GB300 NVL72 runs 14473 at $0.04/M. GB300 NVL72 is 134% cheaper per token; GB300 NVL72 delivers 138% more tok/s/chip.
Setting 280 tok/s/user as the target on Qwen 3.5 397B-A17B, B300 produces 4523 tok/s/chip ($0.14 per million tokens) and GB300 NVL72 produces 9831 ($0.06). GB300 NVL72 is 121% cheaper per token; GB300 NVL72 delivers 117% more tok/s/chip. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:10448.1GB300 NVL72:32370.9 | B300:6076.7GB300 NVL72:14472.9 | B300:4522.8GB300 NVL72:9831.2 |
| Cost ($/M tok) | B300:$0.060GB300 NVL72:$0.020 | B300:$0.104GB300 NVL72:$0.044 | B300:$0.139GB300 NVL72:$0.063 |
| tok/s/MW | B300:5499026GB300 NVL72:15269282 | B300:3198243GB300 NVL72:6826840 | B300:2380421GB300 NVL72:4637358 |
| Concurrency | B300:~22GB300 NVL72:~1151 | B300:~16GB300 NVL72:~243 | B300:~4GB300 NVL72:~135 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.