Nvidia driver, Cuda, Pytorch installtion.md

Read this

• main problem is compatibility issues of driver-cuda-pytorch
• Currently direct pytorch installation support 12.1 CUDA, above it you need to install manually
• Thus we will select 12.1 CUDA
• it requires driver >530

Install Nvidia driver

• https://askubuntu.com/questions/524242/how-to-find-out-which-nvidia-gpu-i-have
  • sudo update-pciids
  • lspci -nn | grep '\[03'
• ![[Pasted image 20240605144929.png]]
• RTX A6000
• https://ubuntu.com/server/docs/nvidia-drivers-installation
  • sudo ubuntu-drivers list
  • sudo ubuntu-drivers install nvidia:535
    • select the version
• This gives available versions with ubuntu package manager, but if you want latest drivers you can go here
  • https://www.nvidia.com/download/index.aspx
  • select your gpu download the driver
  • sudo sh <filename>.run
• After installation
  • ![[Pasted image 20240605160623.png]]
  • CUDA version written here is maximum version compatible with this driver (its not yet installed)
  • ![[Pasted image 20240605145434.png]]

Installing CUDA

• https://docs.nvidia.com/cuda/cuda-installation-guide-linux/
• check GCC is installed
  • ![[Pasted image 20240605145610.png]]
• update kernels
  • sudo apt-get install linux-headers-$(uname -r)
• It is recommended to use the distribution-specific packages,
• Download and install CUDA (search on google with the exact version you want) - choose upto 12.1 version as for later version you need to install pytorch manually from source which would be difficult
  • https://developer.nvidia.com/cuda-downloads
  • ![[Pasted image 20240605145747.png]]
  • ![[Pasted image 20240605153616.png]]
  • check the version mentioned in commands here it is 12.1
  • sudo apt-get update should show cuda repo
    • ![[Pasted image 20240605154327.png]]
  • install with following command
    • sudo apt install cuda-toolkit-12-1 and not sudo apt-get -y install cuda as driver is already installed so it will conflict with the driver
  • reboot after installation
• After reboot do following
  • https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions
  • Environment Setup``
    • echo "export PATH=/usr/local/cuda-12.1/bin${PATH:+:${PATH}}" >> ~/.bashrc
    • restart the terminal
  • Verify installation
    • cat /proc/driver/nvidia/version
    • nvcc --version
      • ![[Pasted image 20240605162637.png]]
  • Install Third-party Libraries
    • sudo apt-get install g++ freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libglu1-mesa-dev libfreeimage-dev libglfw3-dev

Install Pytorch

• check which version is compatible with the cuda
  • https://pytorch.org/get-started/previous-versions/
• https://pytorch.org/get-started/locally/
• create a conda env and install in it
  • ![[Pasted image 20240605153859.png]]
  • conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia