MiniMax M2.5/M2.7 — GB300 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) on MiniMax M2.5/M2.7. 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 27–104 tok/s/user interactivity band, at 46 tok/s/user on MiniMax M2.5/M2.7: GB300 NVL72 runs 5802 tok/s/GPU at $0.13/M tokens, MI325X runs 1643 at $0.22/M. GB300 NVL72 is 68% cheaper per token; GB300 NVL72 delivers 253% more tok/s/GPU.
Setting 65 tok/s/user as the target on MiniMax M2.5/M2.7, GB300 NVL72 produces 2996 tok/s/GPU ($0.25 per million tokens) and MI325X produces 757 ($0.48). GB300 NVL72 is 91% cheaper per token; GB300 NVL72 delivers 296% more tok/s/GPU.
At 85 tok/s/user interactivity on MiniMax M2.5/M2.7, GB300 NVL72 delivers 1310 tok/s/GPU at $0.58 per million tokens; MI325X delivers 325 tok/s/GPU at $1.09. GB300 NVL72 is 88% cheaper per token; GB300 NVL72 delivers 303% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:5801.7MI325X:1642.6 | GB300 NVL72:2996.0MI325X:757.2 | GB300 NVL72:1310.0MI325X:324.7 |
| Cost ($/M tok) | GB300 NVL72:$0.129MI325X:$0.217 | GB300 NVL72:$0.251MI325X:$0.479 | GB300 NVL72:$0.582MI325X:$1.094 |
| tok/s/MW | GB300 NVL72:2762692MI325X:753479 | GB300 NVL72:1426674MI325X:347355 | GB300 NVL72:623824MI325X:148934 |
| Concurrency | GB300 NVL72:~593MI325X:~147 | GB300 NVL72:~128MI325X:~38 | GB300 NVL72:~64MI325X:~16 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.