Nvidia CUDA Toolkit supports CentOS 7 and it is relatively simple to install it. Nvidia provides three types of installation – a big setup file, a big rpm file and an official Nvidia repository, which we are going to use it in this article. The Nvidia repository contains the Nvidia video driver for the Nvidia video cards like GeForce, GTX, RTX and so on. You may need CUDA Toolkit if you are a game developer or you want to build yourself some of the mining software like XMR-STAK.
In this article, we are going to use the NVIDIA official repository for CUDA and the video driver module. There are other ways to install CUDA, which are not the purpose of this article. Using an official repository is the best practice for installing software on your system.
STEP 1) Update and install the NVIDIA official repository.
yum update -y yum install -y yum-utils yum-config-manager --add-repo http://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo
The real output in our server:
[root@localhost ~]# yum -y update Loaded plugins: fastestmirror Loading mirror speeds from cached hostfile * base: mirror.host.ag * extras: mirror.host.ag * updates: mirror.host.ag No packages marked for update [root@localhost ~]# yum install -y yum-utils Loaded plugins: fastestmirror Loading mirror speeds from cached hostfile * base: mirror.host.ag * extras: mirror.host.ag * updates: mirror.host.ag Resolving Dependencies --> Running transaction check ---> Package yum-utils.noarch 0:1.1.31-52.el7 will be installed --> Processing Dependency: python-kitchen for package: yum-utils-1.1.31-52.el7.noarch --> Processing Dependency: libxml2-python for package: yum-utils-1.1.31-52.el7.noarch --> Running transaction check ---> Package libxml2-python.x86_64 0:2.9.1-6.el7_2.3 will be installed ---> Package python-kitchen.noarch 0:1.1.1-5.el7 will be installed --> Processing Dependency: python-chardet for package: python-kitchen-1.1.1-5.el7.noarch --> Running transaction check ---> Package python-chardet.noarch 0:2.2.1-3.el7 will be installed --> Finished Dependency Resolution Dependencies Resolved ============================================================================================================================================================================ Package Arch Version Repository Size ============================================================================================================================================================================ Installing: yum-utils noarch 1.1.31-52.el7 base 121 k Installing for dependencies: libxml2-python x86_64 2.9.1-6.el7_2.3 base 247 k python-chardet noarch 2.2.1-3.el7 base 227 k python-kitchen noarch 1.1.1-5.el7 base 267 k Transaction Summary ============================================================================================================================================================================ Install 1 Package (+3 Dependent packages) Total download size: 862 k Installed size: 4.3 M Downloading packages: (1/4): python-chardet-2.2.1-3.el7.noarch.rpm | 227 kB 00:00:00 (2/4): yum-utils-1.1.31-52.el7.noarch.rpm | 121 kB 00:00:00 (3/4): libxml2-python-2.9.1-6.el7_2.3.x86_64.rpm | 247 kB 00:00:00 (4/4): python-kitchen-1.1.1-5.el7.noarch.rpm | 267 kB 00:00:00 ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Total 1.4 MB/s | 862 kB 00:00:00 Running transaction check Running transaction test Transaction test succeeded Running transaction Installing : python-chardet-2.2.1-3.el7.noarch 1/4 Installing : python-kitchen-1.1.1-5.el7.noarch 2/4 Installing : libxml2-python-2.9.1-6.el7_2.3.x86_64 3/4 Installing : yum-utils-1.1.31-52.el7.noarch 4/4 Verifying : python-kitchen-1.1.1-5.el7.noarch 1/4 Verifying : yum-utils-1.1.31-52.el7.noarch 2/4 Verifying : libxml2-python-2.9.1-6.el7_2.3.x86_64 3/4 Verifying : python-chardet-2.2.1-3.el7.noarch 4/4 Installed: yum-utils.noarch 0:1.1.31-52.el7 Dependency Installed: libxml2-python.x86_64 0:2.9.1-6.el7_2.3 python-chardet.noarch 0:2.2.1-3.el7 python-kitchen.noarch 0:1.1.1-5.el7 Complete! [root@localhost ~]# yum-config-manager --add-repo http://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo Loaded plugins: fastestmirror adding repo from: http://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo grabbing file http://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo to /etc/yum.repos.d/cuda-rhel7.repo repo saved to /etc/yum.repos.d/cuda-rhel7.repo [root@localhost ~]# yum install -y epel-release Loaded plugins: fastestmirror Loading mirror speeds from cached hostfile * base: mirrors.daticum.com * extras: mirrors.daticum.com * updates: mirrors.daticum.com Resolving Dependencies --> Running transaction check ---> Package epel-release.noarch 0:7-11 will be installed --> Finished Dependency Resolution Dependencies Resolved ============================================================================================================================================================================ Package Arch Version Repository Size ============================================================================================================================================================================ Installing: epel-release noarch 7-11 extras 15 k Transaction Summary ============================================================================================================================================================================ Install 1 Package Total download size: 15 k Installed size: 24 k Downloading packages: epel-release-7-11.noarch.rpm | 15 kB 00:00:00 Running transaction check Running transaction test Transaction test succeeded Running transaction Installing : epel-release-7-11.noarch 1/1 Verifying : epel-release-7-11.noarch 1/1 Installed: epel-release.noarch 0:7-11 Complete! [root@localhost ~]# yum clean all Loaded plugins: fastestmirror Cleaning repos: base cuda epel extras updates Cleaning up list of fastest mirrors
STEP 2) Install the CUDA and the NVIDIA video driver.
As you can see the driver uses DKMS – https://en.wikipedia.org/wiki/Dynamic_Kernel_Module_Support. In simple words, the DKMS allows installing kernel modules, which source does not reside in the kernel tree. NVIDIA uses proprietary video driver and DKMS to load the kernel module and the proprietary video driver.
yum -y install nvidia-driver-latest-dkms cuda
The above line will initiate the installation of around 209 packages including GNU GCC, mesa, perl, java (openjdk), Xorg server and multiple libraries and more. The total downloaded size is 2.4 Gbytes and installed size is 4.7 Gbytes. In the end, there is a section with the full installation log.
STEP 3) Enable the NVIDIA video driver.
At this stage, we have NVIDIA Cuda installed and ready for use, but in most cases, you probably would like to use the compiled code on the same computer, so you need the NVIDIA video driver loaded. The driver is just installed but cannot be used. We have to manually compile it once and after the first compile when updating the kernel it will compile the module automatically.
First install the last dependency:
[root@localhost ~]# yum install kernel-devel Loaded plugins: fastestmirror, nvidia #### NVIDIA #### Loading mirror speeds from cached hostfile * base: mirror.wwfx.net * epel: mirrors.neterra.net * extras: mirror.wwfx.net * updates: mirror.wwfx.net Resolving Dependencies --> Running transaction check ---> Package kernel-devel.x86_64 0:3.10.0-1062.1.1.el7 will be installed --> Finished Dependency Resolution Dependencies Resolved ============================================================================================================================================================================ Package Arch Version Repository Size ============================================================================================================================================================================ Installing: kernel-devel x86_64 3.10.0-1062.1.1.el7 updates 18 M Transaction Summary ============================================================================================================================================================================ Install 1 Package Total download size: 18 M Installed size: 38 M Is this ok [y/d/N]: y Downloading packages: Delta RPMs disabled because /usr/bin/applydeltarpm not installed. kernel-devel-3.10.0-1062.1.1.el7.x86_64.rpm | 18 MB 00:00:02 Running transaction check Running transaction test Transaction test succeeded Running transaction Installing : kernel-devel-3.10.0-1062.1.1.el7.x86_64 1/1 Verifying : kernel-devel-3.10.0-1062.1.1.el7.x86_64 1/1 Installed: kernel-devel.x86_64 0:3.10.0-1062.1.1.el7 Complete!
