Closed David-337 closed 1 year ago
Also having same issue with RTX 2080 ti -- 11gb vram
Update: solved by fully nuking Nvidia drivers, CUDA toolkit and any other nvidia or cuda related packages from the system. Then reinstalling latest nvidia from rpmfusion and cuda 11.8 from https://developer.download.nvidia.com/compute/cuda/repos/fedora35/x86_64/.
Is there an existing issue for this?
What happened?
Around a month ago everything was working fine, but recently I booted up A1111 and the extension on their latest main branch commits, and I get an instant OOM whenever I try to train a model, on my 24 GB 3090 Ti, even if I use settings that worked for me a month ago.
I have tried:
Steps to reproduce the problem
Commit and libraries
Command Line Arguments
Console logs
Additional information
I'm using Fedora 38 with CUDA 11.8 and Nvidia drivers 535 (although I also tried downgrading to 525)