MiniMax M3 428B — B200 vs MI355X
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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.
B200 / MI355X on MiniMax M3 428B at 18 tok/s/user: 2384 / 2324 tok/s/GPU, $0.23 / $0.18 per million tokens. MI355X is 28% cheaper per token; B200 delivers 3% more tok/s/GPU.
Around the middle of the 9–46 tok/s/user interactivity band, at 27 tok/s/user on MiniMax M3 428B: B200 runs 1995 tok/s/GPU at $0.27/M tokens, MI355X runs 1934 at $0.21/M. MI355X is 31% cheaper per token; B200 delivers 3% more tok/s/GPU.
Setting 37 tok/s/user as the target on MiniMax M3 428B, B200 produces 1603 tok/s/GPU ($0.33 per million tokens) and MI355X produces 1514 ($0.28). MI355X is 19% cheaper per token; B200 delivers 6% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:2384.2MI355X:2324.1 | B200:1995.1MI355X:1933.9 | B200:1602.5MI355X:1513.8 |
| Cost ($/M tok) | B200:$0.227MI355X:$0.177 | B200:$0.274MI355X:$0.209 | B200:$0.335MI355X:$0.281 |
| tok/s/MW | B200:1098700MI355X:877033 | B200:919410MI355X:729789 | B200:738491MI355X:571247 |
| Concurrency | B200:~256MI355X:~294 | B200:~168MI355X:~158 | B200:~87MI355X:~97 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.