MiniMax M3 428B — H100 vs MI355X
Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and MI355X (AMD CDNA 4) 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.
Throughput at 75 tok/s/user on MiniMax M3 428B: H100 hits 993 tok/s/GPU, MI355X hits 4536. Per-million costs land at $0.36 and $0.09 respectively. MI355X is 302% cheaper per token; MI355X delivers 357% more tok/s/GPU.
H100 / MI355X on MiniMax M3 428B at 129 tok/s/user: 699 / 2944 tok/s/GPU, $0.51 / $0.14 per million tokens. MI355X is 265% cheaper per token; MI355X delivers 321% more tok/s/GPU.
Toward the upper edge of the 22–235 tok/s/user interactivity band, at 182 tok/s/user on MiniMax M3 428B: H100 runs 544 tok/s/GPU at $0.67/M tokens, MI355X runs 1798 at $0.23/M. MI355X is 190% cheaper per token; MI355X delivers 230% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | H100:992.7MI355X:4535.9 | H100:698.6MI355X:2943.9 | H100:544.3MI355X:1798.0 |
| Cost ($/M tok) | H100:$0.365MI355X:$0.091 | H100:$0.513MI355X:$0.140 | H100:$0.669MI355X:$0.230 |
| tok/s/MW | H100:724633MI355X:2170289 | H100:509946MI355X:1408581 | H100:397317MI355X:860283 |
| Concurrency | H100:~13MI355X:~33 | H100:~6MI355X:~12 | H100:~3MI355X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.