Machine Learning Corner
    Menu
    • Link Icon Link Icon Link Icon ERPComputing.com
      • External link opens in new tab or window Link Icon Link Icon Link Icon Menu

        How to setup deep learning Python (3) environment with NVidia GPU

        Dr. George Jen


        If your computer has NVIDIA GPU,  chances are you do, and runs Ubuntu, following are steps to setup:

         

        Latest version CUDA

        ·        NVIDIA GPU Driver

        ·        cuDNN (NVidia CUDA Deep Neural Network library)

        ·        Anaconda3 Python 3.6

        ·        Tensorflow (for GPU)

        ·        Keras (for GPU)

        ·        CNTK (for GPU)

        ·        Theano

        ·        PyTorch

        ·        along with other required software libraries

         

        Then you have an environment to pursue machine learning running on GPU, which can be typically more than 10 times faster than running on CPU.

         

        This demo environment is under Ubuntu 18.04 (if you have different version of Ubuntu, there could be slight difference in setting things up; if you have other Linux, such as Redhat or CentOS, procedures can be more different)

         

        1. Check OS release:

         

        cat /etc/*release*

         

        DISTRIB_ID=Ubuntu

        DISTRIB_RELEASE=18.04

        DISTRIB_CODENAME=bionic

        DISTRIB_DESCRIPTION="Ubuntu 18.04 LTS"

        NAME="Ubuntu"

        VERSION="18.04 LTS (Bionic Beaver)"

        ID=ubuntu

        ID_LIKE=debian

        PRETTY_NAME="Ubuntu 18.04 LTS"

        VERSION_ID="18.04"

        HOME_URL="https://www.ubuntu.com/"

        SUPPORT_URL="https://help.ubuntu.com/"

        BUG_REPORT_URL="https://bugs.launchpad.net/ubuntu/"

        PRIVACY_POLICY_URL="https://www.ubuntu.com/legal/terms-and-policies/privacy-policy"

        VERSION_CODENAME=bionic

        UBUNTU_CODENAME=bionic

         

        2. Confirm NVidia GPU plugged in, using command lspci -v to show it.

         

        lspci -v | grep VGA

         

        02:00.0 VGA compatible controller: NVIDIA Corporation GK208 [GeForce GT 710B] (rev a1) (prog-if 00 [VGA controller])

         

        3. Pre-requisite:

         

        developer tools such as g++, gcc, git and make etc are already installed. If not, you can install them

         

        sudo apt update

        sudo apt upgrade

        sudo apt install build-essential

        sudo apt install git

         

        4.  Clean up/remove previous NVIDIA software including drivers:

         

        sudo apt-get remove --purge nvidia* cuda-drivers libcuda1-396 cuda-runtime-9-2 cuda-9.2 cuda-demo-suite-9-2  cuda

         

        5. Install latest CUDA kit:

         

        For detail about CUDA:

         

        https://developer.nvidia.com/cuda-zone

         

        sudo apt install nvidia-cuda-toolkit

         

        become root to make /opt/local/cuda/bin/nvcc available for GPU driver install:

        sudo –i

        cd /usr/local

        mkdir cuda

        cd cuda

        mkdir bin

        ln -s /usr/bin/nvcc nvcc

         

        6. Install latest NVIDIA GPU driver:

         

        sudo add-apt-repository ppa:graphics-drivers/ppa

        sudo apt update

        ubuntu-drivers devices

        sudo ubuntu-drivers autoinstall

         

        Reboot to load Nvidia driver installed, when comes back, run command "nvidia-smi" to verify enabled NVIDIA GPU


         

         

        Install CUDA SDK kit include source sample code:

         

        Mkdir ~/cuda

        cd cuda

        wget External link opens in new tab or windowhttps://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda-repo-ubuntu1804-10-0-local-10.0.130-410.48_1.0-1_amd64

         sudo dpkg -i ./cuda-repo-ubuntu1804-10-0-local-10.0.130-410.48_1.0-1_amd64

         sudo apt-key add /var/cuda-repo-10-0-local-10.0.130-410.48/7fa2af80.pub

         sudo apt update

         sudo apt install cuda

         

        becomes root:

        sudo –i

        cd /usr/local/cuda-10.0/samples

         

        Build all sample codes:

         

        make

         

        Reboot. After come back, run deviceQuery

         

        /usr/local/cuda-10.0/samples/bin/x86_64/linux/release# /usr/local/cuda-10.0/samples/bin/x86_64/linux/release/deviceQuery


         

         

        7. Install cuDNN (The NVIDIA CUDA Deep Neural Network library):

         

        For detail, see:

         

        https://developer.nvidia.com/cudnn

         

        Download cdDNN, you need to register a developer membership with NVIDIA, which is free before download below deb files. Once downloaded, run below to install.

         

        sudo dpkg -i libcudnn7_7.1.4.18-1+cuda9.2_amd64.deb

        sudo dpkg -i libcudnn7-dev_7.1.4.18-1+cuda9.2_amd64.deb

        sudo dpkg -i libcudnn7-doc_7.1.4.18-1+cuda9.2_amd64.deb

         

        cp -r /usr/src/cudnn_samples_v7/ $HOME

        cd  $HOME/cudnn_samples_v7/mnistCUDNN

        make clean && make

         

        Test mnist number recognition using cuDNN

        ./mnistCUDNN

         

        ~/cudnn_samples_v7/mnistCUDNN$ ./mnistCUDNN

         

        cudnnGetVersion() : 7104 , CUDNN_VERSION from cudnn.h : 7104 (7.1.4)

        Host compiler version : GCC 7.3.0

        There are 1 CUDA capable devices on your machine :

        device 0 : sms  1  Capabilities 3.5, SmClock 954.0 Mhz, MemSize (Mb) 2000, MemClock 900.0 Mhz, Ecc=0, boardGroupID=0

        Using device 0

         

        Testing single precision

        Loading image data/one_28x28.pgm

        Performing forward propagation ...

