Qwen2.5 Coder 32B: how much VRAM?
In Q4_K_M with 8k tokens of context, Qwen2.5 Coder 32B needs about 23 GB of memory. The cheapest option: 2 × RTX 3060 12 GB (€1,460 for the complete machine).
Model sheet
ollama run qwen2.5-coder:32bWeights on Hugging Face: Qwen/Qwen2.5-Coder-32B-Instruct
VRAM by precision and context
| Precision | 4k | 8k | 32k | 128k |
|---|---|---|---|---|
| FP16 | 70 GB | 71 GB | 77 GB | 101 GB |
| FP8 | 36 GB | 37 GB | 43 GB | 67 GB |
| Q8_0 | 38 GB | 39 GB | 45 GB | 69 GB |
| Q6_K | 30 GB | 31 GB | 37 GB | 61 GB |
| Q4_K_M | 22 GB | 23 GB | 29 GB | 53 GB |
| Q3_K_M | 18 GB | 19 GB | 25 GB | 49 GB |
One user. Each simultaneous user adds their own context cache.
Frequently asked questions
How much VRAM does Qwen2.5 Coder 32B need?
With an 8k-token context and one user, Qwen2.5 Coder 32B needs about 23 GB in Q4, 39 GB in Q8 and 71 GB in FP16.
What is the cheapest hardware to run Qwen2.5 Coder 32B?
In Q4, the cheapest option in our catalogue to run Qwen2.5 Coder 32B is 2 × RTX 3060 12 GB, about €1,460 for the complete machine.
How fast does Qwen2.5 Coder 32B run?
Expect around 15 tokens per second for one user on 2 × RTX 3060 12 GB in Q4 (estimate based on memory bandwidth).
On which hardware?
| Hardware | Cards | Speed | Machine |
|---|---|---|---|
| RTX 3060 12 GB | 2 cards | ~15 tok/s | €1,460 |
| RTX 3090 24 GB (used) | 1 is enough | ~26 tok/s | €1,650 |
| RTX 4060 Ti 16 GB | 2 cards | ~12 tok/s | €1,800 |
| Radeon RX 7900 XTX 24 GB | 1 is enough | ~27 tok/s | €1,850 |
| Radeon RX 7800 XT 16 GB | 2 cards | ~26 tok/s | €1,900 |
| RTX 4070 12 GB | 2 cards | ~21 tok/s | €2,000 |
| Mini-PC Ryzen AI Max+ 395 128 GB | 1 is enough | ~7.1 tok/s | €2,300 |
| RTX 4070 Ti SUPER 16 GB | 2 cards | ~28 tok/s | €2,600 |
| RTX 4090 24 GB | 1 is enough | ~28 tok/s | €2,900 |
| RTX 4080 SUPER 16 GB | 2 cards | ~31 tok/s | €3,000 |
| RTX 5080 16 GB | 2 cards | ~40 tok/s | €3,300 |
| RTX 5090 32 GB | 1 is enough | ~50 tok/s | €3,500 |
| NVIDIA DGX Spark 128 GB | 1 is enough | ~7.6 tok/s | €4,000 |
| Mac Studio M4 Max 128 GB | 1 is enough | ~15 tok/s | €4,400 |
| RTX 6000 Ada 48 GB | 1 is enough | ~27 tok/s | €7,900 |
| RTX PRO 6000 Blackwell 96 GB | 1 is enough | ~50 tok/s | €9,900 |
| Mac Studio M3 Ultra 512 GB | 1 is enough | ~23 tok/s | €11,500 |
| L40S 48 GB | 1 is enough | ~24 tok/s | €13,500 |
| A100 80 GB | 1 is enough | ~57 tok/s | €17,000 |
| H100 80 GB | 1 is enough | ~93 tok/s | €34,000 |
| H200 141 GB | 1 is enough | ~133 tok/s | €39,000 |
| B200 192 GB | 1 is enough | ~222 tok/s | €48,000 |
Estimated complete machine: cards + host PC (€900 for 1–2 consumer cards, €2,800 beyond, a €6,000 server for datacenter cards). Macs and mini-PCs are sold complete.
Similar models
From: RTX 3060 12 GB · €1,180
From: RTX 3060 12 GB · €1,180
From: RTX 3060 12 GB · €1,180
From: RTX 4060 Ti 16 GB · €1,350