MiniMax M3 428B — GB300 NVL72 vs MI355X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) 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.
Near the low end of the 9–129 tok/s/user interactivity band, at 39 tok/s/user on MiniMax M3 428B: GB300 NVL72 runs 1828 tok/s/GPU at $0.37/M tokens, MI355X runs 1574 at $0.26/M. MI355X is 44% cheaper per token; GB300 NVL72 delivers 16% more tok/s/GPU.
Setting 69 tok/s/user as the target on MiniMax M3 428B, GB300 NVL72 produces 511 tok/s/GPU ($1.43 per million tokens) and MI355X produces 510 ($0.81). MI355X is 78% cheaper per token; throughput per GPU is essentially tied.
At 100 tok/s/user interactivity on MiniMax M3 428B, GB300 NVL72 delivers 235 tok/s/GPU at $3.11 per million tokens; MI355X delivers 183 tok/s/GPU at $2.24. MI355X is 39% cheaper per token; GB300 NVL72 delivers 29% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:1827.9MI355X:1574.0 | GB300 NVL72:511.0MI355X:509.8 | GB300 NVL72:235.1MI355X:182.9 |
| Cost ($/M tok) | GB300 NVL72:$0.371MI355X:$0.258 | GB300 NVL72:$1.432MI355X:$0.806 | GB300 NVL72:$3.106MI355X:$2.239 |
| tok/s/MW | GB300 NVL72:870439MI355X:593952 | GB300 NVL72:243355MI355X:192372 | GB300 NVL72:111947MI355X:69008 |
| Concurrency | GB300 NVL72:~287MI355X:~86 | GB300 NVL72:~30MI355X:~33 | GB300 NVL72:~7MI355X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.