Last active
September 16, 2022 13:58
-
-
Save w2sv/d96cf1afdc021ff7443b9e0067c69b33 to your computer and use it in GitHub Desktop.
Setting up tensorflow-gpu on Ubuntu 22.04
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# $$$$$$$$$$ Pre-Installation $$$$$$$$$$$$$$$ | |
# Ensure presence of cuda-compatible GPU | |
lspci | grep -i nvidia # should output something like '01:00.0 3D controller: <YOUR_NVIDIA_GPU_KIND> (rev a1)' | |
# Ensure gcc being installed | |
gcc --version | |
# Optional: install the nvidia cuda toolkit | |
sudo apt-get update | |
sudo apt-get install nvidia-cuda-toolkit | |
# $$$$$$$$$$ Installation $$$$$$$$$$ | |
# Head over to https://www.tensorflow.org/install/source#gpu to determine the cuda and cuDNN versions | |
# which are compatible with your desired tensorflow & python version, and set them as session variables | |
# for usage by subsequent commands. | |
cudnn_version={{e.g. 8.1.1.*}} # NOTE: the specified version must comprise the 4 specification levels (or 3 + '.*', respectively) | |
cuda_version={{e.g. 11.2}} | |
# 1. Install CUDA | |
sudo apt-get update | |
sudo apt-get install cuda-${cuda_version} | |
# if you've installed the nvidia cuda toolkit, running 'nvcc --version' should now reflect your installed CUDA version | |
# 2. Install cuDNN, following https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#package-manager-ubuntu-install | |
# Enable repository | |
OS=ubuntu2204 | |
wget https://developer.download.nvidia.com/compute/cuda/repos/${OS}/x86_64/cuda-${OS}.pin | |
sudo mv cuda-${OS}.pin /etc/apt/preferences.d/cuda-repository-pin-600 | |
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/${OS}/x86_64/3bf863cc.pub | |
sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/${OS}/x86_64/ /" | |
sudo apt-get update | |
# Install cuDNN library | |
sudo apt-get install libcudnn8=${cudnn_version}-1+cuda${cuda_version} | |
# 3. Install tensorflow-gpu in a virtual environment hosted by e.g. conda. | |
conda create -n tf-gpu python={{PY_VERSION}} -y | |
conda activate tf-gpu | |
conda install tensorflow-gpu={{TF_VERSION}} # if desired version not hosted by conda, just install it via pip instead by running: | |
# pip install tensorflow-gpu=={{TF_VERSION}} | |
# $$$$$$$$$$ Post-Installation $$$$$$$$$$$$$$$ | |
# verify installation | |
python -c 'import tensorflow as tf; tf.config.experimental.list_physical_devices("GPU")' | |
# monitor your GPU via the nvidia system management interface. Note: the CUDA version displayed in the top right corner of the | |
# generated output may differ from the one you've just installed manually. | |
nvidia-smi |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Those are simply the steps I found to have worked for me. The file ain't actually supposed to be run as a script, but to provide all the necessary command blocks required for the set-up. Particularly the desired cuda, cudnn, python and tf-gpu versions, all marked by double-curly braces, must be manually provided beforehand.