NVIDIA Optimus: A tale of two graphic cards
This is a short summary on how I managed to use the nvidia graphic card with my Debian Testing installation. I’ve cooked this blog since it may help other (and a future me) during their debian configuration stage
L’Entrée
Most Linux users are familiar with the hassle of making Nvidia GPU’s working with their linux distributions. I’ve bought a Laptop that happens to run a new – at least for me – technology known as NVIDIA Optimus which allows to have an internal Intel Graphics Card running with dedicated Nvidia GPU, each independent of the other, so we can selectively render graphics at each card according to our performance needs, thus optimizing the use of resources mostly for Laptops like mine.
You tought having only a nvidia graphics card was troublesome? No wonder that making Optimus work is not an exception.
Main course
First of all, you need to check that your system is optimus compatible by checking which graphic cards you have
$ lspci | grep -i VGA
# Should output something like
00:02.0 VGA compatible controller: Intel Corporation UHD Graphics 630 (Mobile)
And then
$ lspci | grep -i 3D
# Should output something like
01:00.0 3D controller: NVIDIA Corporation GP107M [GeForce GTX 1050 Ti Mobile] (rev a1)
So, you have two graphic units. In my case, an Intel Graphics Card, and the Nvidia GPU.
Spoiler Alert!: Installing just the nvidia drivers might work, but is not the essence of having an optimus system. As a fallback alternative I’ve been running manually installed nvidia drivers for my laptop and is not the most energy saving solution
In order to check the GLX implementation we need to install mesa-utils since they include the glxinfo command,
$ glxinfo | grep OpenGL
# Should output something like
OpenGL vendor string: Intel Open Source Technology Center
OpenGL renderer string: Mesa DRI Intel(R) UHD Graphics 630 (Coffeelake 3x8 GT2)
OpenGL core profile version string: 4.5 (Core Profile) Mesa 19.2.3
OpenGL core profile shading language version string: 4.50
OpenGL core profile context flags: (none)
OpenGL core profile profile mask: core profile
OpenGL core profile extensions:
OpenGL version string: 3.0 Mesa 19.2.3
OpenGL shading language version string: 1.30
OpenGL context flags: (none)
OpenGL extensions:
OpenGL ES profile version string: OpenGL ES 3.2 Mesa 19.2.3
OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.20
OpenGL ES profile extensions:
Now, lets install bumblebee with the nvidia drivers. Bumblebee aims to provide support for NVIDIA Optimus laptops for GNU/Linux distributions.
sudo apt install nvidia-driver bumblebee-nvidia primus libgl1-nvidia-glx nvidia-smi
Now, we can check that optirun passes the execution to the nvidia card
optirun glxinfo | grep OpenGL
I won’t tell you that things worked straightforward, the first time I ran this I got:
$ optirun glxinfo | grep OpenGL
[23794.793713] [ERROR]Cannot access secondary GPU - error: Could not load GPU driver
[23794.793750] [ERROR]Aborting because fallback start is disabled.
Here’s where the adventure begins, and I feel really thankful with the people of the Arch Linux community (see the resources below)
Dessert
Here comes the tricky part. I’ll be adding almost ad literam what’ve found on the arch linux forum. There is not really much to elaborate here.
In order to verify the performnace, we need to install two additional things
# apt install tlp powertop
And perform the following configurations
/etc/default/tlp
Add GPU to TLP with variable RUNTIME_PM_BLACKLIST at /etc/default/tlp. To that end we need the slot for the nvidia card (the 01:00.0 when one does lspci over the nvidia)
RUNTIME_PM_BLACKLIST="01:00.0"
/etc/bumblebee/bumblebee.conf
Driver=nvidia
And in nvidia section:
PMMethod=none
/etc/tempfiles.d/nvidia_pm.conf
Allow gpu to poweroff on boot
w /sys/bus/pci/devices/0000:01:00.0/power/control - - - - auto
/etc/X11/xorg.conf.d/01-noautogpu.conf
Section "ServerFlags"
Option "AutoAddGPU" "off"
EndSection
/etc/X11/xorg.conf.d/20-intel.conf
Section "Device"
Identifier "Intel Graphics"
Driver "modesetting"
EndSection
Create blacklist files
/etc/modprobe.d/blacklist.conf
blacklist nouveau
blacklist rivafb
blacklist nvidiafb
blacklist rivatv
blacklist nv
blacklist nvidia
blacklist nvidia-drm
blacklist nvidia-modeset
blacklist nvidia-uvm
blacklist ipmi_msghandler
blacklist ipmi_devintf
/etc/modprobe.d/disable-ipmi.conf
These modules are loaded together with nvidia and block its unloading. I do not need ipmi therefore I simply disabled this functionality.
install ipmi_msghandler /usr/bin/false
install ipmi_devintf /usr/bin/false
/etc/modprobe.d/disable-nvidia.conf
install nvidia /bin/false
Create GPU management scripts
GPU management scripts were created by tyrells to which manipulation of blacklist config was added.
Create two following management scripts. Creation of aliases is recommended.
enableGpu.sh
#!/bin/sh
# allow to load nvidia module
mv /etc/modprobe.d/disable-nvidia.conf /etc/modprobe.d/disable-nvidia.conf.disable
# remove NVIDIA card (currently in power/control = auto)
echo -n 1 > /sys/bus/pci/devices/0000\:01\:00.0/remove
sleep 1
# change PCIe power control
echo -n on > /sys/bus/pci/devices/0000\:00\:01.0/power/control
sleep 1
# rescan for NVIDIA card (defaults to power/control = on)
echo -n 1 > /sys/bus/pci/rescan
disableGpu.sh
modprobe -r nvidia_drm
modprobe -r nvidia_uvm
modprobe -r nvidia_modeset
modprobe -r nvidia
# change NVIDIA card power control
echo -n auto > /sys/bus/pci/devices/0000\:01\:00.0/power/control
sleep 1
# change PCIe power control
echo -n auto > /sys/bus/pci/devices/0000\:00\:01.0/power/control
sleep 1
# lock system form loading nvidia module
mv /etc/modprobe.d/disable-nvidia.conf.disable /etc/modprobe.d/disable-nvidia.conf
Create service which locks GPU on shutdown
Service which locks GPU on shutdown / restart when it is not disabled by disableGpu.sh script is necessary. Otherwise on next boot nvidia will be loaded together with ipmi modules (even if we have blacklist with install command for them) and it would not be possible to unload them.
/etc/systemd/system/disable-nvidia-on-shutdown.service
[Unit]
Description=Disables Nvidia GPU on OS shutdown
[Service]
Type=oneshot
RemainAfterExit=true
ExecStart=/bin/true
ExecStop=/bin/bash -c "mv /etc/modprobe.d/lock-nvidia.conf.disable /etc/modprobe.d/lock-nvidia.conf || true"
[Install]
WantedBy=multi-user.target
Enabling
Reload systemd daemons and enable service:
systemctl daemon-reload
systemctl enable disable-nvidia-on-shutdown.service
Now, after rebooting the system I ran
# chmod 755 enableGpu.sh disableGpu.sh
and then, tested running
$ nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.
Now, the moment of truth, lets see how it goes afet running doing,
# ./enableGpu.sh
$ nvidia-smi
Tue Nov 12 17:48:28 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 430.50 Driver Version: 430.50 CUDA Version: N/A |
|-------------------------------+----------------------+----------------------+
| 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 105... Off | 00000000:01:00.0 Off | N/A |
| N/A 56C P0 N/A / N/A | 0MiB / 4040MiB | 1% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
Voilá! Now the GPU can be enabled or disable as needed