torch==1.13.1+cu117 DOES NOT EQUAL torch==1.13.1+cu117

[aifartist]

If you clone A1111, create a venv, and do a manual "pip3 install -r requirements.txt" you end up with a different setup than if you simply run webui.sh and let it install everything on the first run. In the former case the cuDNN lib's end up in:
venv/lib/python3.10/site-packages/nvidia/cudnn/lib
and in the later:
venv/lib/python3.10/site-packages/torch/lib
What's even more scary is that the libraries aren't identical for these two different 1.13.1+cu117 things.
If I'm guessing then the former comes from a nvidia repro and the later from the pytorch org and who knows if they are bug compatible, optimized the same, have the same pytorch features enabled, or even used the same compiler when built. Does it matter? I have no idea but I don't like it.

Env 1:
pip3 install torch
pip3 list | egrep "torch|cudnn"

nvidia-cudnn-cu11 8.5.0.96
torch             1.13.1

python3 -c "import torch;print(torch.__version__)"
1.13.1+cu117

Env 2:
pip3 install torch==1.13.1+cu117 --extra-index-url https://download.pytorch.org/whl/cu117
pip3 list | egrep "torch|cudnn"
torch 1.13.1+cu117
python3 -c "import torch;print(torch.__version__)"
1.13.1+cu117

[Sakura-Luna]

You're kidding, there is only one CUDA in the world and it's Nvidia's, the differences you're talking about are just different versions.

[aifartist]

Try printing the "VERSION" used to "IDENTIFY" these two different things and see if there is a difference.
Also, I believe I was talking about the VERSION of "torch". FYI, Torch and CUDA are not the same thing.

[vladmandic]

cudnn is not the same as cuda. torch is build against specific version of cuda, but then dynamically loads cudnn.
check with

python -c 'import torch; import platform; print("- Python:", platform.python_version(), "Torch:", torch.version, "CUDA:", torch.version.cuda, "cuDNN:", torch.backends.cudnn.version(), "GPU:", torch.cuda.get_device_name(torch.cuda.current_device()), "Arch:", torch.cuda.get_device_capability());'

[FurkanGozukara]

24.) Automatic1111 Web UI - PC - Free
For downgrade to older version if you don't like Torch 2 : first delete venv, let it reinstall, then activate venv and run this command pip install -r "path_of_SD_Extension\requirements.txt"
How To Install New DREAMBOOTH & Torch 2 On Automatic1111 Web UI PC For Epic Performance Gains Guide

this method installs latest cuda dll files too

test py

import torch

print(f"torch {torch.__version__}, cuda {torch.version.cuda}, cudnn {torch.backends.cudnn.version()}")