Installing successfully CUDA 10.1 and Tensorflow 1.14 to enable GPU processing

6 min readSep 10, 2019

Here I will present to you how to set up an environment to train your models using GPU with Cuda 10.x and Tensorflow 1.14.

First, you need to install your GPU driver in your operating system (in my case is Ubuntu 18.04 and GTX 1070).

You should check the compatibility of the TensorFlow and Cuda versions here.

Some prerequisites:

sudo apt update && sudo apt install gcc

sudo apt update && sudo apt install build-essential

sudo apt update && sudo apt install libglvnd-dev pkg-config

sudo apt update && sudo apt install freeglut3 freeglut3-dev libxi-dev libxmu-dev

Compatibility TF — CUDA — GPU

You can check the compatibility here.

Compatibility TF — CUDA — CPU

First of all, we check our GPU description (here shows you if you have NVIDIA or ATI/AMD processor):

lspci | grep ' VGA ' | cut -d" " -f 1 | xargs -i lspci -v -s {}

The output should show the GPU name and the driver, we have NVIDIA GPU

Next, Add and Update your GPU driver repository:

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update

Now, we need to check which driver should we install, just run to see recommendations:

ubuntu-drivers devices

Recommended drivers for your operating system and your GPU device

Then. you can install the driver recommended by the OS:

sudo ubuntu-drivers autoinstall

Or install the driver (in our case, they recommend to install NVidia-430 for a GeForce GTX 1070):

sudo apt-get install nvidia-driver-430

Or click here to see NVIDIA recommendations and download the .run file to install it.
While you are installing NVIDIA, it asks for a password (just simple security) and if when you are rebooting appears a blue screen asking for MOK password (see the section below).

Dealing with MOK (only for UEFI Secure Boot enabled devices)

If you were asked to set up a secure boot password, you’ll see a blue screen that says something about “MOK management”. It’s a complicated topic and I’ll try to explain it in simpler terms.

MOK (Machine Owner Key) is needed due to the secure boot feature that requires all kernel modules to be signed. Ubuntu does that for all the kernel modules that it ships in the ISO. Because you installed a new module (the additional driver) or made a change in the kernel modules, your security system may treat it as an unwarranted/foreign change in your system and may refuse to boot.

If you select “Continue boot”, chances are that your system will boot like normal and you won’t have to do anything at all. But it’s possible that not all features of the new driver work correctly.

This is why you should choose Enroll MOK.

First screen when you reboot your system after install GPU driver

It will ask you to Continue on the next screen followed by asking a password. Use the password you had set while installing the additional drivers (in this case NVIDIA driver). You’ll be asked to reboot now.

enroll MOK -> continue -> enter password -> reboot

Continue the installation

Then, Reboot your computer. To verify the installation, open a terminal and run the following command:

If your nvidia-smi failed to communicate but you’ve installed the driver so many times, check prime-select:
Run prime-select query to get all possible options. You should see at least nvidia | intel.
Choose prime-select nvidia.
If it says nvidia is already selected, select a different one, e.g. prime-select intel, then switch back to Nvidia prime-select nvidia.
Reboot and check nvidia-smi.

How to install CUDA 10 (and toolkits by default)

Once installed nvidia-drivers, we need to install cuda library, first, download file installer (check here):

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/10.1/Prod/local_installers/cuda-repo-ubuntu1804-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo apt-key add /var/cuda-repo-10-1-local-10.1.243-418.87.00/7fa2af80.pub
sudo apt-get update
sudo apt-get -y install cuda

At this time you will have CUDA already installed.

CUDADNN

To install CUDA NN deep learning for neural networks in CUDA (you need to have an account):

Go to cudnn or here.
Once you login into the system, go to download page, you can select cuDNN dev or lib compatible with your CUDA version
Select CUDNN 7.6.3 for CUDA 10.x
Download the cuDNN v7.6.3 Library for Linux (deb file)
Open a terminal in the directory the tar file is located and run (in order):

sudo dpkg -i libcudnn7_7.6.3.30-1+cuda10.1_amd64.deb
sudo dpkg -i libcudnn7-dev_7.6.3.30-1+cuda10.1_amd64.deb

I hope you enjoy it!

How to verify

To verify cuda installation, just run:

Then, we write the helloWorldCUDA.cu program:

#include <stdio.h>
       
 const int N = 16; 
 const int blocksize = 16; 
         
 __global__ 
 void hello(char *a, int *b) {
   a[threadIdx.x] += b[threadIdx.x];
 }
         
 int main(){
   char a[N] = "Hello \0\0\0\0\0\0";
   int b[N] = {15, 10, 6, 0, -11, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
   char *ad;
   int *bd;
   const int csize = N*sizeof(char);
   const int isize = N*sizeof(int);
       
   printf("%s", a);
       
   cudaMalloc( (void**)&ad, csize ); 
   cudaMalloc( (void**)&bd, isize ); 
   cudaMemcpy( ad, a, csize, cudaMemcpyHostToDevice ); 
   cudaMemcpy( bd, b, isize, cudaMemcpyHostToDevice ); 
       
   dim3 dimBlock( blocksize, 1 );
   dim3 dimGrid( 1, 1 );
   hello<<<dimGrid, dimBlock>>>(ad, bd);
   cudaMemcpy( a, ad, csize, cudaMemcpyDeviceToHost ); 
   cudaFree( ad );
   cudaFree( bd );
       
   printf("%s\n", a);
   return EXIT_SUCCESS;
 }

Compile and execute with:

nvcc helloWorld.cu -o helloWorld
./helloWorld

Or using TensforFlow:

TF version

import tensorflow as tf
tf.test.is_gpu_available()

If you have some errors during installation

This happened to me, I installed 2 versions of CUDA 9 and 10.

If you have some error when you are installing cuda-toolkit like the image below:

This happened to me when I have 2 versions installed at the same time

You can fixe up it with the following command:

sudo apt-get -o Dpkg::Options::="--force-overwrite" install --fix-broken

Then, modify .bashrc file, write these lines:

export PATH=$PATH:/usr/local/cuda-10.1/bin
export CUDADIR=/usr/local/cuda-10.1
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.1/lib64

After modifying that file, run:

source .bashrc

That’s all.

Regards.