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 10842 tok/s/chip for $0.06 per million tokens at 101 tok/s/user on MiniMax M3 428B; H200 posts 1171 tok/s/chip for $0.29. GB300 NVL72 is 389% cheaper per token; GB300 NVL72 delivers 826% more tok/s/chip.
Throughput at 180 tok/s/user on MiniMax M3 428B: GB300 NVL72 hits 2293 tok/s/chip, H200 hits 833. Per-million costs land at $0.29 and $0.41 respectively. GB300 NVL72 is 41% cheaper per token; GB300 NVL72 delivers 175% more tok/s/chip.
GB300 NVL72 / H200 on MiniMax M3 428B at 260 tok/s/user: 1681 / 382 tok/s/chip, $0.38 / $0.91 per million tokens. GB300 NVL72 is 139% cheaper per token; GB300 NVL72 delivers 340% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:10842.4H200:1171.0 | GB300 NVL72:2292.7H200:833.2 | GB300 NVL72:1681.2H200:381.9 |
| Cost ($/M tok) | GB300 NVL72:$0.059H200:$0.289 | GB300 NVL72:$0.291H200:$0.410 | GB300 NVL72:$0.380H200:$0.908 |
| tok/s/MW | GB300 NVL72:5114337H200:854713 | GB300 NVL72:1081467H200:608171 | GB300 NVL72:793003H200:278730 |
| Concurrency | GB300 NVL72:~257H200:~6 | GB300 NVL72:~12H200:~3 | GB300 NVL72:~8H200:~1 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.