AMD Instinct MI250X meets the minimum VRAM requirement for Q4 inference of Qwen/Qwen2.5-72B-Instruct. Review the quantization breakdown below to see how higher precision settings impact VRAM and throughput.
AMD Instinct MI250X can run Qwen/Qwen2.5-72B-Instruct with Q4 quantization. At approximately 92 tokens/second, you can expect Good speed - acceptable for interactive use.
You have 93GB headroom, which is sufficient for system overhead and smooth operation.
| Quantization | VRAM needed | VRAM available | Estimated speed | Verdict |
|---|---|---|---|---|
| Q4 | 35GB | 128GB | 91.85 tok/s | ✅ Fits comfortably |
| Q8 | 70GB | 128GB | 67.35 tok/s | ✅ Fits comfortably |
| FP16 | 141GB | 128GB | 33.03 tok/s | ❌ Not recommended |