Setting up Tensorflow-gpu on a GCP VM

ubuntu-16-04-LTS-tensorflow-gpu-gcp.md

Overview

Installation instructions for setting up a GCP VM with Tensorflow-gpu.

Instructions

GCP VM Instance

1. Select Compute Engine > VM Instances > Create Instance.
2. Specify Name, Region, and Zone. Note, the zone must support GPU instances.
3. Under Machine Type, select Customize.
   1. Choose the number of VCPUs.
   2. Choose the amount of memory.
   3. Choose the number and type of GPUs.
4. Under Boot disk, select Change.
   1. Choose Ubuntu 16.04 LTS.
   2. Specify the boot disk size.
5. Configure any other desired settings for the instance.
6. Select Create.

CUDA setup

1. SSH into the instance.
2. Download and install the CUDA 9.0 package.

   curl -O http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
   sudo dpkg -i ./cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
   sudo apt-key adv --fetch-keys http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub
   sudo apt-get update
   sudo apt-get install cuda-9-0 -y
   export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64:$LD_LIBRARY_PATH
   export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

CUDNN setup

1. Download cudnn-9.0-linux-x64-v7.2.1.38.tgz from the Nvidia website.
2. SCP the tar file to the server.

   gcloud compute --project "${PROJECT_NAME}" scp --zone "${GCP_ZONE}" ${CUDNN_TAR_FILE} ${INSTANCE_NAME}:~/

3. SSH into the instance.
4. Install CUDNN.

   tar -xvf ${CUDNN_TAR_FILE}
   sudo cp cuda/include/cudnn.h /usr/local/cuda/include
   sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
   sudo chmod a+r /usr/local/cuda/include/cudnn.h 

Tensorflow-gpu setup

1. Install tensorflow-gpu.

   sudo apt-get install python-pip python-dev python-virtualenv
   mkdir ~/tensorflow 
   cd ~/tensorflow 
   virtualenv --system-site-packages
   source ~/tensorflow/venv/bin/activate  
   pip install --upgrade tensorflow-gpu

2. Verify the installation.

   python -c "import tensorflow as tf; print(tf.__version__)"