MiniMax M3 428B — GB300 NVL72 vs MI300X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) 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.
Throughput at 56 tok/s/user on MiniMax M3 428B: GB300 NVL72 hits 16377 tok/s/chip, MI300X hits 668. Per-million costs land at $0.04 and $0.40 respectively. GB300 NVL72 is 911% cheaper per token; GB300 NVL72 delivers 2350% more tok/s/chip.
GB300 NVL72 / MI300X on MiniMax M3 428B at 92 tok/s/user: 12575 / 530 tok/s/chip, $0.05 / $0.50 per million tokens. GB300 NVL72 is 881% cheaper per token; GB300 NVL72 delivers 2272% more tok/s/chip.
Toward the upper edge of the 22–162 tok/s/user interactivity band, at 127 tok/s/user on MiniMax M3 428B: GB300 NVL72 runs 4476 tok/s/chip at $0.16/M tokens, MI300X runs 339 at $0.76/M. GB300 NVL72 is 388% cheaper per token; GB300 NVL72 delivers 1220% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:16377.2MI300X:668.4 | GB300 NVL72:12575.2MI300X:530.1 | GB300 NVL72:4475.7MI300X:339.0 |
| Cost ($/M tok) | GB300 NVL72:$0.039MI300X:$0.397 | GB300 NVL72:$0.051MI300X:$0.501 | GB300 NVL72:$0.156MI300X:$0.761 |
| tok/s/MW | GB300 NVL72:7725089MI300X:480849 | GB300 NVL72:5931695MI300X:381341 | GB300 NVL72:2111157MI300X:243899 |
| Concurrency | GB300 NVL72:~2048MI300X:~15 | GB300 NVL72:~776MI300X:~6 | GB300 NVL72:~57MI300X:~3 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.