Ubuntu Server Installation Tensorflow-gpu

Catalogue

• Graphics driver Installation
• Cuda Installation
• CUDNN Installation
• TENSORFLOW-GPU Installation

this time using the host configuration:

CPU:i7-8700k graphics :gtx-1080ti

First, install the video driver

Open a Command Window (ctrl+alt+t)

sudo apt-get purge nvidia*sudo add-apt-repository ppa:graphics-drivers/ppasudo apt-sudoinstall nvidia-384 nvidia-settings

if the error Add-apt-repository does not exist, run the following code to resolve it:

sudo Install Software-properties-common python-software-properties

Verify that the installation is successful

Nvidia-smi

The following results show that the installation was successful

second, the installation of Cuda

As the cuda9.0+ version of the NVIDIA official website is defective, the installation option CUDA8.0

To Download the installation package:

Address: https://developer.nvidia.com/cuda-80-ga2-download-archive

After entering the download page, download CUDA8.0 (cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64.deb) installation packages and

An upgrade package (CUDA-REPO-UBUNTU1604-8-0-LOCAL-CUBLAS-PERFORMANCE-UPDATE_8.0.61-1_AMD64.DEB)

Installation Steps :

1. Install the base package

sudo dpkg-i cuda-repo-ubuntu1604-8-0-local-ga2_8. 0.61-1_amd64.debsudo apt-get updatesudoinstall Cuda

2. Install the upgrade package

sudo dpkg-i cuda-repo-ubuntu1604-8-0-local-cublas-performance-update_8. 0.61-1_amd64.debsudo apt-get update  sudo apt-get upgrade Cuda

Set Environment variables:

Add in. bashrc

Export path=/usr/local/cuda/bin${path:+: ${path}}export ld_library_path=/usr/local/cuda/lib64${ld_ library_path:+: ${ld_library_path}}export cuda_home=/usr/local/cuda

View the installation version

Cat /usr/local/cuda/version.txt

Output

8.0.  A

third, installation Cudnn

cuDNN6.0 and CUDA8.0 are the best partners, so this time choose cuDNN6.0

To Download the installation package :

Address: Https://developer.nvidia.com/rdp/cudnn-download

Installation:

1. Copy the downloaded compressed package to the Cuda directory

sudo CP cudnn-8.0-linux-x64-v6. 0. Tgz/usr/local/cuda

2. Enter/usr/local/cuda to unzip the file and establish the connection

tar -zxvf cudnn-8.0-linux-x64-v6. 0  sudocp cuda/include/cudnn.h/usr/local/cuda/include/sudoCP cuda/lib64/libcudnn*/usr/local/cuda/lib64/-D

View CUDNN version information:

Cat grep 2

Output

#define Cudnn_major      6#define cudnn_minor      0#define cudnn_patchlevel 21--#define cudnn_ VERSION    (cudnn_major * + Cudnn_minor * + cudnn_patchlevel)"driver_types.h "

detects if Cuda and CUDNN are installed successfully:

Go to test Catalog

// go to test Catalog:cd/usr/local/cuda-8.0/samples/1_utilities/devicequery// compilation environment   Make-J4// run:./devicequery

result = PASS indicates successful installation

iv. installation of Tensorflow-gpu

// Install the command, to install which version of TENSORFLOW-GPU, use ' tensorflow-gpu==xx ' on the line Install tensorflow-gpu==1.4// Uninstall command pip uninstall Tensorflow-gpu

Test if the GPU is installed successfully

= TF. Session (CONFIG=TF. Configproto (Log_device_placement=true))

The output has the name of the GPU, memory and other information, indicating that TensorFlow can use the GPU

Ubuntu Server Installation Tensorflow-gpu