Kimi K2.5/K2.6/K2.7-Code 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/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.
B200 posts 520 tok/s/GPU for $1.06 per million tokens at 40 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T; MI325X posts 89 tok/s/GPU for $4.03. B200 is 282% cheaper per token; B200 delivers 486% more tok/s/GPU.
Throughput at 44 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B200 hits 465 tok/s/GPU, MI325X hits 71. Per-million costs land at $1.18 and $4.99 respectively. B200 is 323% cheaper per token; B200 delivers 551% more tok/s/GPU.
B200 / MI325X on Kimi K2.5/K2.6/K2.7-Code 1T at 48 tok/s/user: 415 / 59 tok/s/GPU, $1.31 / $6.10 per million tokens. B200 is 366% cheaper per token; B200 delivers 605% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:519.9MI325X:88.7 | B200:465.2MI325X:71.4 | B200:415.2MI325X:58.9 |
| Cost ($/M tok) | B200:$1.055MI325X:$4.028 | B200:$1.178MI325X:$4.986 | B200:$1.309MI325X:$6.102 |
| tok/s/MW | B200:239604MI325X:40676 | B200:214401MI325X:32768 | B200:191317MI325X:27029 |
| Concurrency | B200:~55MI325X:~9 | B200:~45MI325X:~7 | B200:~36MI325X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.