Skip to content

Qwen-Image VRAM requirements

How much VRAM Qwen-Image needs
8-bit: 20.9 GB4-bit GGUF: 12.6 GBAt 1328 × 1328
Type
Image model
Parameters
20.4B
Text encoder
Qwen2.5-VL-7B-Instruct
Released
Aug 2025
License
Apache 2.0

How much VRAM Qwen-Image needs

Memory Qwen-Image needs at each precision
PrecisionWeightsPeak at 1328 × 1328
16-bit BF1638.1 GB39.2 GB
8-bit FP8 · Q8_019.7 GB20.9 GB
6-bit GGUF Q6_K15.7 GB16.8 GB
5-bit GGUF Q5_K_M13.6 GB14.7 GB
4-bit GGUF Q4_K_M · NVFP411.4 GB12.6 GB

The Qwen2.5-VL-7B-Instruct text encoder (7.7 GB at FP8) runs first and then waits in system RAM.

GPUs and Macs that run Qwen-Image

How Qwen-Image runs on common GPUs, Macs and AI boxes
MachineBest fitPer imageReport
RTX 4060Too bigPDF: Qwen-Image on RTX 4060 report
RTX 5060Too bigPDF: Qwen-Image on RTX 5060 report
RTX 3070Too bigPDF: Qwen-Image on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Qwen-Image on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Qwen-Image on RTX 3060 12 GB report
Arc B580Too bigPDF: Qwen-Image on Arc B580 report
RTX 4070Too bigPDF: Qwen-Image on RTX 4070 report
RTX 4070 SuperToo bigPDF: Qwen-Image on RTX 4070 Super report
RTX 5070Too bigPDF: Qwen-Image on RTX 5070 report
RTX 4060 Ti 16 GB5-bit15 minPDF: Qwen-Image on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB5-bit14 minPDF: Qwen-Image on RTX 5060 Ti 16 GB report
Arc A770 16 GB5-bit12 minPDF: Qwen-Image on Arc A770 16 GB report
Radeon RX 9070 XT5-bit5.8 minPDF: Qwen-Image on Radeon RX 9070 XT report
RTX 4070 Ti Super5-bit7.5 minPDF: Qwen-Image on RTX 4070 Ti Super report
RTX 4080 Super5-bit6.3 minPDF: Qwen-Image on RTX 4080 Super report
RTX 5070 Ti5-bit7.5 minPDF: Qwen-Image on RTX 5070 Ti report
RTX 50805-bit5.9 minPDF: Qwen-Image on RTX 5080 report
Radeon RX 7900 XTX8-bit8.0 minPDF: Qwen-Image on Radeon RX 7900 XTX report
RTX 30908-bit8.1 minPDF: Qwen-Image on RTX 3090 report
RTX 40908-bit3.5 minPDF: Qwen-Image on RTX 4090 report
Radeon AI PRO R97008-bit5.1 minPDF: Qwen-Image on Radeon AI PRO R9700 report
RTX 50908-bit2.7 minPDF: Qwen-Image on RTX 5090 report
RTX PRO 5000 Blackwell 48 GB16-bit2.2 minPDF: Qwen-Image on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada16-bit1.6 minPDF: Qwen-Image on RTX 6000 Ada report
L40S16-bit1.6 minPDF: Qwen-Image on L40S report
RTX PRO 6000 Blackwell16-bit70 sPDF: Qwen-Image on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Qwen-Image on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Qwen-Image on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Qwen-Image on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Qwen-Image on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit75 minPDF: Qwen-Image on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit75 minPDF: Qwen-Image on Mac mini M4 Pro 64 GB report
M4 Max 128 GB16-bit37 minPDF: Qwen-Image on M4 Max 128 GB report
M5 Max 128 GB16-bit11 minPDF: Qwen-Image on M5 Max 128 GB report
M3 Ultra 96 GB16-bit21 minPDF: Qwen-Image on M3 Ultra 96 GB report
M3 Ultra 256 GB16-bit21 minPDF: Qwen-Image on M3 Ultra 256 GB report
M3 Ultra 512 GB16-bit21 minPDF: Qwen-Image on M3 Ultra 512 GB report
M5 Ultra 512 GB16-bit6.6 minPDF: Qwen-Image on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit17 minPDF: Qwen-Image on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB16-bit17 minPDF: Qwen-Image on Ryzen AI Max+ 395 128 GB report
RTX Spark16-bit9.2 minPDF: Qwen-Image on RTX Spark report
DGX Spark16-bit9.2 minPDF: Qwen-Image on DGX Spark report
DGX Station GB30016-bit15 sPDF: Qwen-Image on DGX Station GB300 report
A100 80 GB SXM16-bit1.8 minPDF: Qwen-Image on A100 80 GB SXM report
H100 SXM16-bit36 sPDF: Qwen-Image on H100 SXM report

1328 × 1328, 50 steps. Times are rough, plus or minus 30%.

Where to get it

Questions

How much VRAM does Qwen-Image need?

Qwen-Image needs about 20.9 GB at 8-bit, or 12.6 GB as a 4-bit GGUF, at its default output size.

What is the minimum VRAM for Qwen-Image?

The minimum is about 12.6 GB of VRAM, as a 4-bit file at its default output size, with the text encoder parked in system RAM.

How big is Qwen-Image?

The 20.4B denoiser is 38.1 GB at 16-bit and 19.7 GB at 8-bit; the Qwen2.5-VL-7B-Instruct text encoder comes on top.

Can Qwen-Image run on an RTX 4090?

Yes. At 8-bit it takes about 3.5 min per image.

More models