Why does SD still occupy GPUs when there is not enough memory to run or when SD is not running? No image generation operation during this period，Sleep mode required

[wzgrx]

When running multiple images, the memory usage gradually increases, displaying OutOfMemoryError: CUDA out of memory Tried to allocate 324.00 MiB (GPU 0; 8.00 GiB total capacity; 4.82 GiB already allocated; 0 bytes free; 7.17 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_ split_ size_ mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_ CUDA_ ALLOC_ CONF, but with the same settings as the first one, the image can be run, and after an error is reported due to insufficient memory, the task manager displays that the GPU still occupies a high amount, resulting in the inability to generate images. Therefore, it is necessary to completely restart SD.

Why not release the GPU when there is no sleep mode or when no images are generated? Restarting SD every time is troublesome.

[wzgrx]

When I don't use SD and use other applications to occupy GPUs, I don't have enough GPUs to use it. If I hang SD and don't operate it, it will also occupy a lot of GPUs. I don't understand why?

[fin-gal-25]

Good day, SD using memory because checkpoint is loaded in VRAM even if not generating images.
Also SD2.1 models use more VRAM than SD1.5
Did you try to add this to your webui-user.bat: ?

set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
set CUDA_MODULE_LOADING="LAZY"

Also you can use [Unload SD checkpoint to free VRAM] in Settings -> Action

[wzgrx]

Good day, SD using memory because checkpoint is loaded in VRAM even if not generating images. Also SD2.1 models use more VRAM than SD1.5 Did you try to add this to your webui-user.bat: ?

set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
set CUDA_MODULE_LOADING="LAZY"

Also you can use [Unload SD checkpoint to free VRAM] in Settings -> Action
Okay, thank you. I'll try it out