MiniMax M3 428B — H100 vs MI300X
Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and MI300X (AMD CDNA 3) on MiniMax M3 428B. 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.
At 52 tok/s/user interactivity on MiniMax M3 428B, H100 delivers 1130 tok/s/GPU at $0.32 per million tokens; MI300X delivers 601 tok/s/GPU at $0.52. H100 is 62% cheaper per token; H100 delivers 88% more tok/s/GPU at this point.
H100 posts 946 tok/s/GPU for $0.38 per million tokens at 82 tok/s/user on MiniMax M3 428B; MI300X posts 494 tok/s/GPU for $0.63. H100 is 65% cheaper per token; H100 delivers 92% more tok/s/GPU.
Throughput at 112 tok/s/user on MiniMax M3 428B: H100 hits 767 tok/s/GPU, MI300X hits 335. Per-million costs land at $0.47 and $0.93 respectively. H100 is 98% cheaper per token; H100 delivers 129% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H100:1130.0MI300X:601.3 | H100:946.0MI300X:493.7 | H100:767.1MI300X:335.4 |
| Cost ($/M tok) | H100:$0.320MI300X:$0.517 | H100:$0.383MI300X:$0.632 | H100:$0.469MI300X:$0.931 |
| tok/s/MW | H100:653183MI300X:335950 | H100:546825MI300X:275812 | H100:443435MI300X:187349 |
| Concurrency | H100:~27MI300X:~14 | H100:~11MI300X:~6 | H100:~7MI300X:~3 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.