MiniMax M3 428B · GPU comparison

MiniMax M3 428B — B300 vs H200

Head-to-head AI inference benchmark comparison of B300 (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.

Setting 82 tok/s/user as the target on MiniMax M3 428B, B300 produces 2017 tok/s/GPU ($0.33 per million tokens) and H200 produces 666 ($0.58). B300 is 80% cheaper per token; B300 delivers 203% more tok/s/GPU.

At 153 tok/s/user interactivity on MiniMax M3 428B, B300 delivers 1031 tok/s/GPU at $0.63 per million tokens; H200 delivers 331 tok/s/GPU at $1.17. B300 is 85% cheaper per token; B300 delivers 212% more tok/s/GPU at this point.

B300 posts 495 tok/s/GPU for $1.31 per million tokens at 223 tok/s/user on MiniMax M3 428B; H200 posts 166 tok/s/GPU for $2.43. B300 is 86% cheaper per token; B300 delivers 198% 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.)

View performance-per-dollar view →

Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
Metric
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Throughput (tok/s/gpu)
B300:2016.9H200:666.5
B300:1031.0H200:330.7
B300:495.1H200:166.1
Cost ($/M tok)
B300:$0.325H200:$0.584
B300:$0.632H200:$1.169
B300:$1.309H200:$2.429
tok/s/MW
B300:1061549H200:486476
B300:542654H200:241422
B300:260569H200:121223
Concurrency
B300:~68H200:~18
B300:~14H200:~4
B300:~8H200:~3

Inference Performance

Inference performance metrics across different models, hardware configurations, and serving parameters.

Vendor:
Aggregation:
Spec Decoding: