Qwen 3.5 397B-A17B — GB300 NVL72 vs MI355X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (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.
GB300 NVL72 / MI355X on Qwen 3.5 397B-A17B at 73 tok/s/user: 8529 / 3448 tok/s/GPU, $0.09 / $0.12 per million tokens. GB300 NVL72 is 41% cheaper per token; GB300 NVL72 delivers 147% more tok/s/GPU.
Around the middle of the 44–161 tok/s/user interactivity band, at 103 tok/s/user on Qwen 3.5 397B-A17B: GB300 NVL72 runs 4678 tok/s/GPU at $0.16/M tokens, MI355X runs 2693 at $0.15/M. MI355X is 5% cheaper per token; GB300 NVL72 delivers 74% more tok/s/GPU.
Setting 132 tok/s/user as the target on Qwen 3.5 397B-A17B, GB300 NVL72 produces 2941 tok/s/GPU ($0.24 per million tokens) and MI355X produces 2090 ($0.20). MI355X is 23% cheaper per token; GB300 NVL72 delivers 41% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:8529.0MI355X:3448.3 | GB300 NVL72:4678.2MI355X:2692.7 | GB300 NVL72:2941.0MI355X:2090.0 |
| Cost ($/M tok) | GB300 NVL72:$0.086MI355X:$0.121 | GB300 NVL72:$0.157MI355X:$0.150 | GB300 NVL72:$0.244MI355X:$0.198 |
| tok/s/MW | GB300 NVL72:4061452MI355X:1301263 | GB300 NVL72:2227701MI355X:1016122 | GB300 NVL72:1400468MI355X:788678 |
| Concurrency | GB300 NVL72:~225MI355X:~14 | GB300 NVL72:~45MI355X:~6 | GB300 NVL72:~23MI355X:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.