[PYTHON] Put TensorFlow in P2 instance with pip3

Just put Python3 in your p2 instance and pip into TensorFlow. Play with the Tesla K80.

Requirements

TensorFlow is as usual

• Python 2.7 and Python 3.3+
• Cuda toolkit >= 7.0
• cuDNN >= v3

As for compute capability, Tesla K80 has 3.7, so you can rest assured. CUDA GPUs | NVIDIA Developer

Introduction method

Set up an instance

TensorFlow seems to like Ubuntu, so I searched for AMI from Ubuntu official. Ubuntu Amazon EC2 AMI Finder

I tried the hvm: ebs-ssd guy on Ubuntu 16.10.

Put what you need

Classic

sudo apt-get update
sudo apt-get upgrade

Python3 (with 3.5.2+)

sudo apt-get install python3-pip python3-dev

The latest version of CUDA is 8. You can find the commands you can bring with wget by following the flow of CUDA 8.0 Downloads | NVIDIA Developer.

sudo apt-get wget
wget https://developer.nvidia.com/compute/cuda/8.0/prod/local_installers/cuda-repo-ubuntu1604-8-0-local_8.0.44-1_amd64-deb
sudo dpkg -i cuda-repo-ubuntu1604-8-0-local_8.0.44-1_amd64-deb
sudo apt-get update
rm cuda-repo-ubuntu1604-8-0-local_8.0.44-1_amd64-deb
sudo apt-get install cuda

For cuDNN, you have to create an account and bring the cuDNN Library for Linux from cuDNN Download | NVIDIA Developer to upload it.

tar xvzf cudnn-8.0-linux-x64-v5.1-tgz
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 /usr/local/cuda/lib64/libcudnn*

cd /usr/local/cuda/lib64
sudo ln -s libcudart.so libcudart.so.7.5

The last two lines are ʻImportError: libcudart.so.7.5: cannot open shared object file: No such file or directory`. Below CUDA 7.5 fails with pip install and docker (Ubuntu 14.04) · Issue # 20 · tensorflow / tensorflow.

Leave it through the path.

export CUDA_HOME=/usr/local/cuda
export CUDA_ROOT=/usr/local/cuda
export PATH=$PATH:$CUDA_ROOT/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_ROOT/lib64

I put the TensorFlow main body with GPU.

export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.11.0rc0-cp35-cp35m-linux_x86_64.whl
pip3 install --ignore-installed --upgrade $TF_BINARY_URL

References

Installing TensorFlow on an AWS EC2 P2 GPU Instance - ExpressionFlowExpressionFlow This is the place to start from the source code.