Intel integrated GPU + NVidia dedicated GPU.
- 1. prime-select nvidia
- 2. prime-select intel
  - 2.1 monitor connected to intel GPU, HMD connected to intel GPU
  - 3. Nvidia driver may fail due to damaged HMD display cable.
AMD integrated GPU + AMD dedicated GPU, wayland session
- 1. Wayland FOSS drivers, monitor connected to iGPU, HMD connected to dGPU

Intel integrated GPU + NVidia dedicated GPU.

The following configurations have been tested on Ubuntu 20.04 with UHD Graphics 610 using Mesa 20.0.8 and a GeForce GTX 1050 Ti using nvidia 440.64.

sudo prime-select nvidia and sudo prime-select intel can switch between using the nvidia or the intel GPU as primary GPU (the GPU the X Server primarily runs on). Reboot required after switch.

1. prime-select nvidia

1.1 monitor connected to intel gpu, HMD connected to nvidia gpu

Note: nvidia may choose an nvidia-only setup when there are monitors connected to nvidia. When connecting monitors to the intel GPU, connect ONLY a HMD to the nvidia gpu.

monado-service on nvidia gpu: Direct mode works
- Vulkan client on nvidia GPU: works
- Vulkan client on intel GPU: vkAllocateMemory: VK_ERROR_INVALID_EXTERNAL_HANDLE
- OpenGL client on nvidia GPU: works
- OpenGL client on intel GPU: n/a (unknown how to start an application like this)
monado-service on intel gpu: no direct mode, compositor window renders only black/locks up?
- Vulkan client: locks up, likely blocked on blocked compositor
- XRT_COMPOSITOR_FORCE_NVIDIA=1 will not work, intel driver will not detect HMD on nvidia GPU

1.2 monitor connected to nvidia GPU, HMD connected to intel GPU

Same as 1.1.

1.3 monitor connected to intel GPU, HMD connected to intel GPU

monado-service on intel GPU: vkAcquireXlibDisplayEXT: VK_ERROR_INITIALIZATION_FAILED
- probable explanation: intel driver can see HMD display output, but running with nvidia as primary GPU, direct mode is not properly supported in this reverse prime configuration
monado-service on nvidia GPU: X Error
- probable explanation: nvidia driver sees HMD display connected to intel GPU, but fails to use it with direct mode

X Error of failed request:  BadMatch (invalid parameter attributes)
  Major opcode of failed request:  156 (NV-GLX)
  Minor opcode of failed request:  31 ()
  Serial number of failed request:  31
  Current serial number in output stream:  31

Note for 1.1 and 1.2

When running in prime-select nvidia mode, other Vulkan applications like vkmark also lock up with a black window when attempting to run on the intel GPU. (see https://github.com/vkmark/vkmark/pull/28 for GPU selection in vkmark)

2. prime-select intel

Completely disables the nvidia GPU on my desktop PC. Please contribute to this article if you have an Optimus Laptop that behaves differently.

2.1 monitor connected to intel GPU, HMD connected to intel GPU

monado-service on intel gpu: Direct mode works
- Vulkan client on intel GPU: works
- OpenGL client on nvidia GPU: works

3. Nvidia driver may fail due to damaged HMD display cable.

Should your system be affected by visual snow or a failing tether cable nvidia may throw a Mode is rejected: DP Bandwidth check failed. in X logs indicating a refusal to expose the display.

AMD integrated GPU + AMD dedicated GPU, wayland session

The following configuraion was tested on mesa 23.0.0, KDE plasma 5.27.2, wayland, kernel 6.1, 5900HX & RX 6800M.

For KDE:

KWIN_DRM_DEVICES=/dev/dri/card0:/dev/dri/card1 KWIN_DRM_DEVICES=/dev/dri/card1:/dev/dri/card0

For Gnome:

WLR_DRM_DEVICES=/dev/dri/card0:/dev/dri/card1 WLR_DRM_DEVICES=/dev/dri/card1:/dev/dri/card0

The above enviornment variables may be used to select a card in which the desktop session will be run on under wayland when placed in /etc/enviornment.

Depending on your system inserting either but not both of the following configurations (as examples) will select the primary GPU of your desktop session. This should not be necessary but can come in handy if you ever need to debug multiGPU issues.

Please note that all apps in the monado session must be run on the same GPU that monado runtime is operating on. For performance reasons nearly all applications will utilize vkFormat OPTIMAL tiling which can introduce artifacting when accessing cross-GPU resources.

The solution should simply be to either attempt to provide vkFormat LINEAR between GPUs in your own software implementation or to ensure all applications are operating on the same GPU with the DRI_PRIME enviornment variable.

1. Wayland FOSS drivers, monitor connected to iGPU, HMD connected to dGPU

Wayland provides fairly frictionless access to displays located across different GPUs. In the tested configuration an LCD was attached to the iGPU and the HMD attached to the dGPU over USB-C Alt-DP mode and were able to function seamlessly and acquire direct mode to the HMD.

Monado