How to install MXnet in Ubuntu 14.04 LTS

Hardware Configutation

S7050GM2NR

CPU: E5-2650 v3 x1

RAM: DDR4 8GB x4

OS: Ubuntu 14.04 LTS Desktop 64bit

GPU: Nvidia Tesla K80 x1 (Driver: v352.93)

Setup Environment

1.      Install Ubuntu 14.04 LTS Desktop

2.      Boot to text mode

vi /etc/default/grub

mark GRUB_CMDLINE_LINUX_DEFAULT=”quiet”

change GRUB_CMDLINE_LINUX=”text”

umark GRUB_TERMINAL=console

save

update grub via command:

update-grub

reboot system

Install K80 Driver

     ./NVIDIA-Linux-x86_64-352.93.run

     apt-get update

     apt-get install –y build-essential git libblas-dev libopencv-dev libatlas-base-dev

Download MXnet

1.      git clone –recursive https://github.com/dmlc/mxnet

Install CUDA 7.5 environment

     wget http://developer.download.nvidia/compute/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb

    dpkg –I cuda-repo-ubuntu1404_7.5-18_amd64.deb

You can also type follows command if you have CUDA 7.5 source file already

    dpkg –I cuda-repo-ubuntu1404-7.5-local_7.5-18_amd64.deb

    apt-get update

    apt-get install cuda

Set CUDA environment

     export PATH=/usr/local/cuda-7.5/bin:$PATH

     export LD_LIBRARY_PATH=/usr/local/cuda-7.5/lib64:$LD_LIBRARY_PATH

Check CUDA environment

     printenv LD_LIBRARY_PATH

     nvidia-smi

Set CUDA for MXnet

     cd /temp/mxnet

     cp /temp/mxnet/make/config.mk /temp/mxnet/.

Modify config.mk

     USE_CUDA = 1

     USE_CUDA_PATH = /usr/local/cuda

     USE_BLAS = atlas

Complie MXNET

     make –j4

Install Python

     apt-get install python-setuptools python-dev python-pip

     pip install numpy

     cd python; python setup.py install

Test MXnet

     cd /temp/mxnet/example/image-classificatopm

     python train_mnist.py            <= use CPU only

     python train_mnist.py --gpus “0,1” <=use GPUs