Difference between revisions of "HPC:CUDA"

Revision as of 09:00, 22 April 2018

CUDA Development

• HPC:CUDA Development Documentation & Tutorials

CUDA installation process in worker nodes

• Main doc: http://docs.nvidia.com/cuda/index.html
• Installation docs: https://developer.nvidia.com/cuda-downloads
• Full installation guide [1]

Base Installation

This procedure is intended for CentOS 7.3, but some updates cited below allows it to configure a CentOS 7.4 worker node:

• Install EPEL (Extra Packages for Enterprise Linux) - Result: OK
  • Note: this fulfill the requirement of the dkms package for CentOS 7
  • Site: https://developer.nvidia.com/cuda-zone

# wget http://dl.fedoraproject.org/pub/epel/7/x86_64/e/epel-release-7-9.noarch.rpm
# rpm -ivh epel-release-7-9.noarch.rpm

• 'Update (alternative) for CentOS 7.4: it is necessary to use epel v. 7.11 and to install nVidia graphics driver'

# wget http://dl.fedoraproject.org/pub/epel/7/x86_64/e/epel-release-7-11.noarch.rpm
# rpm -ivh epel-release-7-11.noarch.rpm

• • Download graphics driver NVIDIA-Linux-x86_64-390.46.run from http://www.nvidia.com/Download/index.aspx

# bash NVIDIA-Linux-x86_64-390.46.run

• Get CUDA local installation RPM file - Result: OK

# wget https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda-repo-rhel7-8-0-local-ga2-8.0.61-1.x86_64-rpm
# mv cuda-repo-rhel7-8-0-local-ga2-8.0.61-1.x86_64-rpm cuda-repo-rhel7-8-0-local-ga2-8.0.61-1.x86_64.rpm

• Install CUDA package - Result: OK

# rpm -i cuda-repo-rhel7-8-0-local-ga2-8.0.61-1.x86_64.rpm
# yum clean all
# yum install cuda

Configuration

• Create or edit the file /etc/modprobe.d/blacklist-nouveau.conf with the following contents:

blacklist nouveau
options nouveau modeset=0

• Regenerate the kernel initramfs:

# dracut --force

• Create a file /etc/profile.d/cuda.sh containing

export PATH=/usr/local/cuda-8.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

Test

It seems it's necessary to restart each machine, because the system must load the cuda driver (without the noveau driver) - Reboot OK (node '11')

• Create user testing

# adduser testing
# passwd testing

• Log in as user testing

# su testing

• Create testing directory

$ cd
$ mkdir cuda
$ mkdir cuda/testing

• Copy testing examples

$ cd cuda/testing
$ cuda-install-samples-8.0.sh ./

• Enter an arbitrary example and run

$ cd NVIDIA_CUDA-8.0_Samples/0_Simple/matrixMul
$ make
$ ./matrixMul