DeepSeek V4 Pro 1.6T — MI325X vs MI355X
Head-to-head AI inference benchmark comparison of MI325X (AMD CDNA 3) and MI355X (AMD CDNA 4) on DeepSeek V4 Pro 1.6T. 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 35 tok/s/user on DeepSeek V4 Pro 1.6T, MI355X delivers 4286 tok/s/GPU at $0.10 per million tokens; MI325X hasn't been benchmarked at this target.
MI355X hits 1782 tok/s/GPU for $0.23 per million tokens at 67 tok/s/user on DeepSeek V4 Pro 1.6T. No MI325X data at this operating point.
MI355X: 853 tok/s/GPU, $0.48 per million tokens at 99 tok/s/user on DeepSeek V4 Pro 1.6T. MI325X is unmeasured here. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MI325X:—MI355X:4286.3 | MI325X:—MI355X:1781.8 | MI325X:—MI355X:852.5 |
| Cost ($/M tok) | MI325X:—MI355X:$0.096 | MI325X:—MI355X:$0.230 | MI325X:—MI355X:$0.482 |
| tok/s/MW | MI325X:—MI355X:2050877 | MI325X:—MI355X:852545 | MI325X:—MI355X:407903 |
| Concurrency | MI325X:—MI355X:~512 | MI325X:—MI355X:~34 | MI325X:—MI355X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.