Kimi K2.5/K2.6/K2.7-Code 1T — H200 vs MI300X
Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) and MI300X (AMD CDNA 3) on Kimi K2.5/K2.6/K2.7-Code 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.
Near the low end of the 34–48 tok/s/user interactivity band, at 37 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: H200 runs 465 tok/s/GPU at $0.86/M tokens, MI300X runs 80 at $3.90/M. H200 is 353% cheaper per token; H200 delivers 484% more tok/s/GPU.
Setting 41 tok/s/user as the target on Kimi K2.5/K2.6/K2.7-Code 1T, H200 produces 409 tok/s/GPU ($0.98 per million tokens) and MI300X produces 68 ($4.65). H200 is 375% cheaper per token; H200 delivers 505% more tok/s/GPU.
At 45 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, H200 delivers 361 tok/s/GPU at $1.09 per million tokens; MI300X delivers 57 tok/s/GPU at $5.57. H200 is 411% cheaper per token; H200 delivers 531% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H200:465.0MI300X:79.6 | H200:409.0MI300X:67.6 | H200:360.7MI300X:57.1 |
| Cost ($/M tok) | H200:$0.862MI300X:$3.904 | H200:$0.978MI300X:$4.646 | H200:$1.089MI300X:$5.566 |
| tok/s/MW | H200:268812MI300X:44450 | H200:236439MI300X:37789 | H200:208514MI300X:31923 |
| Concurrency | H200:~53MI300X:~9 | H200:~42MI300X:~7 | H200:~33MI300X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.