Kimi K2.5/K2.6 1T — H200 vs MI325X
Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) and MI325X (AMD CDNA 3) 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.
Throughput at 38 tok/s/user on Kimi K2.5/K2.6 1T: H200 hits 451 tok/s/GPU, MI325X hits 100. Per-million costs land at $0.89 and $3.58 respectively. H200 is 304% cheaper per token; H200 delivers 350% more tok/s/GPU.
H200 / MI325X on Kimi K2.5/K2.6 1T at 42 tok/s/user: 400 / 79 tok/s/GPU, $0.99 / $4.49 per million tokens. H200 is 352% cheaper per token; H200 delivers 405% more tok/s/GPU.
Toward the upper edge of the 34–51 tok/s/user interactivity band, at 47 tok/s/user on Kimi K2.5/K2.6 1T: H200 runs 346 tok/s/GPU at $1.13/M tokens, MI325X runs 62 at $5.81/M. H200 is 415% cheaper per token; H200 delivers 460% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H200:450.8MI325X:100.1 | H200:399.5MI325X:79.2 | H200:346.0MI325X:61.8 |
| Cost ($/M tok) | H200:$0.886MI325X:$3.581 | H200:$0.993MI325X:$4.490 | H200:$1.128MI325X:$5.811 |
| tok/s/MW | H200:260550MI325X:45923 | H200:230947MI325X:36311 | H200:199990MI325X:28337 |
| Concurrency | H200:~50MI325X:~11 | H200:~40MI325X:~8 | H200:~30MI325X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.