Hires Fix for 4GB VRAM

[AniMoster]

I have a Nvidia GTX 1650Ti 4GB VRAM card

This is what I use in my webui-user.bat file

@echo off

set PYTHON="E:AI\stable-diffusion-webui\venv\Scripts\python.exe"
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --update-all-extensions --lowvram --disable-nan-check --disable-safe-unpickle --opt-split-attention
set 'PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512'
git pull

call webui.bat

And when I use Hires Fix I get the following error:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 768.00 MiB (GPU 0; 4.00 GiB total capacity; 2.21 GiB already allocated; 0 bytes free; 3.35 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

Anyone has any idea why I'm unable to use Hires Fix even after using --lowvram & PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512

I'm looking forward for your suggestions. Thank you

[w-e-w]

that's surprising, just hardware limitation

it literally tells you that you're out of memory
aka means that you cannot generate images that large

you can give Tiled VAE a try

[AniMoster]

I thought the same until I saw this HERE and it gave me some hope. So I thought of asking that here.

I have installed Tiled VAE as you suggested. I will try it out.
And doesn't Tiled VAE also use Hires Fix?
Thank you

[w-e-w]

I just did the clean install and tested on my 1650 4G (non it)
with --xformar --lowvram
it was able to generate 1024x1024 (which is near the limit)
with tiled vae I was able to do an image at 2048x2048
not sure of the actual limit and the quality, just saying it's possible
and I also don't recommend because higher resolution takes far longer on a weak GPU like the 1650

[leizim182]

I have a notebook with GTX 1650 4GB too. Heres my commands:
COMMANDLINE_ARGS =--medvram --xformers --autolaunch

I use only these commands above. I put the image in 512x768 or 768x512, changing it to portrait or landscape. For hires fix use 1.5, and then use upscale extra menu with 2x or 2.5x value.
I didn't update the automatic 1111 with this last update. I saw that a lot of people had a problem, and since it's working like this, I decided to wait for future updates.

[w-e-w]

I just did a clean install on my laptop whitch has a 1650 4G
I didn't experience any particular issues including vram issues with the current 1.1.1
there were some known issues that if you didn't update properly with 1.1.0 but it has been resolved since 1.1.1

[leizim182]

Acabei de fazer uma instalação limpa no meu laptop whitch tem um 1650 4GEu não experimentei nenhum problema específico, incluindo problemas de vram com o atual 1.1.1Havia alguns problemas conhecidos que se você não atualizou corretamente com a 1.1.0, mas foi resolvido desde a 1.1.1

Good to know. I will do a separate install when I get home. What arguments are you using in this new installation?