Kimi K2.5/K2.6/K2.7-Code 1T — B200 vs MI355X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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 35–128 tok/s/user interactivity band, at 58 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B200 runs 988 tok/s/GPU at $0.55/M tokens, MI355X runs 930 at $0.44/M. MI355X is 24% cheaper per token; B200 delivers 6% more tok/s/GPU.
Setting 82 tok/s/user as the target on Kimi K2.5/K2.6/K2.7-Code 1T, B200 produces 600 tok/s/GPU ($0.90 per million tokens) and MI355X produces 638 ($0.64). MI355X is 40% cheaper per token; MI355X delivers 6% more tok/s/GPU.
At 105 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, B200 delivers 391 tok/s/GPU at $1.39 per million tokens; MI355X delivers 460 tok/s/GPU at $0.88. MI355X is 58% cheaper per token; MI355X delivers 18% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:987.7MI355X:930.1 | B200:600.0MI355X:637.7 | B200:390.9MI355X:460.2 |
| Cost ($/M tok) | B200:$0.548MI355X:$0.443 | B200:$0.901MI355X:$0.645 | B200:$1.387MI355X:$0.878 |
| tok/s/MW | B200:577628MI355X:445048 | B200:350880MI355X:305139 | B200:228621MI355X:220168 |
| Concurrency | B200:~37MI355X:~33 | B200:~15MI355X:~16 | B200:~8MI355X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.