DeepSeek V4 Pro 1.6T — H200 vs MI355X
Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) 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.
Near the low end of the 1–168 tok/s/user interactivity band, at 43 tok/s/user on DeepSeek V4 Pro 1.6T: H200 runs 219 tok/s/GPU at $1.79/M tokens, MI355X runs 7 at $58.8/M. H200 is 3187% cheaper per token; H200 delivers 3026% more tok/s/GPU.
Setting 85 tok/s/user as the target on DeepSeek V4 Pro 1.6T, H200 produces 98 tok/s/GPU ($4.04 per million tokens) and MI355X produces 7 ($58.8). H200 is 1355% cheaper per token; H200 delivers 1305% more tok/s/GPU.
At 127 tok/s/user interactivity on DeepSeek V4 Pro 1.6T, H200 delivers 62 tok/s/GPU at $6.30 per million tokens; MI355X delivers 7 tok/s/GPU at $58.8. H200 is 832% cheaper per token; H200 delivers 781% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H200:218.7MI355X:7.0 | H200:98.3MI355X:7.0 | H200:61.6MI355X:7.0 |
| Cost ($/M tok) | H200:$1.787MI355X:$58.750 | H200:$4.037MI355X:$58.750 | H200:$6.304MI355X:$58.750 |
| tok/s/MW | H200:126433MI355X:2641 | H200:56813MI355X:2641 | H200:35635MI355X:2641 |
| Concurrency | H200:~21MI355X:~8 | H200:~5MI355X:~8 | H200:~2MI355X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.