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 1–181 tok/s/user interactivity band, at 46 tok/s/user on DeepSeek V4 Pro 1.6T: H200 runs 222 tok/s/GPU at $1.72/M tokens, MI355X runs 7 at $58.8/M. H200 is 3309% cheaper per token; H200 delivers 3071% more tok/s/GPU.
Setting 91 tok/s/user as the target on DeepSeek V4 Pro 1.6T, H200 produces 84 tok/s/GPU ($4.67 per million tokens) and MI355X produces 7 ($58.8). H200 is 1159% cheaper per token; H200 delivers 1098% more tok/s/GPU.
At 136 tok/s/user interactivity on DeepSeek V4 Pro 1.6T, H200 delivers 70 tok/s/GPU at $5.62 per million tokens; MI355X delivers 7 tok/s/GPU at $58.8. H200 is 945% cheaper per token; H200 delivers 899% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H200:221.9MI355X:7.0 | H200:83.9MI355X:7.0 | H200:69.9MI355X:7.0 |
| Cost ($/M tok) | H200:$1.723MI355X:$58.750 | H200:$4.667MI355X:$58.750 | H200:$5.622MI355X:$58.750 |
| tok/s/MW | H200:161974MI355X:3348 | H200:61212MI355X:3348 | H200:51018MI355X:3348 |
| Concurrency | H200:~21MI355X:~8 | H200:~4MI355X:~8 | H200:~2MI355X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.