MiniMax M2.5/M2.7 · GPU comparison

MiniMax M2.5/M2.7 — GB200 NVL72 vs MI300X

Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI300X (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 27–95 tok/s/user interactivity band, at 44 tok/s/user on MiniMax M2.5/M2.7: GB200 NVL72 runs 5835 tok/s/GPU at $0.11/M tokens, MI300X runs 1249 at $0.25/M. GB200 NVL72 is 138% cheaper per token; GB200 NVL72 delivers 367% more tok/s/GPU.

Setting 61 tok/s/user as the target on MiniMax M2.5/M2.7, GB200 NVL72 produces 3207 tok/s/GPU ($0.19 per million tokens) and MI300X produces 779 ($0.40). GB200 NVL72 is 111% cheaper per token; GB200 NVL72 delivers 312% more tok/s/GPU.

At 78 tok/s/user interactivity on MiniMax M2.5/M2.7, GB200 NVL72 delivers 1320 tok/s/GPU at $0.48 per million tokens; MI300X delivers 412 tok/s/GPU at $0.75. GB200 NVL72 is 55% cheaper per token; GB200 NVL72 delivers 220% 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.)

View performance-per-dollar view →

Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
Metric
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Throughput (tok/s/gpu)
GB200 NVL72:5835.3MI300X:1248.6
GB200 NVL72:3206.8MI300X:778.8
GB200 NVL72:1319.9MI300X:411.9
Cost ($/M tok)
GB200 NVL72:$0.105MI300X:$0.250
GB200 NVL72:$0.191MI300X:$0.402
GB200 NVL72:$0.480MI300X:$0.747
tok/s/MW
GB200 NVL72:2778722MI300X:697553
GB200 NVL72:1527044MI300X:435091
GB200 NVL72:628514MI300X:230105
Concurrency
GB200 NVL72:~1012MI300X:~63
GB200 NVL72:~296MI300X:~26
GB200 NVL72:~64MI300X:~11

Inference Performance

Inference performance metrics across different models, hardware configurations, and serving parameters.