Installation of NVIDIA CUDA Drivers and Tensorflow (GPU enabled)

In this document we describe how we installed NVIDIA CUDA drivers and Tensorflow (for GPU). Before installation, make sure your system have CUDA-capable GPU and supported Linux version.

You can find out if you have Nvidia gpu by using command: lspci | grep -i nvidia and checking if your gpu is listed on "Cuda enabled products" list (NVIDIA CUDA Enabled products).

First step was to install latest updates using following commands: sudo apt-get update and sudo apt-get upgrade

After that, you need to download installation packet for CUDA (Site shows commands needed to install selected package) from Nvidia CUDA downloads. More specific installation guide can be found from Nvidia CUDA installation guide.

You also need Nvidia's gpu drivers, usually you can check if you have Nvidia gpu driver installed by running command sudo nvidia-smi. Latest NVIDIA drivers can be downloaded from Nvidia drivers page.

Command for tensorflow installation, using pip(3)

Python version	CPU	GPU
Python 2.7	pip install tensorflow	pip install tensorflow-gpu
Python 3.x	pip3 install tensorflow	pip3 install tensorflow-gpu

You can test if your installation succeeded by running python(3) in terminal and writing import tensorflow, if errors dont appear, installation should be successful.

If you dont have pip installed, you can install pip by using sudo apt install python-pip python-dev (For Python3, replace python with python3)

Better guide for installing tensorflow is available at Installing TensorFlow on Ubuntu.