MiniMax M3 428B — B300 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) 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.
Throughput at 125 tok/s/user on MiniMax M3 428B: B300 hits 3753 tok/s/chip, GB300 NVL72 hits 4610. Per-million costs land at $0.17 and $0.15 respectively. GB300 NVL72 is 14% cheaper per token; GB300 NVL72 delivers 23% more tok/s/chip.
B300 / GB300 NVL72 on MiniMax M3 428B at 229 tok/s/user: 2283 / 1782 tok/s/chip, $0.27 / $0.36 per million tokens. B300 is 32% cheaper per token; B300 delivers 28% more tok/s/chip.
Toward the upper edge of the 22–436 tok/s/user interactivity band, at 333 tok/s/user on MiniMax M3 428B: B300 runs 1074 tok/s/chip at $0.60/M tokens, GB300 NVL72 runs 1030 at $0.62/M. B300 is 2% cheaper per token; B300 delivers 4% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:3752.7GB300 NVL72:4610.3 | B300:2282.9GB300 NVL72:1782.4 | B300:1074.0GB300 NVL72:1030.4 |
| Cost ($/M tok) | B300:$0.167GB300 NVL72:$0.147 | B300:$0.272GB300 NVL72:$0.358 | B300:$0.604GB300 NVL72:$0.619 |
| tok/s/MW | B300:1975113GB300 NVL72:2174679 | B300:1201503GB300 NVL72:840769 | B300:565268GB300 NVL72:486019 |
| Concurrency | B300:~16GB300 NVL72:~60 | B300:~5GB300 NVL72:~8 | B300:~4GB300 NVL72:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.