Kimi K2.5/K2.6/K2.7-Code 1T — B200 vs MI300X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) 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.
Throughput at 40 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B200 hits 520 tok/s/GPU, MI300X hits 70. Per-million costs land at $1.06 and $4.44 respectively. B200 is 321% cheaper per token; B200 delivers 638% more tok/s/GPU.
B200 / MI300X on Kimi K2.5/K2.6/K2.7-Code 1T at 43 tok/s/user: 479 / 62 tok/s/GPU, $1.15 / $5.09 per million tokens. B200 is 344% cheaper per token; B200 delivers 669% more tok/s/GPU.
Toward the upper edge of the 37–48 tok/s/user interactivity band, at 46 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B200 runs 439 tok/s/GPU at $1.24/M tokens, MI300X runs 55 at $5.81/M. B200 is 368% cheaper per token; B200 delivers 704% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:519.9MI300X:70.4 | B200:478.6MI300X:62.3 | B200:439.5MI300X:54.6 |
| Cost ($/M tok) | B200:$1.055MI300X:$4.441 | B200:$1.147MI300X:$5.089 | B200:$1.242MI300X:$5.813 |
| tok/s/MW | B200:239604MI300X:39352 | B200:220547MI300X:34789 | B200:202524MI300X:30525 |
| Concurrency | B200:~55MI300X:~7 | B200:~47MI300X:~6 | B200:~40MI300X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.