MiniMax M2.5/M2.7 — GB200 NVL72 vs H100
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and H100 (NVIDIA Hopper) 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.
At 59 tok/s/user interactivity on MiniMax M2.5/M2.7, GB200 NVL72 delivers 3310 tok/s/GPU at $0.18 per million tokens; H100 delivers 616 tok/s/GPU at $0.58. GB200 NVL72 is 213% cheaper per token; GB200 NVL72 delivers 437% more tok/s/GPU at this point.
GB200 NVL72 posts 1320 tok/s/GPU for $0.48 per million tokens at 78 tok/s/user on MiniMax M2.5/M2.7; H100 posts 380 tok/s/GPU for $0.94. GB200 NVL72 is 95% cheaper per token; GB200 NVL72 delivers 248% more tok/s/GPU.
Throughput at 98 tok/s/user on MiniMax M2.5/M2.7: GB200 NVL72 hits 761 tok/s/GPU, H100 hits 210. Per-million costs land at $0.82 and $1.72 respectively. GB200 NVL72 is 108% cheaper per token; GB200 NVL72 delivers 262% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:3310.2H100:616.0 | GB200 NVL72:1319.9H100:379.6 | GB200 NVL72:760.8H100:210.4 |
| Cost ($/M tok) | GB200 NVL72:$0.184H100:$0.576 | GB200 NVL72:$0.480H100:$0.936 | GB200 NVL72:$0.825H100:$1.717 |
| tok/s/MW | GB200 NVL72:1576291H100:356072 | GB200 NVL72:628514H100:219420 | GB200 NVL72:362304H100:121604 |
| Concurrency | GB200 NVL72:~340H100:~43 | GB200 NVL72:~64H100:~12 | GB200 NVL72:~32H100:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.