DeepSeek V4 Pro 1.6T — GB200 NVL72 vs MI355X
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) on DeepSeek V4 Pro 1.6T. 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.
Throughput at 34 tok/s/user on DeepSeek V4 Pro 1.6T: GB200 NVL72 hits 9617 tok/s/GPU, MI355X hits 2683. Per-million costs land at $0.06 and $0.16 respectively. GB200 NVL72 is 145% cheaper per token; GB200 NVL72 delivers 258% more tok/s/GPU.
GB200 NVL72 / MI355X on DeepSeek V4 Pro 1.6T at 63 tok/s/user: 9177 / 1049 tok/s/GPU, $0.07 / $0.39 per million tokens. GB200 NVL72 is 486% cheaper per token; GB200 NVL72 delivers 775% more tok/s/GPU.
Toward the upper edge of the 6–120 tok/s/user interactivity band, at 92 tok/s/user on DeepSeek V4 Pro 1.6T: GB200 NVL72 runs 7154 tok/s/GPU at $0.09/M tokens, MI355X runs 586 at $0.72/M. GB200 NVL72 is 741% cheaper per token; GB200 NVL72 delivers 1121% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:9616.5MI355X:2683.2 | GB200 NVL72:9177.0MI355X:1049.4 | GB200 NVL72:7154.0MI355X:585.8 |
| Cost ($/M tok) | GB200 NVL72:$0.064MI355X:$0.157 | GB200 NVL72:$0.067MI355X:$0.392 | GB200 NVL72:$0.086MI355X:$0.720 |
| tok/s/MW | GB200 NVL72:5142524MI355X:1283814 | GB200 NVL72:4907501MI355X:502088 | GB200 NVL72:3825657MI355X:280264 |
| Concurrency | GB200 NVL72:~12680MI355X:~79 | GB200 NVL72:~11085MI355X:~16 | GB200 NVL72:~2628MI355X:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.