Qwen 3.5 397B-A17B — B200 vs MI325X
Head-to-head AI inference benchmark comparison of B200 (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.
B200 / MI325X on Qwen 3.5 397B-A17B at 46 tok/s/user: 4600 / 403 tok/s/GPU, $0.12 / $0.86 per million tokens. B200 is 635% cheaper per token; B200 delivers 1042% 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: B200 runs 4061 tok/s/GPU at $0.13/M tokens, MI325X runs 219 at $1.63/M. B200 is 1131% cheaper per token; B200 delivers 1751% more tok/s/GPU.
Setting 64 tok/s/user as the target on Qwen 3.5 397B-A17B, B200 produces 3529 tok/s/GPU ($0.15 per million tokens) and MI325X produces 134 ($2.64). B200 is 1614% cheaper per token; B200 delivers 2528% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:4600.2MI325X:402.9 | B200:4061.3MI325X:219.4 | B200:3529.1MI325X:134.3 |
| Cost ($/M tok) | B200:$0.118MI325X:$0.864 | B200:$0.133MI325X:$1.631 | B200:$0.154MI325X:$2.638 |
| tok/s/MW | B200:2119899MI325X:184830 | B200:1871590MI325X:100633 | B200:1626333MI325X:61591 |
| Concurrency | B200:~225MI325X:~37 | B200:~166MI325X:~16 | B200:~120MI325X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.