        Testing cudnnGetConvolutionForwardAlgorithm ...

        Fastest algorithm is Algo 2

        Testing cudnnFindConvolutionForwardAlgorithm ...

        ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.054144 time requiring 100 memory

        ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.056032 time requiring 0 memory

        ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.123040 time requiring 57600 memory

        ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.474048 time requiring 207360 memory

        ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 1.869824 time requiring 2057744 memory

        Resulting weights from Softmax:

        0.0000000 0.9999399 0.0000000 0.0000000 0.0000561 0.0000000 0.0000012 0.0000017 0.0000010 0.0000000

        Loading image data/three_28x28.pgm

        Performing forward propagation ...

        Resulting weights from Softmax:

        0.0000000 0.0000000 0.0000000 0.9999288 0.0000000 0.0000711 0.0000000 0.0000000 0.0000000 0.0000000

        Loading image data/five_28x28.pgm

        Performing forward propagation ...

        Resulting weights from Softmax:

        0.0000000 0.0000008 0.0000000 0.0000002 0.0000000 0.9999820 0.0000154 0.0000000 0.0000012 0.0000006

         

        Result of classification: 1 3 5

         

        Test passed!

         

        8. Install Anaconda Python (3.6.5)

         

        Download Anaconda Python 3 from:

         

        https://www.anaconda.com/download/

         

        Once downloaded, run .sh file:

         

        sh ./Anaconda3-5.2.0-Linux-x86_64.sh

        conda update conda

        source ~/.bashrc

         

        9. Install Tensorflow (GPU version) (TensorFlow™ is an open source software library for high performance numerical computation.) using conda commandline:

         

        conda install -c anaconda tensorflow-gpu

         

        Test it with

         

        python -c "import tensorflow"

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcudnn.so.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcufft.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcuda.so.1 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcurand.so.7.5 locally

         

         

        10. Install keras (GPU) (Keras is a high-level neural networks API, written in Python and capable of running on top of TensorFlow, CNTK, or Theano.)

         

        conda install -c anaconda "keras-gpu"

         

        Tested it:

         

        python -c "import keras"

        ..anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.

          from ._conv import register_converters as _register_converters

        Using TensorFlow backend.

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcudnn.so.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcufft.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcuda.so.1 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcurand.so.7.5 locally

         

        11. Install Theano (A Python library that allows you to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays) using conda commandline:

         

        conda install -c anaconda theano

         

        Verify installation by:

         

        python -c "import theano"

         

        12. Install Pytorch (Tensors and Dynamic neural networks in Python with strong GPU acceleration) using conda commandline:

         

        conda install -c anaconda pytorch

         

        Verify installation by:

         

        python -c "import torch"

         

        13. Install CNTK (Microsoft Cognitive Toolkit), GPU version:

         

        sudo apt-get install openmpi-bin

        pip install cntk-gpu

        Install Open MPI Library (Open Message Passing Interface Library)

         

        Get the installation sources:

         

        wget https://www.open-mpi.org/software/ompi/v1.10/downloads/openmpi-1.10.3.tar.gz

        tar -xzvf ./openmpi-1.10.3.tar.gz

        cd openmpi-1.10.3

        ./configure --prefix=/usr/local/mpi

        make -j all

        sudo make install

         

        Verify installation

         

        python -c "import cntk"

         

        14. Other libraries and modules:

         

        Protobuf Protocol Buffers for serialization. For installation please follow these steps:

         

        cd ~

        sudo apt-get install autoconf automake libtool curl make g++ unzip

        wget https://github.com/google/protobuf/archive/v3.1.0.tar.gz

        tar -xzf v3.1.0.tar.gz

        cd protobuf-3.1.0

        ./autogen.sh

        ./configure CFLAGS=-fPIC CXXFLAGS=-fPIC --disable-shared --prefix=/usr/local/protobuf-3.1.0

        make -j $(nproc)

        sudo make install

         

        15. install zlib1g-dev

         

        sudo apt-get install zlib1g-dev

         

        16. LIBZIP

         

        wget http://nih.at/libzip/libzip-1.1.2.tar.gz

        tar -xzvf ./libzip-1.1.2.tar.gz

        cd libzip-1.1.2

        ./configure

        make -j all

        sudo make install

         

        17. Boost Library

        The Boost Library is a prerequisite for building the Microsoft Cognitive Toolkit.

         

        sudo apt-get install libbz2-dev

        sudo apt-get install python-dev

        wget -q -O - https://sourceforge.net/projects/boost/files/boost/1.60.0/boost_1_60_0.tar.gz/download | tar -xzf -

        cd boost_1_60_0

        ./bootstrap.sh --prefix=/usr/local/boost-1.60.0

        sudo ./b2 -d0 -j"$(nproc)" install  

         

        18.  Install h5py, requires for saving model

         

        conda install -c anaconda h5py

         

        19.  Now you can test import all these modules:

         

        python -c "import torch"

        python -c "import tensorflow"

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcudnn.so.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcufft.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcuda.so.1 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcurand.so.7.5 locally

        python -c "import theano"

        python -c "import keras"

        .../anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.

          from ._conv import register_converters as _register_converters

        Using TensorFlow backend.

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcudnn.so.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcufft.so.7.5 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcuda.so.1 locally

        I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcurand.so.7.5 locally

        python -c "import cntk"

         

        Now your machine is now a Python work horse doing machine learning on NVIDIA GPU that could have hundreds or thousands of CUDA cores depending on how much you pay for the GPU card(s).

         


         

         


          close lightbox