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 85 tok/s/user: 11010 / 3894 tok/s/chip, $0.06 / $0.11 per million tokens. GB300 NVL72 is 84% cheaper per token; GB300 NVL72 delivers 183% more tok/s/chip.
Around the middle of the 40–222 tok/s/user interactivity band, at 131 tok/s/user on Qwen 3.5 397B-A17B: GB300 NVL72 runs 5267 tok/s/chip at $0.12/M tokens, MI355X runs 2671 at $0.15/M. GB300 NVL72 is 23% cheaper per token; GB300 NVL72 delivers 97% more tok/s/chip.
Setting 177 tok/s/user as the target on Qwen 3.5 397B-A17B, GB300 NVL72 produces 2887 tok/s/chip ($0.22 per million tokens) and MI355X produces 1660 ($0.25). GB300 NVL72 is 13% cheaper per token; GB300 NVL72 delivers 74% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:11009.8MI355X:3893.7 | GB300 NVL72:5266.8MI355X:2670.5 | GB300 NVL72:2886.9MI355X:1659.9 |
| Cost ($/M tok) | GB300 NVL72:$0.058MI355X:$0.107 | GB300 NVL72:$0.125MI355X:$0.154 | GB300 NVL72:$0.222MI355X:$0.250 |
| tok/s/MW | GB300 NVL72:5193302MI355X:1862998 | GB300 NVL72:2484317MI355X:1277761 | GB300 NVL72:1361762MI355X:794204 |
| Concurrency | GB300 NVL72:~1158MI355X:~21 | GB300 NVL72:~122MI355X:~9 | GB300 NVL72:~40MI355X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.