Qwen 3.5 397B-A17B — GB200 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) 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.
GB200 NVL72 / MI325X on Qwen 3.5 397B-A17B at 44 tok/s/user: 12698 / 902 tok/s/chip, $0.04 / $0.34 per million tokens. GB200 NVL72 is 731% cheaper per token; GB200 NVL72 delivers 1307% more tok/s/chip.
Around the middle of the 36–68 tok/s/user interactivity band, at 52 tok/s/user on Qwen 3.5 397B-A17B: GB200 NVL72 runs 12192 tok/s/chip at $0.04/M tokens, MI325X runs 652 at $0.46/M. GB200 NVL72 is 989% cheaper per token; GB200 NVL72 delivers 1770% more tok/s/chip.
Setting 60 tok/s/user as the target on Qwen 3.5 397B-A17B, GB200 NVL72 produces 11690 tok/s/chip ($0.04 per million tokens) and MI325X produces 456 ($0.68). GB200 NVL72 is 1437% cheaper per token; GB200 NVL72 delivers 2464% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:12697.9MI325X:902.5 | GB200 NVL72:12192.3MI325X:652.2 | GB200 NVL72:11689.8MI325X:455.9 |
| Cost ($/M tok) | GB200 NVL72:$0.041MI325X:$0.338 | GB200 NVL72:$0.042MI325X:$0.462 | GB200 NVL72:$0.044MI325X:$0.679 |
| tok/s/MW | GB200 NVL72:6790311MI325X:534017 | GB200 NVL72:6519957MI325X:385893 | GB200 NVL72:6251204MI325X:269744 |
| Concurrency | GB200 NVL72:~2304MI325X:~20 | GB200 NVL72:~1876MI325X:~12 | GB200 NVL72:~1415MI325X:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.