Kimi K2.5/K2.6 1T — H200 vs MI355X
Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) and MI355X (AMD CDNA 4) on Kimi K2.5/K2.6 1T. 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.
Setting 36 tok/s/user as the target on Kimi K2.5/K2.6 1T, H200 produces 478 tok/s/GPU ($0.83 per million tokens) and MI355X produces 48 ($8.46). H200 is 916% cheaper per token; H200 delivers 893% more tok/s/GPU.
At 38 tok/s/user interactivity on Kimi K2.5/K2.6 1T, H200 delivers 451 tok/s/GPU at $0.89 per million tokens; MI355X delivers 46 tok/s/GPU at $8.88. H200 is 902% cheaper per token; H200 delivers 884% more tok/s/GPU at this point.
H200 posts 412 tok/s/GPU for $0.97 per million tokens at 41 tok/s/user on Kimi K2.5/K2.6 1T; MI355X posts 43 tok/s/GPU for $9.45. H200 is 877% cheaper per token; H200 delivers 852% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H200:478.0MI355X:48.1 | H200:450.8MI355X:45.8 | H200:411.8MI355X:43.3 |
| Cost ($/M tok) | H200:$0.833MI355X:$8.464 | H200:$0.886MI355X:$8.883 | H200:$0.967MI355X:$9.448 |
| tok/s/MW | H200:276324MI355X:18158 | H200:260550MI355X:17292 | H200:238033MI355X:16330 |
| Concurrency | H200:~56MI355X:~6 | H200:~50MI355X:~5 | H200:~42MI355X:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.