MiniMax M3 428B — B300 vs MI300X
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and MI300X (AMD CDNA 3) 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.
At 47 tok/s/user interactivity on MiniMax M3 428B, B300 delivers 3503 tok/s/GPU at $0.19 per million tokens; MI300X delivers 517 tok/s/GPU at $0.59. B300 is 218% cheaper per token; B300 delivers 577% more tok/s/GPU at this point.
B300 posts 2086 tok/s/GPU for $0.32 per million tokens at 81 tok/s/user on MiniMax M3 428B; MI300X posts 202 tok/s/GPU for $1.53. B300 is 384% cheaper per token; B300 delivers 932% more tok/s/GPU.
Throughput at 115 tok/s/user on MiniMax M3 428B: B300 hits 1143 tok/s/GPU, MI300X hits 142. Per-million costs land at $0.57 and $2.14 respectively. B300 is 276% cheaper per token; B300 delivers 703% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:3502.6MI300X:517.4 | B300:2085.7MI300X:202.2 | B300:1142.6MI300X:142.4 |
| Cost ($/M tok) | B300:$0.186MI300X:$0.590 | B300:$0.317MI300X:$1.534 | B300:$0.570MI300X:$2.145 |
| tok/s/MW | B300:1843451MI300X:372210 | B300:1097711MI300X:145436 | B300:601372MI300X:102431 |
| Concurrency | B300:~470MI300X:~47 | B300:~70MI300X:~11 | B300:~27MI300X:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.