This guide will show you how to use Intel graphics for rendering display and NVIDIA graphics for CUDA computing on Ubuntu 18.04 / 20.04 desktop.
I made this work on an ordinary gaming PC with two graphics devices, an Intel UHD Graphics 630 plus an NVIDIA GeForce GTX 1080 Ti.
Both of them can be shown via lspci | grep VGA.
00:02.0 VGA compatible controller: Intel Corporation Device 3e92
01:00.0 VGA compatible controller: NVIDIA Corporation GP102 [GeForce GTX 1080 Ti] (rev a1)
This guide can be summarized into two steps:
- To install NVIDIA drivers without OpenGL files.
- To configure Xorg to use the Intel graphics.
I haven't tested on different hardware configurations, but it should work similarly. See this section for more discussion.
Before operating within Linux, you need to make some configuration on your hardware. Make sure monitors are plugged to the motherboard the instead of dedicated display card. Configure the BIOS to make Intel graphics as the primary display device (usually select IGFX instead of PEG or PCIE). Make sure your computer could boot to GUI and be logged in to desktop successfully under this setting.
I suggest installing the driver in either of the following ways. If you would like to follow your own method, just make sure the OpenGL files are not installed.
Common steps for both methods to avoid possible conflicts.
-
If you have installed via PPA repositories
sudo apt purge nvidia* # some interactive operations sudo apt autoremove # some interactive operations
Check remaining packages related to NVIDIA.
dpkg -l | grep nvidiaIf some packages are not removed, manually remove them by
sudo dpkg -P <package_name>
If you have add third party repositories, e.g. ones from NVIDIA, remove them too. This could be done by removing related files under
/etc/apt/source.list.d/or viappa-purgeutility. -
If you have installed via binary installers
sudo nvidia-uninstall # some interactive operations -
Reboot.
-
Add the
ppa:graphics-drivers/pparepository.sudo add-apt-repository ppa:graphics-drivers/ppa # some interactive operationsOn ubuntu 18.04
sudo apt updateis automatically performed after adding a PPA, thus not manually required. -
Install the headless NVIDIA driver
sudo apt install nvidia-headless-418 nvidia-utils-418
Version 418 is the latest when I write this page. Changing to the latest version available is a good idea.
IMPORTANT
The
nvidia-headless-418contains only the driver, while the fullnvidia-driver-418package contain everything including display component such OpenGL libraries. If you hope to connect the display to a NVIDIA display card, install the full package, otherwise, install only the driver.The
nvidia-utils-418package provide utilities such asnvidia-smi. -
Reboot. If the installation is successful, command
nvidia-smiwill show all NVIDIA GPUs.
My previous post, thought old, is detailed and still work.
To summary:
-
Download the binary installer from NVIDIA official site and make it executable.
-
Disable nouveau driver by creating file
/etc/modprobe.d/blacklist-nouveau.confwith contentblacklist nouveau options nouveau modeset=0or just executing the installer and it will prompt to create it for you.
-
Execute
sudo update-initramfs -uand then reboot. -
After booting, switch to tty1. Stop display services such as
gdm,gdm3,lightdm. Kill all process that is using the graphic card, such asXorgorX. -
Execute the installer with
--no-opengl-filessuffix (IMPORTANT) to avoid installation of all OpenGL related files, likesudo ./NVIDIA-Linux-x86_64-418.56.run --no-opengl-files
Or if you would like to display from an NVIDIA graphic card, execute the installer without any arguments, like
sudo ./NVIDIA-Linux-x86_64-418.56.run
-
After a successful installation, command
nvidia-smiwill show all NVIDIA GPUs.
The installed NVIDIA driver and configurations will hint Xorg to start with NVIDIA devices. Depending on whether NVIDIA related display libraries are well installed, the X server would failed to start or success to start but still use NVIDIA devices, both of which are unwanted.
We can force Xorg to use Intel graphics by creating a configuration file with the following contents and save it to /etc/X11/xorg.conf.
Section "ServerLayout"
Identifier "Layout0"
Screen 0 "Screen0"
EndSection
Section "Screen"
Identifier "Screen0"
Device "Device0"
EndSection
Section "Device"
Identifier "Device0"
Driver "intel"
VendorName "Intel Corporation"
BusID "PCI:0:2:0
EndSection
The key point is the "BusID" option. It indicates the PCI bus id that the Intel graphics connects to.
