Kimi K2.5/K2.6 1T — MI300X vs MI355X
Head-to-head AI inference benchmark comparison of MI300X (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.
Near the low end of the 14–43 tok/s/user interactivity band, at 21 tok/s/user on Kimi K2.5/K2.6 1T: MI300X runs 121 tok/s/GPU at $2.56/M tokens, MI355X runs 111 at $3.78/M. MI300X is 48% cheaper per token; MI300X delivers 9% more tok/s/GPU.
Setting 28 tok/s/user as the target on Kimi K2.5/K2.6 1T, MI300X produces 110 tok/s/GPU ($2.83 per million tokens) and MI355X produces 67 ($6.14). MI300X is 117% cheaper per token; MI300X delivers 64% more tok/s/GPU.
At 36 tok/s/user interactivity on Kimi K2.5/K2.6 1T, MI300X delivers 83 tok/s/GPU at $3.75 per million tokens; MI355X delivers 48 tok/s/GPU at $8.46. MI300X is 126% cheaper per token; MI300X delivers 72% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MI300X:121.4MI355X:110.9 | MI300X:110.0MI355X:67.0 | MI300X:83.0MI355X:48.1 |
| Cost ($/M tok) | MI300X:$2.557MI355X:$3.778 | MI300X:$2.829MI355X:$6.139 | MI300X:$3.748MI355X:$8.464 |
| tok/s/MW | MI300X:67843MI355X:41855 | MI300X:61428MI355X:25293 | MI300X:46350MI355X:18158 |
| Concurrency | MI300X:~23MI355X:~23 | MI300X:~16MI355X:~10 | MI300X:~10MI355X:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.