Tensorflow gpu setup for ubuntu

tldr.md

How to setup tensorflow gor gpu on Ubuntu

To check if tensorflow can detect any GPU just run

python -c "import tensorflow as tf;print(tf.test.is_gpu_available(True))"

If the GPUs are available it should print True.

Setup anaconda

The easiest way to setup cuda and tensorflow-gpu is to install everything using anaconda. Anaconda do not requires root permissions and create a self-contained folder in $HOME/anaconda3.

# Download anaconda
wget https://repo.anaconda.com/archive/Anaconda3-2020.02-Linux-x86_64.sh -O anaconda.sh
# Quiet setup, the default installation path is $HOME/anaconda3
bash ./anaconda.sh -b
# Setup anaconda as the default python
echo "export PATH=\$PATH:$HOME/anaconda3/bin" >> $HOME/.bashrc
# Activate the default anaconda environment
$HOME/anaconda3/bin/conda init

Check that anaconda is installed correctly

If everything went well the shell should look like something like this:

(base) user$

and to double check just run python:

(base) user$ python
Python 3.8.3 (default, Jul  2 2020, 16:21:59)
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>>

It should print the GCC version and Anaconda in the Header.

Install the drivers

This step is not needed for SLURM clusters.

If the drivers are available and correctly installed you should be able to run:

nvidia-smi

Which should print something like:

Tue Aug  4 08:17:14 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.100      Driver Version: 440.100      CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 207...  Off  | 00000000:09:00.0  On |                  N/A |
|  0%   40C    P8    26W / 235W |    546MiB /  7979MiB |      6%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1228      G   /usr/lib/Xorg                                262MiB |
|    0      1473      G   /usr/bin/gnome-shell                         133MiB |
|    0      2653      G   ...AAAAAAAAAAAACAAAAAAAAAA= --shared-files    61MiB |
|    0     44439      G   ...uest-channel-token=13355294416564996066    57MiB |
|    0     45602      G   /usr/bin/alacritty                            13MiB |
|    0     50497      G   /usr/bin/alacritty                            13MiB |
+-----------------------------------------------------------------------------+

Before installing we must install all the dependancies

sudo apt-get install build-essential freeglut3 freeglut3-dev libxi-dev libxmu-dev linux-headers-$(uname -r)

To install the cuda toolkit for Cuda 10.1 for ubuntu 18.04 run:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget http://developer.download.nvidia.com/compute/cuda/11.0.2/local_installers/cuda-repo-ubuntu1804-11-0-local_11.0.2-450.51.05-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804-11-0-local_11.0.2-450.51.05-1_amd64.deb 
sudo apt-key add /var/cuda-repo-ubuntu1804-11-0-local/7fa2af80.pub
sudo apt-get update
sudo apt-get -y install cuda

For other versions check this website and follow the instructions.

Install the python libraries

First of all you must uninstall tensorflow if already installed. If you install tensorflow-gpu over a previous version the installation will work but you won't be able to use the GPUs.

pip uninstall tensorflow

Install the drivers:

conda install cuda

Install tensorflow-gpu

conda install tensorflow-gpu

Check the correct installation

If everything went well this should output True:

python -c "import tensorflow as tf;print(tf.test.is_gpu_available(True))"