Build the DKMS module:
[root@localhost ~]# dkms status nvidia, 418.87.00: added [root@localhost ~]# dkms build nvidia/418.87.00 Kernel preparation unnecessary for this kernel. Skipping... Building module: cleaning build area... 'make' -j2 module SYSSRC=/lib/modules/3.10.0-1062.1.1.el7.x86_64/build IGNORE_XEN_PRESENCE=1 IGNORE_PREEMPT_RT_PRESENCE=1 IGNORE_CC_MISMATCH=1..................... cleaning build area... DKMS: build completed. [root@localhost ~]# dkms status nvidia, 418.87.00, 3.10.0-1062.1.1.el7.x86_64, x86_64: built
Enable the driver by uncommenting the line:
options nvidia-drm modeset=1
in file – “/usr/lib/modprobe.d/nvidia.conf”. If the line is missing just add it at the end of the file. In the same file the there is a line “blacklist nouveau” to instruct the kernel not to load the open-source module for the NVIDIA video driver.
And then reboot the machine. After reboot, the NVIDIA driver will be loaded.
Check if the driver is loaded successfully:
[root@localhost ~]# dmesg|grep -i nvidia
[ 6.383444] input: HDA NVidia HDMI/DP,pcm=3 as /devices/pci0000:00/0000:00:03.1/0000:0a:00.1/sound/card0/input12
[ 6.383579] input: HDA NVidia HDMI/DP,pcm=7 as /devices/pci0000:00/0000:00:03.1/0000:0a:00.1/sound/card0/input13
[ 6.383826] input: HDA NVidia HDMI/DP,pcm=8 as /devices/pci0000:00/0000:00:03.1/0000:0a:00.1/sound/card0/input14
[ 6.384343] input: HDA NVidia HDMI/DP,pcm=9 as /devices/pci0000:00/0000:00:03.1/0000:0a:00.1/sound/card0/input15
[ 6.777798] nvidia: loading out-of-tree module taints kernel.
[ 6.777804] nvidia: module license 'NVIDIA' taints kernel.
[ 6.832333] nvidia: module verification failed: signature and/or required key missing - tainting kernel
[ 6.890453] nvidia-nvlink: Nvlink Core is being initialized, major device number 240
[ 6.994744] NVRM: loading NVIDIA UNIX x86_64 Kernel Module 418.87.00 Thu Aug 8 15:35:46 CDT 2019
[ 7.044154] nvidia-uvm: Loaded the UVM driver in 8 mode, major device number 238
[ 7.102688] nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms 418.87.00 Thu Aug 8 15:27:13 CDT 2019
[ 7.111439] [drm] [nvidia-drm] [GPU ID 0x00000a00] Loading driver
[ 7.111488] nvidia 0000:0a:00.0: irq 340 for MSI/MSI-X
[ 7.661829] caller os_map_kernel_space.part.5+0x85/0xb0 [nvidia] mapping multiple BARs
[ 8.353066] [drm] Initialized nvidia-drm 0.0.0 20160202 for 0000:0a:00.0 on minor 0
[root@localhost ~]# ls -altr /dev/nvidia*
crw-rw-rw-. 1 root root 195, 255 24 Sep 1,56 /dev/nvidiactl
crw-rw-rw-. 1 root root 238, 0 24 Sep 1,56 /dev/nvidia-uvm
crw-rw-rw-. 1 root root 238, 1 24 Sep 1,56 /dev/nvidia-uvm-tools
crw-rw-rw-. 1 root root 195, 254 24 Sep 1,56 /dev/nvidia-modeset
crw-rw-rw-. 1 root root 195, 0 24 Sep 1,56 /dev/nvidia0
[root@localhost ~]# ls -altr /dev/dri/
total 0
drwxr-xr-x. 2 root root 80 24 Sep 1,56 .
crw-rw----. 1 root video 226, 128 24 Sep 1,56 renderD128
crw-rw----. 1 root video 226, 0 24 Sep 1,56 card0
drwxr-xr-x. 21 root root 3580 24 Sep 1,56 ..
[root@localhost ~]# nvidia-smi
Tue Sep 24 02:10:52 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.87.00 Driver Version: 418.87.00 CUDA Version: 10.1 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 108... Off | 00000000:0A:00.0 On | N/A |
| 0% 46C P8 13W / 275W | 2MiB / 11174MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
If the driver is not loaded all of the above directories in /dev would be missing and the nvidia smi would report the driver is not loaded.