DeepSeek V4 Pro 1.6T — MI300X vs MI355X
Head-to-head AI inference benchmark comparison of MI300X (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.
MI355X: 4286 tok/s/GPU, $0.10 per million tokens at 35 tok/s/user on DeepSeek V4 Pro 1.6T. MI300X is unmeasured here.
At 67 tok/s/user on DeepSeek V4 Pro 1.6T, MI355X delivers 1782 tok/s/GPU at $0.23 per million tokens; MI300X hasn't been benchmarked at this target.
MI355X hits 853 tok/s/GPU for $0.48 per million tokens at 99 tok/s/user on DeepSeek V4 Pro 1.6T. No MI300X data at this operating point. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MI300X:—MI355X:4286.3 | MI300X:—MI355X:1781.8 | MI300X:—MI355X:852.5 |
| Cost ($/M tok) | MI300X:—MI355X:$0.096 | MI300X:—MI355X:$0.230 | MI300X:—MI355X:$0.482 |
| tok/s/MW | MI300X:—MI355X:2050877 | MI300X:—MI355X:852545 | MI300X:—MI355X:407903 |
| Concurrency | MI300X:—MI355X:~512 | MI300X:—MI355X:~34 | MI300X:—MI355X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.