Nvidia CUDA and tensorflow-GPU installation on Ubuntu 16.04

NVIDIA DRIVER INSTALLATION

Firstly install Nvidia driver, open terminal and run following commands.

1. sudo apt update
2. sudo add-apt-repository ppa:graphics-drivers/ppa
3. sudo apt update
4. sudo apt upgrade
5. sudo ubuntu-drivers devices
6. sudo apt install nvidia-410

After installation reboot your PC. Then if you have a problem in login (loop in login);

Open terminal in login screen with CTRL + ALT + F1 and then query your graphics card

• sudo prime-select query

select intel graphics card, if nvidia is selected

• sudo prime-select intel

Turn back to login screen (CTRL + ALT + F7), now you must be able to login succesfully.

Open Terminal, activate and check driver;

• sudo prime-select intel
• nvidia-smi

If everthing is OK, you must an output similar to;

CUDA TOOLKIT INSTALLATION

Now we are ready for CUDA installation;

• download CUDA 10.0 for Linux-Ubuntu-16.04: https://developer.nvidia.com/cuda-10.0-download-archive

Run sudo sh cuda_10.0.130_410.48_linux.run (When it asks to install driver, answer NO, because we have already did it!) Answer YES for all other questions

• download cudnn-7.4.2 for CUDA 10.0 ( cuDNN Library for Linux )-> https://developer.nvidia.com/rdp/cudnn-archive

Then extract the downloaded cuDNN tar file and copy the lib files to /usr/local/cuda-10.0 default path with following commands

• tar -xf cudnn-10.0-linux-x64-v7.4.2.24.tgz
• sudo cp -R cuda/include/* /usr/local/cuda-10.0/include
• sudo cp -R cuda/lib64/* /usr/local/cuda-10.0/lib64

PYTHON AND TENSORFLOW-GPU INSTALLATION

We suggest using conda distribution (download Python 3.7): https://www.anaconda.com/distribution/#download-section

After installation, add anaconda path to environment: export PATH=~/anaconda3/bin:$PATH
Then run : conda in Terminal to check and run: conda activate in your terminal to activate anaconda python environment

To install tensorflow for GPU:

• pip install tensorflow-gpu==1.13.1 (This version is suitable for CUDA 10.0)