MiniMax M3 428B — B300 vs H100
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H100 (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.
Near the low end of the 15–260 tok/s/user interactivity band, at 76 tok/s/user on MiniMax M3 428B: B300 runs 2502 tok/s/GPU at $0.27/M tokens, H100 runs 537 at $0.67/M. B300 is 147% cheaper per token; B300 delivers 366% more tok/s/GPU.
Setting 138 tok/s/user as the target on MiniMax M3 428B, B300 produces 1104 tok/s/GPU ($0.59 per million tokens) and H100 produces 237 ($1.52). B300 is 157% cheaper per token; B300 delivers 367% more tok/s/GPU.
At 199 tok/s/user interactivity on MiniMax M3 428B, B300 delivers 603 tok/s/GPU at $1.07 per million tokens; H100 delivers 157 tok/s/GPU at $2.31. B300 is 116% cheaper per token; B300 delivers 285% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:2502.2H100:536.7 | B300:1103.7H100:236.6 | B300:603.4H100:156.6 |
| Cost ($/M tok) | B300:$0.272H100:$0.673 | B300:$0.590H100:$1.518 | B300:$1.067H100:$2.307 |
| tok/s/MW | B300:1316956H100:391761 | B300:580912H100:172683 | B300:317574H100:114281 |
| Concurrency | B300:~79H100:~31 | B300:~18H100:~8 | B300:~8H100:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.