CUDA Installation Based on RHEL
===============================
Overview
--------
CUDA (Compute Unified Device Architecture) is a parallel computing platform and programming model created by NVIDIA. It can be used by the graphics processing units (GPUs) for general purpose processing.
xCAT supports CUDA installation for Ubuntu 14.04.3 and RHEL 7.2LE on PowerNV (Non-Virtualized) for both diskless and diskful nodes. The CUDA packages provided by NVIDIA include both the runtime libraries for computing and development tools for programming and monitoring. The full package set is very large, so in xCAT, it's suggested that the packages be split into two package sets:
#. **cudaruntime** package set
#. **cudafull** package set
It's suggested to only installing the **cudaruntime** package set on the Compute Nodes (CNs), and the **cudafull** package set on the Management Node or the monitor/development nodes.
In this documentation, xCAT will provide CUDA installation based on RHEL 7.2LE running on IBM Power Systems S822LC nodes.
CUDA Repository
---------------
The NIVIDIA CUDA Toolkit is available at http://developer.nvidia.com/cuda-downloads. User can download the CUDA Toolkit based on the target platform.
Prepare a local repository directory which contains all the CUDA packages and repository meta-data. User can either
* Create a repository on the Management node installing the CUDA toolkit: ::
mkdir -p /install/cuda-repo/cuda-7-5
rpm -ivh cuda-repo-rhel7-7-5-local-7.5-18.ppc64le.rpm
cp /var/cuda-repo-7-5-local/* /install/cuda-repo/cuda-7-5
cd /install/cuda-repo/cuda-7-5
createrepo .
* Create a repository on the Management node without installing the CUDA toolkit: ::
mkdir -p /tmp/cuda
cd /tmp/cuda
rpm2cpio /root/cuda-repo-rhel7-7-5-local-7.5-18.ppc64le.rpm | cpio -i -d
cp /tmp/cuda/var/cuda-repo-7-5-local/* /install/cuda-repo/cuda-7-5
cd /install/cuda-repo/cuda-7-5
createrepo .
The NVIDIA driver RPM packages depend on other external packages, such as DKMS and maybe EPEL (firestone node doesn't need this package). Users need to download those package to the directory. ::
mkdir -p /install/cuda-repo/cuda-deps
ls -ltr /install/cuda-repo/cuda-deps
-rw-r--r-- 1 root root 79048 Oct 5 10:58 dkms-2.2.0.3-30.git.7c3e7c5.el7.noarch.rpm
cd /install/cuda-repo/cuda-deps
createrepo .
CUDA osimage
------------
User can generate a new CUDA osimage object based on another osimage definition or just modify an existing osimage. xCAT provides some sample CUDA pkglist files:
* diskful provisioning in ``/opt/xcat/share/install/rh/`` for ``cudafull`` and ``cudaruntime``: ::
#cat /opt/xcat/share/xcat/install/rh/cudafull.rhels7.pkglist
#INCLUDE:compute.rhels7.pkglist#
#For Cuda 7.5
kernel-devel
gcc
pciutils
dkms
cuda
#cat /opt/xcat/share/xcat/install/rh/cudaruntime.rhels7.pkglist
#INCLUDE:compute.rhels7.pkglist#
#For Cuda 7.5
kernel-devel
gcc
pciutils
dkms
cuda-runtime-7-5
* diskless provisioning in ``/opt/xcat/share/xcat/netboot/rh`` for ``cudafull`` and ``cudaruntime``: ::
#cat /opt/xcat/share/xcat/netboot/rh/cudafull.rhels7.ppc64le.pkglist
#INCLUDE:compute.rhels7.ppc64.pkglist#
#For Cuda 7.5
kernel-devel
gcc
pciutils
dkms
cuda
#cat /opt/xcat/share/xcat/netboot/rh/cudaruntime.rhels7.ppc64le.pkglist
#INCLUDE:compute.rhels7.ppc64.pkglist#
#For Cuda 7.5
kernel-devel
gcc
pciutils
dkms
cuda-runtime-7-5
**NOTE: After CUDA are installed, the nodes require a reboot**
* For the diskful installation, the CUDA packages should be included in the ``pkglist`` field so a reboot happens automatically after the OS is installed.
* For the diskless installation, the CUDA package can be included either in ``otherpkglist`` or ``pktlist`` fields.
