mGPU (discrete AMD GPU + discrete Nvidia GPU)

Wondering how to set this up on Garuda

I have a RX 6900XT which I am using as my default GPU, I would like to add a RTX 2080ti to PC to use as a CUDA and NVEC accelerator. So basically I want the Nvidia card to be always available for use with specific applications while using the AMD GPU. So for example I could be gaming on the AMD GPU and use the Nvidia GPU for NVEC encoding in OBS to capture what the AMD card is rendering. Or I use the AMD GPU as primary GPU for displaying my work as well as doing OpenCL compute work while at the same time running CUDA compute workloads in the background on the Nvidia GPU.

When I was on Elementary OS/ Ubuntu 18.04 I could do this easily by just installing the nvidia drivers, install the GPU and the primary GPU would be selected based on the order of PCIe slots no switching required both GPUs will just work as expected.

With Arch/Garuda or KDE not sure which, I am being forced to have the hybrid switching thing. I thought the hybrid switching was only for APU/iGPU and discrete GPU combination not setups with two discrete GPUs. So I can't seem to get them set up as I want because Nvidia Optimus keeps being installed and taking control of the system. Under Elementary/Ubuntu Nvidia Optimus was never installed with the core drivers.

Am I missing something about how to use the hybrid switching or does mGPU just not work?

What is forcing this?

1 Like

Nvidia Optimus I believe

Do you mean that Nvidia has hacked your system, or Nvidia Optimus is a Garuda utility?

You can change video drivers using Garuda Hardware Manager. Does it fail?

4 Likes

Basically, when the Nvidia drivers are installed Optimus gets installed and GPU switching is the default. So while using the AMD card the Nvidia card is not found by any application in the system. I can only use AMD or Nvidia not both.

When I initially tried last year, I installed the Nvidia version of Garuda and was forced into GPU switching.

I have been trying again over the past few days using the non-Nvidia Garuda build from last year. The Hardware configuration installer (both with auto install proprietary drivers or manually selecting to install dkms) fails to install properly and doesn't find the Nvidia and other modules, though Nvidia settings installs and hardware manager says the drivers are installed. The AUR install via pacman sudo pacman -S nvidia seems to install without errors. However, just like with Optimus, nothing detects the Nvidia GPU with the AMD card as primary.

There's one more thing I am going to try to double check my system is up to date though I know it is seeing I just updated it this week and again yesterday.

[EDIT: Only under the LTS kernel does the OS seem to find the Nvidia GPU under Nvidia settings, the GPU isn't found using Zen, tgk, XanMod kernels]

I believe I found a solution to this now.

The hardware configuration tool seems not to be installing the dkms drivers so I had to install from terminal using

sudo pacman -S nvidia-dkms

Before it installed everything properly except for the tkg kernel which I don't use and can't remember why I installed.

I now just need to confirm that the CUDA toolkit installs

This topic was automatically closed 2 days after the last reply. New replies are no longer allowed.