Kimi K2.5/K2.6 1T — B200 vs MI325X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) 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 41 tok/s/user on Kimi K2.5/K2.6 1T: B200 hits 543 tok/s/GPU, MI325X hits 84. Per-million costs land at $1.00 and $4.26 respectively. B200 is 324% cheaper per token; B200 delivers 550% more tok/s/GPU.
B200 / MI325X on Kimi K2.5/K2.6 1T at 44 tok/s/user: 506 / 71 tok/s/GPU, $1.08 / $4.99 per million tokens. B200 is 363% cheaper per token; B200 delivers 609% more tok/s/GPU.
Toward the upper edge of the 38–51 tok/s/user interactivity band, at 48 tok/s/user on Kimi K2.5/K2.6 1T: B200 runs 459 tok/s/GPU at $1.18/M tokens, MI325X runs 59 at $6.10/M. B200 is 416% cheaper per token; B200 delivers 679% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:543.5MI325X:83.6 | B200:506.3MI325X:71.4 | B200:458.7MI325X:58.9 |
| Cost ($/M tok) | B200:$1.003MI325X:$4.259 | B200:$1.077MI325X:$4.986 | B200:$1.183MI325X:$6.102 |
| tok/s/MW | B200:250455MI325X:38345 | B200:233311MI325X:32768 | B200:211390MI325X:27029 |
| Concurrency | B200:~56MI325X:~8 | B200:~49MI325X:~7 | B200:~40MI325X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.