MiniMax M3 428B — MI300X vs MI325X
Head-to-head AI inference benchmark comparison of MI300X (AMD CDNA 3) and MI325X (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.
MI300X / MI325X on MiniMax M3 428B at 55 tok/s/user: 362 / 513 tok/s/GPU, $0.87 / $0.69 per million tokens. MI325X is 25% cheaper per token; MI325X delivers 42% more tok/s/GPU.
Around the middle of the 13–182 tok/s/user interactivity band, at 97 tok/s/user on MiniMax M3 428B: MI300X runs 168 tok/s/GPU at $1.85/M tokens, MI325X runs 203 at $1.75/M. MI325X is 6% cheaper per token; MI325X delivers 21% more tok/s/GPU.
Setting 140 tok/s/user as the target on MiniMax M3 428B, MI300X produces 115 tok/s/GPU ($2.71 per million tokens) and MI325X produces 113 ($3.13). MI300X is 15% cheaper per token; MI300X delivers 1% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MI300X:361.9MI325X:512.9 | MI300X:168.1MI325X:203.0 | MI300X:114.6MI325X:113.3 |
| Cost ($/M tok) | MI300X:$0.871MI325X:$0.695 | MI300X:$1.847MI325X:$1.750 | MI300X:$2.709MI325X:$3.128 |
| tok/s/MW | MI300X:202206MI325X:235281 | MI300X:93909MI325X:93105 | MI300X:64030MI325X:51985 |
| Concurrency | MI300X:~29MI325X:~41 | MI300X:~7MI325X:~9 | MI300X:~4MI325X:~3 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.