install tensorflow-gpu in docker on RHEL7

0. make sure that you have installed cuda

Check environment

1. Verify you have a CUDA-capable GPU, run the command below and you will see the graphic info

   lspci | grep -i nvidia

2. Verify you have a supported version of Linux, check the version of Linux to install right version of CUDA

   uname -m && cat /etc/*release

3. Verify the system has gcc Installed. Or install gcc firstly before installing CUDA.

   gcc --version

Install steps (by runfile)

1. (optional)Install kernel headers and development packages for the currently running kernel. If you have installed these packages, you could skip this step.

   sudo yum install kernel-devel-
   (uname -r) kernel-headers-
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
   
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
    
   
   (uname -r)

2. Disable the Nouveau drivers
   a. Create a file at /etc/modprobe.d/blacklist-nouveau.conf with below contents

   blacklist nouveau
   options nouveau modeset=0

   b. Regenerate the kernel initramfs

   sudo dracut –force

3. (optional)Reboot to the graphical interface, and change to command interface with command sudo init 3. This step is to install graphic card driver.
4. Run the installer (runfile) with command sudo sh cuda__linux.run. If you choose to install graphic card driver, you must run step 3 first.
5. Create an xorg.conf file to use the NVIDIA GPU for display:

   sudo nvidia-xconfig

6. Reboot the system to load the graphical interface
7. Set up the development environment by modifying the PATH and LD_LIBRARY_PATH variables

   export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}
   export LD_LIBRARY_PATH=/usr/local/cuda/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

8. Run the nbody sample to make sure whether CUDA works or not.

   cd /usr/local/cuda/samples/5_Simulations/nbody
   sudo make
   ./nbody

1. install nvidia-docker

Refer to https://github.com/nvidia/nvidia-docker/wiki/Installation-(version-2.0)

2. pull image from tensorflow/tensorflow

$ docker pull tensorflow/tensorflow:latest-gpu-py3

3. run docker container

$ docker run -it --runtime=nvidia --rm tensorflow/tensorflow:latest-gpu-py3

4. check if it works or not

$ python -c "import tensorflow as tf; tf.enable_eager_execution(); print(tf.reduce_sum(tf.random_normal([1000, 1000])))"

5. reference

[1] https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html
[2] https://github.com/nvidia/nvidia-docker/wiki
[3] https://www.tensorflow.org/install/docker

附： docker-compose.yml

version '2.3'
services:
    tensorflow-gpu:
        image: tensorflow/tensorflow:latest-gpu-py3
        runtime: nvidia
        environment:
            - NVIDIA_VISIBLE_DEVICES=all