MiniMax M3 428B — GB300 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on MiniMax M3 428B. 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 3034 tok/s/GPU for $0.24 per million tokens at 43 tok/s/user on MiniMax M3 428B; H200 posts 1229 tok/s/GPU for $0.32. GB300 NVL72 is 33% cheaper per token; GB300 NVL72 delivers 147% more tok/s/GPU.
Throughput at 82 tok/s/user on MiniMax M3 428B: GB300 NVL72 hits 794 tok/s/GPU, H200 hits 666. Per-million costs land at $0.94 and $0.58 respectively. H200 is 62% cheaper per token; GB300 NVL72 delivers 19% more tok/s/GPU.
GB300 NVL72 / H200 on MiniMax M3 428B at 120 tok/s/user: 274 / 419 tok/s/GPU, $2.64 / $0.93 per million tokens. H200 is 183% cheaper per token; H200 delivers 53% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:3033.8H200:1229.0 | GB300 NVL72:794.2H200:666.5 | GB300 NVL72:274.4H200:418.6 |
| Cost ($/M tok) | GB300 NVL72:$0.239H200:$0.319 | GB300 NVL72:$0.943H200:$0.584 | GB300 NVL72:$2.644H200:$0.934 |
| tok/s/MW | GB300 NVL72:1431023H200:897083 | GB300 NVL72:374635H200:486476 | GB300 NVL72:129430H200:305548 |
| Concurrency | GB300 NVL72:~883H200:~63 | GB300 NVL72:~94H200:~18 | GB300 NVL72:~24H200:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.