Website • Docs • Forum • Twitter • We're Hiring

Update: I've built a quick tool based on this repo. Start running your Tensorflow project on AWS in <30seconds using Floyd. See www.floydhub.com. It's free to try out.

Happy to take feature requests/feedback and answer questions - mail me [email protected].

Setting up a Deep Learning Machine from Scratch (Software)

A detailed guide to setting up your machine for deep learning research. Includes instructions to install drivers, tools and various deep learning frameworks. This was tested on a 64 bit machine with Nvidia Titan X, running Ubuntu 14.04

There are several great guides with a similar goal. Some are limited in scope, while others are not up to date. This guide is based on (with some portions copied verbatim from):

• Caffe Installation for Ubuntu
• Running a Deep Learning Dream Machine

Table of Contents

• Basics
• Nvidia Drivers
• CUDA
• cuDNN
• Python Packages
• Tensorflow
• OpenBLAS
• Common Tools
• Caffe
• Theano
• Keras
• Torch
• X2Go

Basics

• First, open a terminal and run the following commands to make sure your OS is up-to-date

    sudo apt-get update  
    sudo apt-get upgrade  
    sudo apt-get install build-essential cmake g++ gfortran git pkg-config python-dev software-properties-common wget
    sudo apt-get autoremove 
    sudo rm -rf /var/lib/apt/lists/*
  

Nvidia Drivers

• Find your graphics card model

    lspci | grep -i nvidia
  

• Go to the Nvidia website and find the latest drivers for your graphics card and system setup. You can download the driver from the website and install it, but doing so makes updating to newer drivers and uninstalling it a little messy. Also, doing this will require you having to quit your X server session and install from a Terminal session, which is a hassle.

• We will install the drivers using apt-get. Check if your latest driver exists in the "Proprietary GPU Drivers" PPA. Note that the latest drivers are necessarily the most stable. It is advisable to install the driver version recommended on that page. Add the "Proprietary GPU Drivers" PPA repository. At the time of this writing, the latest version is 361.42, however, the recommended version is 352:

    sudo add-apt-repository ppa:graphics-drivers/ppa
    sudo apt-get update
    sudo apt-get install nvidia-352
  

• Restart your system

    sudo shutdown -r now
  

• Check to ensure that the correct version of NVIDIA drivers are installed

    cat /proc/driver/nvidia/version
  

CUDA

• Download CUDA 7.5 from Nvidia. Go to the Downloads directory and install CUDA

    sudo dpkg -i cuda-repo-ubuntu1404*amd64.deb
    sudo apt-get update
    sudo apt-get install cuda
  

• Add CUDA to the environment variables

    echo 'export PATH=/usr/local/cuda/bin:$PATH' >> ~/.bashrc
    echo 'export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
    source ~/.bashrc
  

• Check to ensure the correct version of CUDA is installed

    nvcc -V
  

• Restart your computer

    sudo shutdown -r now
  

Checking your CUDA Installation (Optional)

• Install the samples in the CUDA directory. Compile them (takes a few minutes):

    /usr/local/cuda/bin/cuda-install-samples-7.5.sh ~/cuda-samples
    cd ~/cuda-samples/NVIDIA*Samples
    make -j $(($(nproc) + 1))
  

Note: (-j $(($(nproc) + 1))) executes the make command in parallel using the number of cores in your machine, so the compilation is faster

• Run deviceQuery and ensure that it detects your graphics card and the tests pass

    bin/x86_64/linux/release/deviceQuery
  

cuDNN

• cuDNN is a GPU accelerated library for DNNs. It can help speed up execution in many cases. To be able to download the cuDNN library, you need to register in the Nvidia website at https://developer.nvidia.com/cudnn. This can take anywhere between a few hours to a couple of working days to get approved. Once your registration is approved, download cuDNN v4 for Linux. The latest version is cuDNN v5, however, not all toolkits support it yet.

