Cuda + Pytorch installation for Ubuntu

This tutorial provides a step-by-step guide to install Cuda and PyTorch on an Ubuntu system. It covers the installation process for both Python and C++ setups, ensuring compatibility with your hardware and software configurations.

System Specifications

The following setup is tailored for systems with:

• Processor: Intel i7 8th Generation
• GPU: Nvidia GTX 1060 6GB
• Memory: 16 GB RAM
• Storage: 512 GB SSD

Software Versions Installed

• GCC Version: 11
• NVIDIA Driver: 470
• Python Version: 3.10.16
• CUDA Toolkit Version: 12.0.76
• PyTorch (Python and C++)

Installation Steps

1. Install GCC

To ensure compatibility with CUDA, install GCC version 11.

wget https://ftp.gnu.org/gnu/gcc/gcc-11.3.0/gcc-11.3.0.tar.gz
tar -xvf gcc-11.3.0.tar.gz
cd gcc-11.3.0

Use the following commands to configure GCC 11 as an alternative (this installs both choices in update-alternatives and allows you to choose either of them based on the compatible versioning):

sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 12
sudo update-alternatives --install /usr/bin/gcc gcc /path/to/gcc-11 11
sudo update-alternatives --config gcc

Verify the installation

gcc --version

Output: Ensure it shows gcc (Ubuntu 11.x).

2. Install NVIDIA GPU Driver

First, remove any existing NVIDIA drivers:

sudo apt remove --purge '^nvidia-.*'
sudo apt autoremove
sudo apt clean

Update the package list and install Compatible NVIDIA Driver XXX (As per my hardware, the most stable version that I can use is 470)

sudo apt update
sudo apt install nvidia-driver-470
sudo reboot

Check installation:

nvidia-smi

Output:

3. Install CUDA Toolkit

Download and isntall the CUDA Toolkit:

wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
chmod +x cuda_11.8.0_520.61.05_linux.run
sudo sh cuda_11.8.0_520.61.05_linux.run --silent --toolkit

Add CUDA paths to your environment:

echo 'export PATH=/usr/local/cuda-11.8/bin:$PATH' >> ~/.bashrc
echo 'export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
source ~/.bashrc

Verify the CUDA installation

nvcc --version

Output:

4. [OPTIONAL] Compatible Python version installation

If not present let's add python repositor:

sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt install python3.10 python3.10-venv python3.10-dev

You can configure alternatives for Python versions if needed.

5. Install PYTORCH with CUDA support using PIP

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Test the installation in python

import torch 
print(torch.cuda.is_available())

Output:

6. Install PYTORCH for C++

Install necessary tools like CMake:

sudo apt install cmake -y

Follow the official Pytorch C++ setup guide here

7. Install cuDNN

Download the appropriate cuDNN package from NVIDIA's website (e.g., version 9.6):

wget https://developer.download.nvidia.com/compute/cudnn/9.6.0/local_installers/cudnn-local-repo-ubuntu2404-9.6.0_1.0-1_amd64.deb
sudo dpkg -i cudnn-local-repo-ubuntu2404-9.6.0_1.0-1_amd64.deb
sudo cp /var/cudnn-local-repo-ubuntu2404-9.6.0/cudnn-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cudnn-cuda-11

Link cuDNN libraries to CUDA paths:

sudo ln -sf /usr/lib/x86_64-linux-gnu/libcudnn* /usr/local/cuda/lib64/