Installing DeepLabCut 3.0 , Pytorch doesn't natively install its own CUDA and cuDNN as advertised, windows 11 #2644
Description
Is there an existing issue for this?
- I have searched the existing issues
Bug description
Trying to install DLC 3.0 with GPU. Here are my specs:
- Windows 11
- GPU: NVIDIA RTX A2000 12GB, GPU driver: 552.55
- CUDA Toolkit 12.1
- Visual Studio 2022
I followed exactly the installation instructions but find that Pytorch is not installing its own nvcc and cudnn though advertised that it should based on both DLC installation Guide and Pytorch documentation.
After installing a compatible driver for NVIDIA, a compatible toolkit with Pytorch's latest version (just in case) and the latest DeepLabCut with either pip or the yml, DLC installs but does not use the GPU.
According to what I read Pytorch installs its own CUDA and cuDNN. So I opened a python command line and imported torch and checked for torch.cuda.is_available() = False and torch.backend.cudnn.is_available()=False. So It seems for some people Pytorch doesn't natively install what is necessary?
Is there any specific configuration required by Pytorch and DLC for Pytorch to install these things automatically? Have other people experienced this? Thanks for your help!
Operating System
operating system Windows 11
DeepLabCut version
dlc version 3.0
DeepLabCut mode
multi animal
Device type
NVIDIA RTX A2000 12GB
Steps To Reproduce
- Install Visual Studio 2022
- Install NVIDIA driver
- Install CUDA Toolkit
- Install deelabcut in conda environment
Relevant log output
No response
Anything else?
No response
Code of Conduct
- I agree to follow this project's Code of Conduct