How to install CUDA, CUDNN and TensorFlow in GPU Server



Introduction

This article will show you how to install CUDA 10.0 + cudnn 7.6 + keras 2.3.1 + tensorflow 1.15.2 + python 3.7.10 in Ubuntu 18 OS. In other Linux OS, the KB is not suitable.

9 Steps to install CUDA, CUDNN and TensorFlow in GPU Server

Step 1: Install GCC

# sudo apt update
# sudo apt install build-essential
# sudo apt-get install manpages-dev
# gcc --version

Step 2: Install GPU driver.(You could upload it from terminal server.)

Note: The version of GPU driver must match the version of CUDA. The following driver is matching the CUDA 10.0.

Note: During this installation, you might encounter the following issue:

Detailed error information:Error 1:

Solution 1:

A: Edit the file /etc/modprobe.d/blacklist-nouveau.conf and add the following contents to this file:
blacklist nouveau
options nouveau modeset=0

B: running the following command to regenerate the initramfs file
sudo update-initramfs -u

C: Reboot the server

Error 2:

Solution 2:
Running the following command:

# sudo apt-get install linux-headers-`uname -r`

Step 3: Install CUDA 10.0

# sudo apt-get install freeglut3 freeglut3-dev libxi-dev libxmu-dev
# wget https://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda_10.0.130_410.48_linux
# wget http://developer.download.nvidia.com/compute/cuda/10.0/Prod/patches/1/cuda_10.0.130.1_linux.run
# mv cuda_10.0.130_410.48_linux cuda_10.0.130_410.48_linux.run
# sudo sh cuda_10.0.130_410.48_linux.run

Step 4: Install CUDA Patch

# sudo sh cuda_10.0.130.1_linux.run

Step 5: Set up the environment path

# export PATH=/usr/local/cuda-10.0/bin:$PATH
# export LD_LIBRARY_PATH=/usr/local/cuda-10.0/64:$LD_LIBRARY_PATH

Step 6: Install CUDNN 7.6

6.1 Upload the .tgz file from our terminal server.

6.2 Running the following commands

# tar xvf cudnn-10.0-linux-x64-v7.4.2.24.tgz
# sudo cp -P cuda/include/cudnn.h /usr/local/cuda/include (This path should be update)
# sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64
# sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

6.3 Check if CUDNN is installed successfully

# cat /usr/local/cuda-10.0/include/cudnn.h |grep CUDNN_MAJOR -A 2

Step 7: Install TensorFlowGPU

7.1 Install anaconda

# wget https://repo.continuum.io/archive/Anaconda3-5.0.1-Linux-x86_64.sh
# bash Anaconda3-5.0.1-Linux-x86_64.sh

During the installation process, keep pressing enter until yes/no appears. Enter yes. When asked whether to add environment variables, be sure to enter y.

7.2 Check if anaconda is installed successfully

Disconnect the server, reconnect to the server, and enter python. If no error is reported, the installation is successful.
Unexpected situation: If the environment variables are not configured successfully, the following operations need to be performed
(1) Enter the following command in the terminal: vim ~/.bashrc Open the bashrc file and edit it
(2) Add this line after the bashrc file. After PATH, add the path of the installed anaconda: export PATH=/home/.../bin:$PATH
(3)Activation file: source ~/.bashrc

7.3 Install TensorFlowGPU 1.15

# conda install --channel https://conda.anaconda.org/hanyucui tensorflow-gpu=1.15

7.4 Check if TensorFlowGPU is installed successfully

# python
>>>import tensorflow as tf
>>> tf.test.is_built_with_cuda()
>>> tf.test.is_gpu_available(cuda_only=False,min_cuda_compute_capability=None)

Step 8: Check the version of TensorFlowGPU

pip list | grep -i tensor
pip list | grep -i keras