Kimi K2.5/K2.6 1T — MI325X vs MI355X
Head-to-head AI inference benchmark comparison of MI325X (AMD CDNA 3) 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 22 tok/s/user as the target on Kimi K2.5/K2.6 1T, MI325X produces 147 tok/s/GPU ($2.43 per million tokens) and MI355X produces 103 ($4.04). MI325X is 66% cheaper per token; MI325X delivers 43% more tok/s/GPU.
At 29 tok/s/user interactivity on Kimi K2.5/K2.6 1T, MI325X delivers 138 tok/s/GPU at $2.60 per million tokens; MI355X delivers 63 tok/s/GPU at $6.51. MI325X is 151% cheaper per token; MI325X delivers 118% more tok/s/GPU at this point.
MI325X posts 112 tok/s/GPU for $3.19 per million tokens at 36 tok/s/user on Kimi K2.5/K2.6 1T; MI355X posts 48 tok/s/GPU for $8.46. MI325X is 166% cheaper per token; MI325X delivers 133% more tok/s/GPU. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MI325X:147.1MI355X:102.7 | MI325X:138.0MI355X:63.2 | MI325X:112.0MI355X:48.1 |
| Cost ($/M tok) | MI325X:$2.433MI355X:$4.043 | MI325X:$2.597MI355X:$6.513 | MI325X:$3.187MI355X:$8.464 |
| tok/s/MW | MI325X:67459MI355X:38749 | MI325X:63290MI355X:23831 | MI325X:51373MI355X:18158 |
| Concurrency | MI325X:~27MI355X:~20 | MI325X:~19MI355X:~9 | MI325X:~13MI355X:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.