Qwen 3.5 397B-A17B — B200 vs MI355X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) 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.
At 63 tok/s/user interactivity on Qwen 3.5 397B-A17B, B200 delivers 8835 tok/s/chip at $0.05 per million tokens; MI355X delivers 5111 tok/s/chip at $0.08. B200 is 50% cheaper per token; B200 delivers 73% more tok/s/chip at this point.
B200 posts 6170 tok/s/chip for $0.08 per million tokens at 116 tok/s/user on Qwen 3.5 397B-A17B; MI355X posts 3258 tok/s/chip for $0.13. B200 is 63% cheaper per token; B200 delivers 89% more tok/s/chip.
Throughput at 169 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 4485 tok/s/chip, MI355X hits 2280. Per-million costs land at $0.11 and $0.18 respectively. B200 is 73% cheaper per token; B200 delivers 97% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:8834.8MI355X:5110.9 | B200:6170.1MI355X:3257.5 | B200:4485.1MI355X:2279.5 |
| Cost ($/M tok) | B200:$0.054MI355X:$0.081 | B200:$0.078MI355X:$0.127 | B200:$0.107MI355X:$0.184 |
| tok/s/MW | B200:5166571MI355X:2445405 | B200:3608252MI355X:1558615 | B200:2622839MI355X:1090670 |
| Concurrency | B200:~68MI355X:~39 | B200:~26MI355X:~14 | B200:~13MI355X:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.