Multi-GPU

General

If your host machine uses multiple GPUs, you may want to use one GPU for rendering all the elements for Hyprland including windows, animations, and another for hardware acceleration for certain applications, etc.

This setup is very common in the likes of gaming laptops, GPU-passthrough (without VFIO) capable hosts, and if you have multiple GPUs in general.

Detecting GPUs

For this case, the writer is taking the example of their laptop.

Running lspci -d ::03xx lists all the PCI display controllers available:

01:00.0 VGA compatible controller: NVIDIA Corporation TU117M [GeForce GTX 1650 Mobile / Max-Q] (rev a1)
06:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Cezanne [Radeon Vega Series / Radeon Vega Mobile Series] (rev c6)

Here it is clear that 2 GPUs are available, the dedicated NVIDIA GTX 1650 Mobile and the integrated AMD Cezanne Radeon Vega Series GPU.

Now, run ls -l /dev/dri/by-path.

 total 0
lrwxrwxrwx 1 root root  8 Jul 14 15:45 pci-0000:01:00.0-card -> ../card0
lrwxrwxrwx 1 root root 13 Jul 14 15:45 pci-0000:01:00.0-render -> ../renderD128
lrwxrwxrwx 1 root root  8 Jul 14 15:45 pci-0000:06:00.0-card -> ../card1
lrwxrwxrwx 1 root root 13 Jul 14 15:45 pci-0000:06:00.0-render -> ../renderD129

From the above outputs, we can see that the path for the AMD card is pci-0000:06:00.0-card, due to the matching 06:00.0 from the first command. Do not use the card1 symlink indicated here. It is dynamically assigned at boot and is subject to frequent change, making it unsuitable as a marker for GPU selection.

Telling Hyprland which GPU to use

After determining which “card” belongs to which GPU, we can now tell Hyprland which GPUs to use by setting the AQ_DRM_DEVICES environment variable.

If you would like to use another GPU, or the wrong GPU is picked by default, set AQ_DRM_DEVICES to a :-separated list of card paths, for example:

hl.env("AQ_DRM_DEVICES", "/dev/dri/card0:/dev/dri/card1")

Here, we tell Hyprland which GPUs it’s allowed to use, in order of priority. For example, card0 will be the primary renderer, but if it isn’t available for whatever reason, then card1 is primary.

Do note that if you have an external monitor connected to, for example card1, that card must be included in AQ_DRM_DEVICES for the monitor to work, though it doesn’t have to be the primary renderer.

You should now be able to use an integrated GPU for lighter GPU loads, including Hyprland, or default to your dGPU if you prefer.

Creating consistent device paths for specific cards

As mentioned above, it’s not recommended to use the /dev/dri/card* device paths since they periodically change which device they are symlinked to. Furthermore, the colons in the actual card device paths are not usable in the AQ_DRM_DEVICES environment variable since colons : are used as a separator for multiple paths.

It’s possible to use udev rules to create reliable symlinks to particular device cards. For example, to create a symlink to an AMD card at the path /dev/dri/amd-igpu, we can create a udev rule at /etc/udev/rules.d/amd-igpu-dev-path.rules programmatically like so:

SYMLINK_NAME="amd-igpu"
RULE_PATH="/etc/udev/rules.d/amd-igpu-dev-path.rules"
AMD_IGPU_ID=$(lspci -d ::03xx | grep 'AMD' | cut -f1 -d' ')

cat <<EOF | sudo tee "$RULE_PATH"
KERNEL=="card*", \
KERNELS=="0000:$AMD_IGPU_ID", \
SUBSYSTEM=="drm", \
SUBSYSTEMS=="pci", \
SYMLINK+="dri/$SYMLINK_NAME"
EOF

Then, reload the udev rules with:

sudo udevadm control --reload
sudo udevadm trigger

There should now be a symlink at /dev/dri/amd-igpu that points to your respective card file:

$ ls -l /dev/dri/amd-igpu
lrwxrwxrwx 1 root root 5 /dev/dri/amd-igpu -> card1

This symlink will automatically update to point to correct card file if it ever changes.

Now it is possible to use the new symlink in the AQ_DRM_DEVICES environment variable:

hl.env("AQ_DRM_DEVICES", "/dev/dri/amd-igpu")