Llama 3.3 70B — MI325X vs MI355X
Head-to-head AI inference benchmark comparison of MI325X (AMD CDNA 3) and MI355X (AMD CDNA 4) on Llama 3.3 70B. 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.
Near the low end of the 34–116 tok/s/user interactivity band, at 54 tok/s/user on Llama 3.3 70B: MI325X runs 1431 tok/s/GPU at $0.25/M tokens, MI355X runs 2216 at $0.18/M. MI355X is 37% cheaper per token; MI355X delivers 55% more tok/s/GPU.
Setting 75 tok/s/user as the target on Llama 3.3 70B, MI325X produces 841 tok/s/GPU ($0.42 per million tokens) and MI355X produces 1536 ($0.26). MI355X is 59% cheaper per token; MI355X delivers 83% more tok/s/GPU.
At 96 tok/s/user interactivity on Llama 3.3 70B, MI325X delivers 495 tok/s/GPU at $0.70 per million tokens; MI355X delivers 750 tok/s/GPU at $0.55. MI355X is 28% cheaper per token; MI355X delivers 51% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MI325X:1430.7MI355X:2215.8 | MI325X:840.8MI355X:1536.1 | MI325X:495.0MI355X:749.9 |
| Cost ($/M tok) | MI325X:$0.250MI355X:$0.182 | MI325X:$0.416MI355X:$0.261 | MI325X:$0.703MI355X:$0.548 |
| tok/s/MW | MI325X:656293MI355X:836163 | MI325X:385669MI355X:579660 | MI325X:227060MI355X:282963 |
| Concurrency | MI325X:~64MI355X:~50 | MI325X:~53MI355X:~46 | MI325X:~21MI355X:~16 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.