MiniMax M2.5/M2.7 — GB300 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H200 (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.
Throughput at 49 tok/s/user on MiniMax M2.5/M2.7: GB300 NVL72 hits 4954 tok/s/GPU, H200 hits 1290. Per-million costs land at $0.15 and $0.31 respectively. GB300 NVL72 is 106% cheaper per token; GB300 NVL72 delivers 284% more tok/s/GPU.
GB300 NVL72 / H200 on MiniMax M2.5/M2.7 at 74 tok/s/user: 2117 / 658 tok/s/GPU, $0.35 / $0.59 per million tokens. GB300 NVL72 is 68% cheaper per token; GB300 NVL72 delivers 222% more tok/s/GPU.
Toward the upper edge of the 25–122 tok/s/user interactivity band, at 98 tok/s/user on MiniMax M2.5/M2.7: GB300 NVL72 runs 901 tok/s/GPU at $0.84/M tokens, H200 runs 377 at $1.04/M. GB300 NVL72 is 23% cheaper per token; GB300 NVL72 delivers 139% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:4954.2H200:1290.4 | GB300 NVL72:2116.8H200:657.6 | GB300 NVL72:901.3H200:376.5 |
| Cost ($/M tok) | GB300 NVL72:$0.149H200:$0.306 | GB300 NVL72:$0.354H200:$0.593 | GB300 NVL72:$0.845H200:$1.041 |
| tok/s/MW | GB300 NVL72:2359141H200:745881 | GB300 NVL72:1008006H200:380123 | GB300 NVL72:429188H200:217651 |
| Concurrency | GB300 NVL72:~512H200:~121 | GB300 NVL72:~128H200:~21 | GB300 NVL72:~32H200:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.