Qwen 3.5 397B-A17B — GB300 NVL72 vs RTX PRO 6000
Head-to-head AI inference benchmark comparison of GB300 NVL72 (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.
GB300 NVL72 / RTX PRO 6000 on Qwen 3.5 397B-A17B at 56 tok/s/user: 37920 / 973 tok/s/chip, $0.02 / $0.18 per million tokens. GB300 NVL72 is 989% cheaper per token; GB300 NVL72 delivers 3797% more tok/s/chip.
Around the middle of the 24–153 tok/s/user interactivity band, at 88 tok/s/user on Qwen 3.5 397B-A17B: GB300 NVL72 runs 36496 tok/s/chip at $0.02/M tokens, RTX PRO 6000 runs 756 at $0.25/M. GB300 NVL72 is 1305% cheaper per token; GB300 NVL72 delivers 4725% more tok/s/chip.
Setting 121 tok/s/user as the target on Qwen 3.5 397B-A17B, GB300 NVL72 produces 29858 tok/s/chip ($0.02 per million tokens) and RTX PRO 6000 produces 518 ($0.40). GB300 NVL72 is 1756% cheaper per token; GB300 NVL72 delivers 5666% more tok/s/chip. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:37920.2RTX PRO 6000:973.1 | GB300 NVL72:36496.2RTX PRO 6000:756.3 | GB300 NVL72:29857.6RTX PRO 6000:517.8 |
| Cost ($/M tok) | GB300 NVL72:$0.017RTX PRO 6000:$0.184 | GB300 NVL72:$0.018RTX PRO 6000:$0.247 | GB300 NVL72:$0.022RTX PRO 6000:$0.400 |
| tok/s/MW | GB300 NVL72:17886892RTX PRO 6000:998045 | GB300 NVL72:17215211RTX PRO 6000:775731 | GB300 NVL72:14083781RTX PRO 6000:531118 |
| Concurrency | GB300 NVL72:~2983RTX PRO 6000:~10 | GB300 NVL72:~1205RTX PRO 6000:~4 | GB300 NVL72:~1088RTX PRO 6000:~2 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.