What GPU do you need to run DeepSeek-R1-Distill-Qwen-32B?
32.76B parameters (Hugging Face, accessed 2026-10-15). Q4_K_M GGUF: 19.85 GB (source).
| Single GPU that fits | VRAM | Speed estimate | Price |
|---|---|---|---|
| NVIDIA GeForce RTX 4090 | 24 GB | ~39 tok/s | low $3,200, avg $4,663 (used) source, 2026-10-16 |
| NVIDIA GeForce RTX 3090 Ti | 24 GB | ~39 tok/s | check price |
| NVIDIA GeForce RTX 3090 | 24 GB | ~36 tok/s | low $1,889, avg $2,404 (used) source, 2026-10-16 |
| AMD Radeon RX 7900 XTX | 24 GB | ~37 tok/s | low $1,350, avg $1,552 (new) source, 2026-10-16 |
| NVIDIA Tesla P40 | 24 GB | ~13 tok/s | check price |
| NVIDIA GeForce RTX 5090 | 32 GB | ~69 tok/s | low $6,500, avg $8,757 (new) source, 2026-10-16 |
| AMD Radeon AI PRO R9700 | 32 GB | ~25 tok/s | check price |
| Apple Mac Studio M5 Max (32-core GPU, 36 GB) | 36 GB | ~18 tok/s | low $2,499 (new) source, 2026-10-16 |
| NVIDIA RTX A6000 | 48 GB | ~29 tok/s | low $6,170 (new) source, 2026-10-16 |
| NVIDIA RTX 6000 Ada Generation | 48 GB | ~37 tok/s | check price |
| Apple Mac Studio M5 Max (40-core GPU, 64 GB) | 64 GB | ~24 tok/s | check price |
| Apple Mac Studio M5 Ultra (64-core GPU, 96 GB) | 96 GB | ~46 tok/s | check price |
| Framework Desktop, AMD Ryzen AI Max+ PRO 495 (Strix Halo), 192 GB | 192 GB | ~10 tok/s | low $6,799 (new) source, 2026-10-16 |
Speed calculator estimate
Estimate = 0.76 × bandwidth ÷ GB read per token (Q4 file size, scaled by active/total parameters for MoE). Fit = file size + 1.5 GB headroom ≤ total VRAM. Multi-GPU: VRAM adds, speed uses one card's bandwidth (layer split). Short context only. How this works
Related
- Build that fits it: 24 GB: one used RTX 3090
- Build that fits it: 32 GB: two Radeon RX 9070
- Build that fits it: 48 GB: dual used RTX 3090