• Extract and copy the files

    cd ~/Downloads/
    tar xvf cudnn*.tgz
    cd cuda
    sudo cp */*.h /usr/local/cuda/include/
    sudo cp */libcudnn* /usr/local/cuda/lib64/
    sudo chmod a+r /usr/local/cuda/lib64/libcudnn*
  

Check

• You can do a check to ensure everything is good so far using the nvidia-smi command. This should output some stats about your GPU

Python Packages

• Install some useful Python packages using apt-get. There are some version incompatibilities with using pip install and TensorFlow ( see tensorflow/tensorflow#2034)

    sudo apt-get update && apt-get install -y python-numpy python-scipy python-nose \
                                            python-h5py python-skimage python-matplotlib \
                                    python-pandas python-sklearn python-sympy
    sudo apt-get clean && sudo apt-get autoremove
    rm -rf /var/lib/apt/lists/*
  

Tensorflow

• This installs v0.8 with GPU support. Instructions below are from here

    sudo apt-get install python-pip python-dev
    sudo pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.8.0-cp27-none-linux_x86_64.whl
  

• Run a test to ensure your Tensorflow installation is successful. When you execute the import command, there should be no warning/error.

    python
    >>> import tensorflow as tf
    >>> exit()
  

OpenBLAS

• OpenBLAS is a linear algebra library and is faster than Atlas. This step is optional, but note that some of the following steps assume that OpenBLAS is installed. You'll need to install gfortran to compile it.

    mkdir ~/git
    cd ~/git
    git clone https://github.com/xianyi/OpenBLAS.git
    cd OpenBLAS
    make FC=gfortran -j $(($(nproc) + 1))
    sudo make PREFIX=/usr/local install
  

• Add the path to your LD_LIBRARY_PATH variable

    echo 'export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH' >> ~/.bashrc
  

Common Tools

• Install some common tools from the Scipy stack

    sudo apt-get install -y libfreetype6-dev libpng12-dev
    pip install -U matplotlib ipython[all] jupyter pandas scikit-image
  

Caffe

• The following instructions are from here. The first step is to install the pre-requisites

    sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler
    sudo apt-get install --no-install-recommends libboost-all-dev
    sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev
  

• Clone the Caffe repo

    cd ~/git
    git clone https://github.com/BVLC/caffe.git
    cd caffe
    cp Makefile.config.example Makefile.config
  

• If you installed cuDNN, uncomment the USE_CUDNN := 1 line in the Makefile

    sed -i 's/# USE_CUDNN := 1/USE_CUDNN := 1/' Makefile.config
  

• If you installed OpenBLAS, modify the BLAS parameter value to open

    sed -i 's/BLAS := atlas/BLAS := open/' Makefile.config
  

• Install the requirements, build Caffe, build the tests, run the tests and ensure that all tests pass. Note that all this takes a while

    sudo pip install -r python/requirements.txt
    make all -j $(($(nproc) + 1))
    make test -j $(($(nproc) + 1))
    make runtest -j $(($(nproc) + 1))
  

• Build PyCaffe, the Python interface to Caffe

    make pycaffe -j $(($(nproc) + 1))
  

• Add Caffe to your environment variable

    echo 'export CAFFE_ROOT=$(pwd)' >> ~/.bashrc
    echo 'export PYTHONPATH=$CAFFE_ROOT/python:$PYTHONPATH' >> ~/.bashrc
    source ~/.bashrc
  

• Test to ensure that your Caffe installation is successful. There should be no warnings/errors when the import command is executed.

    ipython
    >>> import caffe
    >>> exit()
  

Theano

• Install the pre-requisites and install Theano. These instructions are sourced from here

    sudo apt-get install python-numpy python-scipy python-dev python-pip python-nose g++ python-pygments python-sphinx python-nose
    sudo pip install Theano
  

• Test your Theano installation. There should be no warnings/errors when the import command is executed.

    python
    >>> import theano
    >>> exit()
  

Keras

• Keras is a useful wrapper around Theano and Tensorflow. By default, it uses Theano as the backend. See here for instructions on how to change this to Tensorflow.

    sudo pip install keras
  

Torch

• Instructions to install Torch below are sourced from here. The installation takes a little while

    git clone https://github.com/torch/distro.git ~/git/torch --recursive
    cd torch; bash install-deps;
    ./install.sh
  

X2Go

• If your deep learning machine is not your primary work desktop, it helps to be able to access it remotely. X2Go is a fantastic remote access solution. You can install the X2Go server on your Ubuntu machine using the instructions below.

    sudo apt-get install software-properties-common
    sudo add-apt-repository ppa:x2go/stable
    sudo apt-get update
    sudo apt-get install x2goserver x2goserver-xsession
  

• X2Go does not support the Unity desktop environment (the default in Ubuntu). I have found XFCE to work pretty well. More details on the supported environmens here

    sudo apt-get update
    sudo apt-get install -y xfce4 xfce4-goodies xubuntu-desktop
  

• Find the IP of your machine using

    hostname -I
  

• You can install a client on your main machine to connect to your deep learning server using the above IP. More instructions here depending on your Client OS