Kimi K2.5/K2.6 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 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.
Setting 54 tok/s/user as the target on Kimi K2.5/K2.6 1T, B200 produces 1054 tok/s/GPU ($0.52 per million tokens) and MI355X produces 632 ($0.66). B200 is 26% cheaper per token; B200 delivers 67% more tok/s/GPU.
At 74 tok/s/user interactivity on Kimi K2.5/K2.6 1T, B200 delivers 681 tok/s/GPU at $0.79 per million tokens; MI355X delivers 516 tok/s/GPU at $0.80. Cost per token is essentially tied; B200 delivers 32% more tok/s/GPU at this point.
B200 posts 485 tok/s/GPU for $1.08 per million tokens at 94 tok/s/user on Kimi K2.5/K2.6 1T; MI355X posts 396 tok/s/GPU for $1.04. MI355X is 4% cheaper per token; B200 delivers 22% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:1053.8MI355X:631.8 | B200:681.4MI355X:516.1 | B200:485.0MI355X:396.0 |
| Cost ($/M tok) | B200:$0.521MI355X:$0.658 | B200:$0.794MI355X:$0.797 | B200:$1.078MI355X:$1.040 |
| tok/s/MW | B200:485621MI355X:238416 | B200:314019MI355X:194767 | B200:223488MI355X:149436 |
| Concurrency | B200:~42MI355X:~24 | B200:~19MI355X:~14 | B200:~10MI355X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.