SDXL model loading requires too much RAM (32GB+ maybe)

[okolenmi]

I have a very simple SDXL workflow without even a refiner. It fails on first step - model loading. This step is not requiring VRAM (also there is no signs of using VRAM in task manager)

Not enough memory error.
It looks like 19GB of free memory (5GB were used before model loading) is not enough to load 6.8GB model.

Is this behaviour expected at the moment or not? (model, that was used)

[ltdrdata]

Is it RAM or VRAM?

[okolenmi]

It's RAM. VRAM doesn't used during model loading as I can see.

[usigg3]

Make sure you have a windows page or swapfile.

[okolenmi]

It works with 8192MB page file. It requires only 10GB of free RAM after model loading. I think, it can be optimized in some way. But, at least, it works.

[GothWave]

please, what is a windows page and swapfile? how can i search for this? (google only shows "face swap")
update: alright, i managed to understand what is a paging file size. My question now is, i need to allocate in the disk that is running ComfyUI or in the entire system?