Closed 2u843yt385592yjh closed 5 days ago
Same for me. Loading an SDXL model takes just short of 600s whereas a SD1.5 model takes 10s total. I'm on a Mac Pro M1 Max with 32gb. I wonder if I'm running into a memory management issue here. With the model loaded and idle, python is taking 18.3GB of RAM, so I suspect that loading the model takes the rest of the space and swap memory and page files are used.
Solution: Install latest nvidia drivers Nvidia control panel 3D settings -> Set CUDA offloading to RAM policy to disabled
Checklist
What happened?
Issue on version 1.9.4 Hard drive spikes up to 50% usage, specifically when sdxl loras are loaded / unloaded. Once loaded (after 1 image generated and a few minutes), the performance stabilizes again with normal generation times.
Also from https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/15206
Steps to reproduce the problem
What should have happened?
Should sdxl lora loading / unloading freeze the system for a few minutes?
What browsers do you use to access the UI ?
Mozilla Firefox
Sysinfo
Console logs
Additional information
No response