MiniMax M2.5/M2.7 — H200 vs MI300X
Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) and MI300X (AMD CDNA 3) on MiniMax M2.5/M2.7. 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 27–95 tok/s/user interactivity band, at 44 tok/s/user on MiniMax M2.5/M2.7: H200 runs 1806 tok/s/chip at $0.19/M tokens, MI300X runs 1249 at $0.21/M. H200 is 11% cheaper per token; H200 delivers 45% more tok/s/chip.
Setting 61 tok/s/user as the target on MiniMax M2.5/M2.7, H200 produces 855 tok/s/chip ($0.39 per million tokens) and MI300X produces 779 ($0.34). MI300X is 16% cheaper per token; H200 delivers 10% more tok/s/chip.
At 78 tok/s/user interactivity on MiniMax M2.5/M2.7, H200 delivers 562 tok/s/chip at $0.56 per million tokens; MI300X delivers 412 tok/s/chip at $0.63. H200 is 13% cheaper per token; H200 delivers 37% more tok/s/chip at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | H200:1806.2MI300X:1248.6 | H200:854.9MI300X:778.8 | H200:562.5MI300X:411.9 |
| Cost ($/M tok) | H200:$0.192MI300X:$0.212 | H200:$0.395MI300X:$0.341 | H200:$0.559MI300X:$0.633 |
| tok/s/MW | H200:1318370MI300X:898287 | H200:624041MI300X:560298 | H200:410567MI300X:296322 |
| Concurrency | H200:~83MI300X:~63 | H200:~29MI300X:~26 | H200:~16MI300X:~11 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.