Qwen 3.5 397B-A17B — H100 vs MI355X
Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and MI355X (AMD CDNA 4) 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.
Setting 64 tok/s/user as the target on Qwen 3.5 397B-A17B, H100 produces 433 tok/s/GPU ($0.85 per million tokens) and MI355X produces 2131 ($0.19). MI355X is 340% cheaper per token; MI355X delivers 392% more tok/s/GPU.
At 100 tok/s/user interactivity on Qwen 3.5 397B-A17B, H100 delivers 311 tok/s/GPU at $1.15 per million tokens; MI355X delivers 1231 tok/s/GPU at $0.33. MI355X is 252% cheaper per token; MI355X delivers 296% more tok/s/GPU at this point.
H100 posts 236 tok/s/GPU for $1.54 per million tokens at 135 tok/s/user on Qwen 3.5 397B-A17B; MI355X posts 893 tok/s/GPU for $0.46. MI355X is 232% cheaper per token; MI355X delivers 278% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H100:433.3MI355X:2131.3 | H100:310.5MI355X:1230.6 | H100:236.1MI355X:892.7 |
| Cost ($/M tok) | H100:$0.848MI355X:$0.193 | H100:$1.151MI355X:$0.327 | H100:$1.536MI355X:$0.463 |
| tok/s/MW | H100:250452MI355X:804277 | H100:179506MI355X:464371 | H100:136470MI355X:336868 |
| Concurrency | H100:~28MI355X:~68 | H100:~13MI355X:~25 | H100:~7MI355X:~13 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.