The following are some sample osimage definitions:
* The diskful cudafull installation osimage object. ::
#lsdef -t osimage rhels7.2-ppc64le-install-cudafull
Object name: rhels7.2-ppc64le-install-cudafull
imagetype=linux
osarch=ppc64le
osdistroname=rhels7.2-ppc64le
osname=Linux
osvers=rhels7.2
otherpkgdir=/install/post/otherpkgs/rhels7.2/ppc64le
pkgdir=/install/rhels7.2/ppc64le,/install/cuda-repo
pkglist=/opt/xcat/share/xcat/install/rh/cudafull.rhels7.pkglist
profile=compute
provmethod=install
template=/opt/xcat/share/xcat/install/rh/compute.rhels7.tmpl
* The diskful cudaruntime installation osimage object. ::
#lsdef -t osimage rhels7.2-ppc64le-install-cudaruntime
Object name: rhels7.2-ppc64le-install-cudaruntime
imagetype=linux
osarch=ppc64le
osdistroname=rhels7.2-ppc64le
osname=Linux
osvers=rhels7.2
otherpkgdir=/install/post/otherpkgs/rhels7.2/ppc64le
pkgdir=/install/rhels7.2/ppc64le,/install/cuda-repo
pkglist=/opt/xcat/share/xcat/install/rh/cudairuntime.rhels7.pkglist
profile=compute
provmethod=install
template=/opt/xcat/share/xcat/install/rh/compute.rhels7.tmpl
* The diskless cudafull installation osimage object. ::
#lsdef -t osimage rhels7.2-ppc64le-netboot-cudafull
Object name: rhels7.2-ppc64le-netboot-cudafull
imagetype=linux
osarch=ppc64le
osdistroname=rhels7.2-ppc64le
osname=Linux
osvers=rhels7.2
otherpkgdir=/install/post/otherpkgs/rhels7.2/ppc64le
permission=755
pkgdir=/install/rhels7.2/ppc64le,/install/cuda-repo
pkglist=/opt/xcat/share/xcat/netboot/rh/cudafull.rhels7.ppc64le.pkglist
postinstall=/opt/xcat/share/xcat/netboot/rh/compute.rhels7.ppc64le.postinstall
profile=compute
provmethod=netboot
rootimgdir=/install/netboot/rhels7.2/ppc64le/compute
* The diskless cudaruntime installation osimage object. ::
#lsdef -t osimage rhels7.2-ppc64le-netboot-cudaruntime
Object name: rhels7.2-ppc64le-netboot-cudaruntime
imagetype=linux
osarch=ppc64le
osdistroname=rhels7.2-ppc64le
osname=Linux
osvers=rhels7.2
otherpkgdir=/install/post/otherpkgs/rhels7.2/ppc64le
permission=755
pkgdir=/install/rhels7.2/ppc64le,/install/cuda-repo
pkglist=/opt/xcat/share/xcat/netboot/rh/cudaruntime.rhels7.ppc64le.pkglist
postinstall=/opt/xcat/share/xcat/netboot/rh/compute.rhels7.ppc64le.postinstall
profile=compute
provmethod=netboot
rootimgdir=/install/netboot/rhels7.2/ppc64le/compute
Deployment of CUDA node
-----------------------
* To provision diskful nodes: ::
nodeset <node> osimage=rhels7.2-ppc64le-install-cudafull
rsetboot <node> net
rpower <node> boot
* To provision diskless nodes: ::
genimage rhels7.2-ppc64le-netboot-cudafull
packimage rhels7.2-ppc64le-netboot-cudafull
nodeset <node> osimage=rhels7.2-ppc64le-netboot-cudafull
rsetboot <node> net
rpower <node> boot
Verification of CUDA Installation
---------------------------------
**NOTE** For ``cudaruntime`` installation, it only provide the basic libraries that can bee used by other applications which works with GPU. The following verification will not apply to ``cudaruntime`` installations.
After compute node booted, The Environment variable has to be set in order to use the CUDA toolkits. The PATH variable needs to include ``/usr/local/cuda-7.5/bin`` and LD_LIBRARY_PATH variable needs to contain ``/usr/local/cuda-7.5/lib64`` on a 64-bit system, and ``/usr/local/cuda-7.5`` on a 32-bit system.
* To change the environment variables for 64-bit operating systems ::
export PATH=/usr/local/cuda-7.5/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-7.5/lib64:$LD_LIBRARY_PATH
* To change the environment variable for 32-bit operating systems ::
export PATH=/usr/local/cuda-7.5/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-7.5/lib:$LD_LIBRARY_PATH
After Environment variables are set correctly, user can verify the CUDA installation by
* Verify the Driver Version ::
#cat /proc/driver/nvidia/version
NVRM version: NVIDIA UNIX ppc64le Kernel Module 352.39 Fri Aug 14 17:10:41 PDT 2015
GCC version: gcc version 4.8.5 20150623 (Red Hat 4.8.5-4) (GCC)
* The version of the CUDA Toolkits ::
#nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2015 NVIDIA Corporation
Built on Tue_Aug_11_14:31:50_CDT_2015
Cuda compilation tools, release 7.5, V7.5.17
* Compiling the Examples, then can run a `deviceQuery` or `bandwidthTest` or other commands under the bin directory to ensure the system and the CUDA-capable device are able to communicate correctly ::
# mkdir -p /tmp/cuda
# cuda-install-samples-7.5.sh /tmp/cuda
# cd /tmp/cuda/NVIDIA_CUDA-7.5_Samples
# make
# cd bin/ppc64le/linux/release
# ./deviceQuery
./deviceQuery Starting...
CUDA Device Query (Runtime API) version (CUDART static linking)
Detected 4 CUDA Capable device(s)
Device 0: "Tesla K80"
CUDA Driver Version / Runtime Version 7.5 / 7.5
CUDA Capability Major/Minor version number: 3.7
Total amount of global memory: 11520 MBytes (12079136768 bytes)
(13) Multiprocessors, (192) CUDA Cores/MP: 2496 CUDA Cores
GPU Max Clock rate: 824 MHz (0.82 GHz)
Memory Clock rate: 2505 Mhz
Memory Bus Width: 384-bit
L2 Cache Size: 1572864 bytes
............
deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 7.5, CUDA Runtime Version = 7.5, NumDevs = 4, Device0 = Tesla K80, Device1 = Tesla K80, Device2 = Tesla K80, Device3 = Tesla K80
Result = PASS
# ./bandwidthTest
[CUDA Bandwidth Test] - Starting...
Running on...
Device 0: Tesla K80
Quick Mode
Host to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 7765.1
Device to Host Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 7759.6
Device to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 141485.3
Result = PASS
NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.
* The tool `nvidia-smi` providied by NVIDIA driver can be used to do GPU management and monitoring. ::
#nvidia-smi -q
==============NVSMI LOG==============
Timestamp : Mon Oct 5 13:43:39 2015
Driver Version : 352.39
Attached GPUs : 4
GPU 0000:03:00.0
Product Name : Tesla K80
Product Brand : Tesla
...........................