Currently you will need to hotfix/manually update some packages in order to get the performance you deserve when using a lovelace gpu (4080/4090/etc)
- add to webui-user.sh: export TORCH_COMMAND="pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu117"
- source ./venv/bin/activate
- pip install xformers==0.0.16rc424
- pip install triton
- download https://developer.download.nvidia.com/compute/redist/cudnn/v8.7.0/local_installers/11.8/cudnn-linux-x86_64-8.7.0.84_cuda11-archive.tar.xz
- copy all files /cudnn-linux-x86_64-8.7.0.84_cuda11-archive/lib/* to venv/lib/python3.10/site-packages/torch/lib
- add to webui-user.sh: export COMMANDLINE_ARGS="--xformers" s
- using conda env
- conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
- copy cudnn-linux-x86_64-8.7.0.84_cuda11-archive/lib/* to /home/YOURUSER/miniconda3/envs/invokeai/lib/python3.9/site-packages/nvidia/cudnn
- https://docs.getgrist.com/3mjouqRSdkBY/sdperformance
- RTX 4080 (16 GB) Linux 2745 none true 0 false 23.40 10.52 6.29 cuda 11, cudnn 8.7.84 14-01-2023
ln -s libnvrtc-672ee683.so.11.2 libnvrtc.so