These utility scripts aim to make the life easier for nvidia cards users. It started with a revelation that bumblebee in current state offers very poor performance. This solution offers a bit more complicated procedure but offers a full GPU utilization(in terms of linux drivers)
This is a fork of the repository Witko to make his script work in Debian (and Debian only!!!) alongside Bumblebee.
Alongside doesn't mean at the same time: you can't run an nvidia-xrun session while using optirun/primusrun and viceversa.
My advice is to just switch to nvidia-xrun session when is required to start a heavy GPU program (like Gaming) and use primusrun for the middle-range GPU demanding programs.
Do not try to use this in other distros whitouth applying the proper modifications. It just won't work.
This is required to launch nvidia-xrun
without typing any password.
If you installed manually you need to add a system group:
# groupadd --system nvxrun
and then add the user that is going to use nvidia-xrun
to that group:
# adduser [YOUR_USERNAME] nvxrun
you will need to reload sudo service:
# systemctl reload sudo
- switch to free tty
- login
- run
nvidia-xrun [app]
- enjoy
The systemd service can be used to completely remove the card from the kernel
device tree (so that it won't even show in lspci
output), and this will
prevent the nvidia module to be loaded, so that we can take advantage of the
kernel PM features to keep the card switched off.
The service can be enabled with this command:
# systemctl enable nvidia-xrun-pm
but if using along side with bumblebee just don't do that: bumblebee bbswitch method will be used instead.
When the nvidia-xrun command is used, the device is added again to the tree so that the nvidia module can be loaded properly: nvidia-xrun will remove the device and enable PM again after the application terminates.
- nvidia-xrun - uses following dir structure:
- /usr/bin/nvidia-xrun - the executable script
- /usr/lib/nvidia-xrun/nv-card-controller - the executable script for controlling PM of the dGPU
- /usr/lib/nvidia-xrun/nv-modules - the executable script for loading and unloading nvidia modules
- /usr/lib/nvidia-xrun/nv-brightness - if having troubles with monitor backlight when in nvidia-xrun session reference this script to change brightness (possible arguments: [up|down|toggle|get])
- /etc/X11/xorg.nvidia-xrun.conf - the main X confing file
- /etc/X11/xinit/nvidia-xinitrc - xinitrc config file. Contains the setting of provider output source
- /etc/X11/xinit/nvidia-xinitrc.d - custom xinitrc scripts directory
- /etc/X11/xorg.nvidia-xrun.conf.d - custom X config directory
- /etc/sudoers.d/nvidia-xrun-sudoers - sudoers file
- /etc/systemd/system/nvidia-xrun-pm.service systemd service
- /etc/default/nvidia-xrun - nvidia-xrun config file
- /usr/share/xsessions/nvidia-xrun-openbox.desktop - xsession file for openbox
- /usr/share/xsessions/nvidia-xrun-plasma.desktop - xsession file for plasma
- /usr/share/xsessions/nvidia-xrun-gnome.desktop - xsession file for GNOME
- [OPTIONAL] $XDG_CONFIG_HOME/X11/nvidia-xinitrc - user-level custom xinit script file. You can put here your favourite window manager for example
Usually the 1:0:0 bus is correct. If this is not your case(you can find out through lspci or bbswitch output mesages) you can create
a conf script for example nano /etc/X11/xorg.nvidia-xrun.conf.d/30-nvidia.conf
to set the proper bus id:
Section "Device"
Identifier "nvidia"
Driver "nvidia"
BusID "PCI:2:0:0"
EndSection
You can use this command to get the bus id:
lspci | grep -i nvidia | awk '{print $1}'
Note that this prints your bus id in hexadecimal, but the Xorg configuration script requires that you provide it in decimal, so you'll need to covert it. You can do this with bash:
# In this example, my bus id is "3c"
bash -c "echo $(( 16#3c ))"
Also this way you can adjust some nvidia settings if you encounter issues:
Section "Screen"
Identifier "nvidia"
Device "nvidia"
# Option "AllowEmptyInitialConfiguration" "Yes"
# Option "UseDisplayDevice" "none"
EndSection
In order to make power management features work properly, you need to make sure
that bus ids in /etc/default/nvidia-xrun
are correctly set for both the
NVIDIA graphic card and the PCI express controller that hosts it. You should be
able to find both the ids in the output of lshw
: the PCIe controller is
usually displayed right before the graphic card.
For convenience you can create nano ~/.config/X11/nvidia-xinitrc
and put there your favourite window manager:
if [ $# -gt 0 ]; then
$*
else
openbox-session
# startkde
fi
With this you do not need to specify the app and you can simply run:
nvidia-xrun
The Debian package (not Ubuntu and derivates) can be found here
The Arch Linux User Repository package can be found here.
The RPM packages and repository details for all supported distributions can be found on the ekultails/nvidia-xrun COPR overview page.
sudo dnf copr enable ekultails/nvidia-xrun
sudo dnf install nvidia-xrun
Yes unfortunately running Steam directly with nvidia-xrun does not work well - I recommend to use some window manager like openbox.
When using openbox on a HiDPI (i.e. 4k) display, everything could be so small that is difficult to read.
To fix, you can change the DPI settings in ~/.Xresources (~/.Xdefaults)
file by adding/changing Xft.dpi
setting. For example :
Xft.dpi: 192
nouveau
driver should be automatically blacklisted by nvidia
but in case it is not, nvidia
might not get access to GPU. Then you need to manually blacklist nouveau
following Arch wiki https://wiki.archlinux.org/index.php/kernel_modules#Blacklisting.
nvidia
driver may load itself on boot, then nvidia-xrun
will fail to start Xorg session.
To avoid that, you should blacklist it (see link above).
Also sometimes, blacklisting is not enough and you should use some hack to really avoid it to load.
For example, adding install nvidia /bin/false
to /etc/modprobe.d/nvidia.conf
will make every load to fail.
In that case, you should add --ignore-install
to modprobe
calls in nvidia-xrun
script.
Check https://wiki.archlinux.org/index.php/Vulkan
- remove package vulkan-intel
- set VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json