It can be retrieved from lspci.
For example, on my computer, lspci outputs 00:02.0 VGA compatible controller: Intel Corporation Device 3e92, thus the bus id is 0:2:0.
Note that the bus id output from lspci is hexadecimal but the number filled in xorg.conf should be decimal.
For example, if the output from lspci is 82:00.0 VGA com..., you need to fill PCI:130:0:0 in the configuration.
On Ubuntu 20.04 you may want to set the driver to modesetting instead of intel. I met some problem and solved as this link (and links within it) describes.
Setting with multiple monitors would be a little bit complex.
Here is my example of setting with two monitors. Some fields are missing but it works as Xorg will smartly use some default configs.
Anyway, search the Internet to get a proper set of configurations for you. man xorg.conf and ArchLinux wiki are good references.
Section "Device"
Identifier "Device0"
Driver "modesetting"
VendorName "Intel Corporation"
BusID "PCI:0:2:0
Option "TearFree" "true"
Option "monitor-DP-1" "DP"
Option "monitor-HDMI-2" "HDMI"
EndSection
Section "Monitor"
Identifier "DP"
Option "Rotate" "Left"
EndSection
Section "Monitor"
Identifier "HDMI"
Option "RightOf " "DP"
Option "Primary" "true"
EndSection
After the configuration, reboot the computer.
If it successes, you will be able to login to the desktop.
If it fails, you could be locked at the login screen.
Reboot to advance mode, drop to root prompt and check /var/log/Xorg.*.log.* for hints.
After login successfully, open an terminal and execute glxheads.
The displayed rendering devices should be the Intel graphics.
glxheads: exercise multiple GLX connections (any key = exit)
Usage:
glxheads xdisplayname ...
Example:
glxheads :0 mars:0 venus:1
Name: :0
Display: 0x56151a1570e0
Window: 0x2000002
Context: 0x56151a182f60
GL_VERSION: 3.0 Mesa 18.0.5
GL_VENDOR: Intel Open Source Technology Center
GL_RENDERER: Mesa DRI Intel(R) HD Graphics (Coffeelake 3x8 GT2)
Check nvidia-smi, there should be no Xorg process listed and memory occupation should be 0MB.
Thu Nov 22 07:05:55 2018
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.56 Driver Version: 418.56 CUDA Version: 10.0 |
|-------------------------------+----------------------+----------------------+
| 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 | 00000000:01:00.0 Off | N/A |
| 0% 52C P0 57W / 250W | 0MiB / 11178MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
Execute a heavy GPU program and move your mouse during the computation. The display should be smooth, as the computing and display rendering are separated on different devices.
Congratulations!
If glxheads outputs some virtual graphics and nvidia-smi still output Xorg processes, you probably still using the NVIDIA graphics for rendering. It just pass the rendered image to Intel graphics for display. When running heavy GPU programs, the display will still be slow.
Separating display and computing is important for a smooth work.
-
You could use a remote headless server for computing a local client for display. You can connect the remote server via SSH. This could be the simplest way. Under this setting, a properly installed headless driver will make everything work. You don't need to follow this guide.
-
If you have only one computer and there is only one graphic device in your computer. Its unfortunate. You have to use the only device for all tasks and you would suffer severe display latency when running GPU programs.
-
You user a single computer for both computing and display. There are two (or more) graphic devices and they are from the same vendor. Then the problem could be mush easier. Taking NVIDIA as the example, with a properly installed driver, display and computing can be perfectly separated. One can plug monitors to device 0 and use
CUDA_VISIBIE_DEVICE=1environment flag to perform computing on device 1. You probably don't need this guide. -
You user a single computer for both computing and display. There are two (or more) graphic devices but they are from different vendors. For example, an ordinary gaming PC configuration could include a Intel HD Graphics within the CPU and a dedicated GPU from NVIDIA. One need to plug monitors to the motherboard to use the Intel one for display and run CUDA program on the NVIDIA one. Then you are at the right place.
-
For some purpose, you need an NVIDIA GPU for both computing and rendering. Then you probably need to run both of them on a single GPU. You need to install the drvier with OpenGL support and tune some rendering settings in Xorg.
Thank you! This fixed my problem exactly.