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 40 tok/s/user on Kimi K2.5/K2.6 1T: B200 hits 526 tok/s/GPU, MI325X hits 89. Per-million costs land at $1.04 and $4.03 respectively. B200 is 288% cheaper per token; B200 delivers 493% more tok/s/GPU.
B200 / MI325X on Kimi K2.5/K2.6 1T at 44 tok/s/user: 479 / 71 tok/s/GPU, $1.14 / $4.99 per million tokens. B200 is 337% cheaper per token; B200 delivers 570% more tok/s/GPU.
Toward the upper edge of the 37–51 tok/s/user interactivity band, at 48 tok/s/user on Kimi K2.5/K2.6 1T: B200 runs 434 tok/s/GPU at $1.25/M tokens, MI325X runs 59 at $6.10/M. B200 is 388% cheaper per token; B200 delivers 637% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:526.3MI325X:88.7 | B200:478.9MI325X:71.4 | B200:434.2MI325X:58.9 |
| Cost ($/M tok) | B200:$1.038MI325X:$4.028 | B200:$1.140MI325X:$4.986 | B200:$1.251MI325X:$6.102 |
| tok/s/MW | B200:242512MI325X:40676 | B200:220670MI325X:32768 | B200:200104MI325X:27029 |
| Concurrency | B200:~55MI325X:~9 | B200:~46MI325X:~7 | B200:~38MI325X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.