MiniMax M2.5/M2.7 — H100 vs MI325X
Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and MI325X (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 41–104 tok/s/user interactivity band, at 57 tok/s/user on MiniMax M2.5/M2.7: H100 runs 650 tok/s/chip at $0.50/M tokens, MI325X runs 1066 at $0.29/M. MI325X is 75% cheaper per token; MI325X delivers 64% more tok/s/chip.
Setting 73 tok/s/user as the target on MiniMax M2.5/M2.7, H100 produces 417 tok/s/chip ($0.78 per million tokens) and MI325X produces 568 ($0.54). MI325X is 45% cheaper per token; MI325X delivers 36% more tok/s/chip.
At 89 tok/s/user interactivity on MiniMax M2.5/M2.7, H100 delivers 265 tok/s/chip at $1.23 per million tokens; MI325X delivers 271 tok/s/chip at $1.13. MI325X is 9% cheaper per token; MI325X delivers 2% more tok/s/chip at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | H100:649.6MI325X:1066.4 | H100:417.2MI325X:568.0 | H100:265.3MI325X:271.1 |
| Cost ($/M tok) | H100:$0.500MI325X:$0.287 | H100:$0.779MI325X:$0.538 | H100:$1.225MI325X:$1.127 |
| tok/s/MW | H100:474158MI325X:631033 | H100:304528MI325X:336124 | H100:193629MI325X:160430 |
| Concurrency | H100:~46MI325X:~79 | H100:~23MI325X:~8 | H100:~12MI325X:~12 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.