How much VRAM does Qwen-Image 2.1 need?
Qwen-Image 2.1 needs about 8.7 GB at 8-bit, or 5.8 GB as a 4-bit GGUF, at its default output size.
- Type
- Image model
- Parameters
- 7.1B
- Text encoder
- Qwen3-VL-8B-Instruct
- Released
- Sep 2026
- License
- Qwen Research License
Memory at each precision
| Precision | Weights | Peak at 2048 × 2048 |
|---|---|---|
| 16-bit BF16 | 13.3 GB | 15.1 GB |
| 8-bit FP8 · Q8_0 | 6.9 GB | 8.7 GB |
| 6-bit GGUF Q6_K | 5.5 GB | 7.3 GB |
| 5-bit GGUF Q5_K_M | 4.7 GB | 6.6 GB |
| 4-bit GGUF Q4_K_M · NVFP4 | 4.0 GB | 5.8 GB |
The Qwen3-VL-8B-Instruct text encoder (8.2 GB at FP8) runs first and then waits in system RAM.
What runs it
| Machine | Best fit | Per image |
|---|---|---|
| RTX 3060 12 GB | 8-bit | 9.7 min |
| RTX 4060 Ti 16 GB | 16-bit | 5.6 min |
| RTX 3090 | 16-bit | 3.5 min |
| RTX 4090 | 16-bit | 1.5 min |
| RTX 5090 | 16-bit | 72 s |
| RTX PRO 6000 Blackwell | 16-bit | 31 s |
| M4 Max 128 GB | 16-bit | 16 min |
| M5 Max 128 GB | 16-bit | 4.7 min |
| Ryzen AI Max+ 395 128 GB | 16-bit | 7.1 min |
| DGX Spark 128 GB | 16-bit | 4.0 min |
| M3 Ultra 256 GB | 16-bit | 9.1 min |
| M5 Ultra 512 GB | 16-bit | 2.8 min |
| H100 SXM | 16-bit | 16 s |
| DGX Station GB300 | 16-bit | 6.9 s |
2048 × 2048, 40 steps. Times are rough, plus or minus 30%.
Where to get it
Questions
How much VRAM does Qwen-Image 2.1 need?
Qwen-Image 2.1 needs about 8.7 GB at 8-bit, or 5.8 GB as a 4-bit GGUF, at its default output size.
Can Qwen-Image 2.1 run on an RTX 4090?
Yes. At 16-bit it takes about 1.5 min per image.