wolfman1219 / nvidia-cuda-tensorflow-installation

Nvidia-CUDA-Tensorflow-installation

Install nvidia driver

For debian based:

You can check whether your computer has an NVIDIA GPU installed or not. Open the terminal and type

lspci | egrep 'VGA|NVIDIA'

NVIDIA drivers are available in the official contrib and non-free package repositories of Debian 11. To enable the contrib package repository, run the following command:

sudo apt-add-repository contrib
sudo apt-add-repository non-free

To update the APT package database, run the following command:

sudo apt update

Install nvidia-driver:

sudo apt install nvidia-driver

sudo reboot

For arch based:

For checking has GPU or not:

lspci | egrep 'VGA|NVIDIA'

And run this commanlines:

sudo pacman -Syu

Install git and clone this:

sudo pacman -S git
git clone https://gitlab.com/XavierEduardo99/nvidia-drivers-arch-linux-installer

Enter this directory:

cd nvidia-drivers-arch-linux-installer

If your kernel linux:

sudo sh install-nvidia.sh

If your kernel is linux-lts:

sudo sh install-nvidia.sh --lts

and reboot system

Install conda

Just run this commands in your terminal:

curl https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -o Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

Follow the instruction from offical site.

Install cuda and cudnn for tensorflow

Create a virtual environment for your TensorFlow project and activate it.

conda create --name tf-gpu python=3.9
conda activate tf-gpu

Install required packages and libraries:

conda install -c conda-forge cudatoolkit=11.8.0
pip install nvidia-cudnn-cu11==8.6.0.163

Configure the system paths.The system paths will be automatically configured when you activate this conda environment.

mkdir -p $CONDA_PREFIX/etc/conda/activate.d
echo 'CUDNN_PATH=$(dirname $(python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"))' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_PREFIX/lib/:$CUDNN_PATH/lib' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh

LIFEHACK! If you want use cuda and cudnn without activating conda try this steps. WARNING!!If you can't fix after system crash, i don't recommend do this. If you believe in yourself so much, then let's start. Activate cuda installed environment:

conda activate tf-gpu

View cuda path:

python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"

Copy this path and run command.Replace path/to/cuda with copied path. WARNING copy directory path not file path!!!:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/lib/:path/to/cuda/lib'

Well done! You are set cuda path.

If when open new terminal tensorflow don't see GPU. Add export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/lib/:path/to/cuda/lib' to .bashrc file.

Install tensorflow

conda activate tf-gpu
pip install --upgrade pip
pip install tensorflow==2.12.*

If all this steps doing well following command print GPU list:

python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

Good luck