MiniMax M2.5/M2.7 — GB300 NVL72 vs H100
Head-to-head AI inference benchmark comparison of GB300 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.
GB300 NVL72 posts 3909 tok/s/chip for $0.17 per million tokens at 60 tok/s/user on MiniMax M2.5/M2.7; H100 posts 599 tok/s/chip for $0.53. GB300 NVL72 is 220% cheaper per token; GB300 NVL72 delivers 552% more tok/s/chip.
Throughput at 79 tok/s/user on MiniMax M2.5/M2.7: GB300 NVL72 hits 1563 tok/s/chip, H100 hits 353. Per-million costs land at $0.43 and $0.92 respectively. GB300 NVL72 is 113% cheaper per token; GB300 NVL72 delivers 343% more tok/s/chip.
GB300 NVL72 / H100 on MiniMax M2.5/M2.7 at 98 tok/s/user: 901 / 207 tok/s/chip, $0.74 / $1.56 per million tokens. GB300 NVL72 is 112% cheaper per token; GB300 NVL72 delivers 335% more tok/s/chip. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:3909.5H100:599.4 | GB300 NVL72:1563.1H100:353.0 | GB300 NVL72:901.3H100:207.1 |
| Cost ($/M tok) | GB300 NVL72:$0.167H100:$0.534 | GB300 NVL72:$0.431H100:$0.920 | GB300 NVL72:$0.736H100:$1.563 |
| tok/s/MW | GB300 NVL72:1844089H100:437502 | GB300 NVL72:737315H100:257637 | GB300 NVL72:425139H100:151188 |
| Concurrency | GB300 NVL72:~380H100:~41 | GB300 NVL72:~64H100:~18 | GB300 NVL72:~32H100:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.