MiniMax M3 428B — GB200 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI325X (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.
Throughput at 52 tok/s/user on MiniMax M3 428B: GB200 NVL72 hits 1598 tok/s/GPU, MI325X hits 557. Per-million costs land at $0.38 and $0.64 respectively. GB200 NVL72 is 67% cheaper per token; GB200 NVL72 delivers 187% more tok/s/GPU.
GB200 NVL72 / MI325X on MiniMax M3 428B at 87 tok/s/user: 638 / 234 tok/s/GPU, $1.00 / $1.49 per million tokens. GB200 NVL72 is 50% cheaper per token; GB200 NVL72 delivers 173% more tok/s/GPU.
Toward the upper edge of the 17–157 tok/s/user interactivity band, at 122 tok/s/user on MiniMax M3 428B: GB200 NVL72 runs 211 tok/s/GPU at $2.86/M tokens, MI325X runs 143 at $2.45/M. MI325X is 17% cheaper per token; GB200 NVL72 delivers 47% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:1598.2MI325X:556.9 | GB200 NVL72:638.3MI325X:234.1 | GB200 NVL72:210.9MI325X:143.1 |
| Cost ($/M tok) | GB200 NVL72:$0.382MI325X:$0.640 | GB200 NVL72:$0.996MI325X:$1.494 | GB200 NVL72:$2.862MI325X:$2.455 |
| tok/s/MW | GB200 NVL72:761024MI325X:255452 | GB200 NVL72:303973MI325X:107394 | GB200 NVL72:100447MI325X:65652 |
| Concurrency | GB200 NVL72:~180MI325X:~47 | GB200 NVL72:~74MI325X:~11 | GB200 NVL72:~18MI325X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.