Qwen 3.5 397B-A17B — H200 vs MI325X
Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) 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.
H200 / MI325X on Qwen 3.5 397B-A17B at 46 tok/s/user: 523 / 403 tok/s/GPU, $0.74 / $0.86 per million tokens. H200 is 17% cheaper per token; H200 delivers 30% more tok/s/GPU.
Around the middle of the 37–72 tok/s/user interactivity band, at 55 tok/s/user on Qwen 3.5 397B-A17B: H200 runs 463 tok/s/GPU at $0.85/M tokens, MI325X runs 219 at $1.63/M. H200 is 93% cheaper per token; H200 delivers 111% more tok/s/GPU.
Setting 64 tok/s/user as the target on Qwen 3.5 397B-A17B, H200 produces 403 tok/s/GPU ($0.97 per million tokens) and MI325X produces 134 ($2.64). H200 is 171% cheaper per token; H200 delivers 200% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H200:522.9MI325X:402.9 | H200:462.9MI325X:219.4 | H200:402.8MI325X:134.3 |
| Cost ($/M tok) | H200:$0.739MI325X:$0.864 | H200:$0.846MI325X:$1.631 | H200:$0.973MI325X:$2.638 |
| tok/s/MW | H200:302252MI325X:184830 | H200:267580MI325X:100633 | H200:232816MI325X:61591 |
| Concurrency | H200:~45MI325X:~37 | H200:~34MI325X:~16 | H200:~25MI325X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.