CUDA out of Memory

[TomPeak]

I just have 4GB of GPU RAM(Nvidia). If the process wants to render, I get an error. Just few MB missing.

I did
export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:124
in der .rcbash file and start ConfyUI with --cuda-malloc … tried the oposite as well.

Where to set it properly?
Is the async malloc in source ok?

Thx,
Tom

[TomPeak]

confused

[TomPeak]

LINUX:
run: nano ~/.bashrc
add: PYTORCH_CUDA_ALLOC_CONF="max_split_size_mb:128"
124 = MB size of alloc
restart console!

does nothing ... must be the conda enviroment some how.

If you have tips?

I started the animateDiff enviroment before running the gui also sometimes. Is anyhow to set there the VRAM usage?