Skip to content

Latest commit

 

History

History
142 lines (95 loc) · 3.97 KB

README.md

File metadata and controls

142 lines (95 loc) · 3.97 KB

Intro

This short guide shows how to install Nvidia drivers and CUDA for GRID K2 hardware. For more information about GPUs on SURF HPC Cloud please visit SURF HPC Documentation.

We have 2 methods to install Nvidia drivers and CUDA.

  • Using ansible playbook
  • Following the steps and installing manually

Method 1: Installation using ansible playbook

The command below runs the ansible-playbook and installs all necassary software. This ansible-playbook is specifically written for SURF HPC Cloud platform and GRID K2 hardware. However, it can easily be adapted for different platforms and graphics cards.

Copy example hosts file and make necassary changes.

cp hosts.example hosts
docker run --rm -ti -v $PWD:/data --workdir=/data ansible/ansible-runner ansible-playbook playbook-install-cuda-gridk2.yml

id_rsa is your rivate ssh key and hosts is the file which has the connection details of the server.

Method 2: Manual installation

Requirements

CUDA currently officially supports only two versions of Ubuntu: 18.04 and 16.04. This instructions were tested on Ubuntu 18.04.

System info

Distribution info

lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 18.04.4 LTS
Release:        18.04
Codename:       bionic

Kernel version

uname -a
Linux packer-Ubuntu-18 4.15.0-101-generic #102-Ubuntu SMP Mon May 11 10:07:26 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux

GPU hardware information

lspci | grep -i nvidia
01:01.0 VGA compatible controller: NVIDIA Corporation GK104GL [GRID K2] (rev a1)

Pre install

For Grid K2 card we will need CUDA 8.0. CUDA 8.0 only works with only gcc 5.0 so it should be installed before.

To decide what version of CUDA and Nvidia drivers you need, please check the links below.

See what drivers you need: https://www.nvidia.com/Download/index.aspx?lang=en-us

Check compatibility first: https://docs.nvidia.com/deploy/cuda-compatibility/index.html

apt install gcc-5 g++-5

Install Nvidia drivers

Download Nvidia driver (version 367) and install it.

wget http://us.download.nvidia.com/XFree86/Linux-x86_64/367.134/NVIDIA-Linux-x86_64-367.134.run
sh ./NVIDIA-Linux-x86_64-367.134.run --accept-license  -s

Install CUDA

Download CUDA 8.0 installer

# wget https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda_8.0.61_375.26_linux-run

Download Patch release:

wget https://developer.nvidia.com/compute/cuda/8.0/Prod2/patches/2/cuda_8.0.61.2_linux-run

While installing CUDA, we had some issues related to Perl scripts. See: https://forums.developer.nvidia.com/t/cant-locate-installutils-pm-in-inc/46952/10

These commands solves the Perl issues.

sh ./cuda_8.0.61_375.26_linux-run  --tar mxvf
cp InstallUtils.pm /usr/lib/x86_64-linux-gnu/perl-base/
export $PERL5LIB
rm -rf InstallUtils.pm cuda-installer.pl run_files uninstall_cuda.pl

After fixing the Perl issue, we can install CUDA.

sh ./cuda_8.0.61_375.26_linux-run --silent --samples --toolkit --override --verbose

Environment variables

In order to be able to use CUDA, we need to change our environment variables.

Add the lines below to .profile file.

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

Test the installation

CUDA compiler

Check the CUDA compiler version.

nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Nov__3_21:07:56_CDT_2017
Cuda compilation tools, release 9.1, V9.1.85

Example code

See gpu-houston for a simple example code.