Welche GPU für Qwen 2.5 Coder 32B lokal?
Qwen 2.5 Coder 32B hat 32 Mrd. Parameter. Diese GPUs können das Modell in mindestens einer praktikablen Quantisierung fahren — sortiert nach Preis pro GB VRAM.
| GPU | beste Quant | VRAM-Bedarf | Tokens/Sek | Preis | €/GB |
|---|---|---|---|---|---|
| AMD Radeon RX 7900 XTX | Q4_K_M | 20.9 GB | — | — | — |
| Apple Mac mini M4 Pro 64GB | Q8_0 | 34.5 GB | — | — | — |
| Apple Mac Studio M3 Ultra 192GB | Q8_0 | 34.5 GB | — | — | — |
| Apple Mac Studio M3 Ultra 96GB | Q8_0 | 34.5 GB | — | — | — |
| Apple MacBook Pro M4 Max 128GB | Q8_0 | 34.5 GB | — | — | — |
| Apple MacBook Pro M4 Max 64GB | Q8_0 | 34.5 GB | — | — | — |
| NVIDIA H100 80GB | Q8_0 | 34.5 GB | — | — | — |
| NVIDIA L40S | Q8_0 | 34.5 GB | — | — | — |
| NVIDIA GeForce RTX 3090 Ti | Q4_K_M | 20.9 GB | — | — | — |
| NVIDIA GeForce RTX 3090 | Q4_K_M | 20.9 GB | — | — | — |
| NVIDIA GeForce RTX 4090 | Q4_K_M | 20.9 GB | — | — | — |
| NVIDIA GeForce RTX 5090 | Q4_K_M | 20.9 GB | — | — | — |
| NVIDIA RTX 6000 Ada Generation | Q8_0 | 34.5 GB | — | — | — |
| NVIDIA RTX A6000 | Q8_0 | 34.5 GB | — | — | — |
| NVIDIA Tesla M40 24GB | Q4_K_M | 20.9 GB | — | — | — |
| NVIDIA Tesla P40 | Q4_K_M | 20.9 GB | — | — | — |