Arch Linux

nashod

Member

Registered: 2012-03-23

Posts: 40

Need help understanding the landscape of dual GPUs

Hello all,  a fresh returner to arch after a long absence. The return was not planned, hence the Nvidia GPU on a system that will run Hyprland.

So the state of things is as follows:

CPU: Ryzen 5 7600X which includes an iGPU
GPU: Nvidia Geforce RTX 2060

From what I recall from a decade ago, on laptops I used to offload the iGPU to the GPU using sinks or something along those lines (or maybe it was the other way around).

Looking at the wiki and doing some web searching I gather that that is still done using something called Prime, am I in the right direction? for Hyprland itself, I'll need to export an enviroment variable to the renderer called aquamarine and tell it to use the iGPU.

What I'm looking to achieve is essentially having the iGPU handle the desktop and everyday tasks and have the GPU handle gaming possibly (I might do a windows VM with GPU passthrough later on), maybe some light CAD for 3D printing. Physically what do I connect to the monitor? I'm assuming the iGPU? Drivers wise, I'm guessing mesa, vulkan and the proprietery Nvidia driver?

I'm sorry for being all over the place, but that's my understanding of GPUs in general, it feels like I have all the pieces to the puzzle, I just need someone to help me put them together.

Head_on_a_Stick

Member

From: The Wirral

Registered: 2014-02-20

Posts: 8,824

Website

Re: Need help understanding the landscape of dual GPUs

If you had bought an AMD discrete card then switching for graphically intensive applications would have been automagical using the stock open source drivers for best performance.

Unfortunately you bought NVIDIA and they hate open source. Good luck!

---

Jin, Jîyan, Azadî

V1del

Forum Moderator

Registered: 2012-10-16

Posts: 24,115

Re: Need help understanding the landscape of dual GPUs

And on a more useful note, nvidia implemented PRIME and you can use that yes. But I'd generally reccommend you just stick to one GPU. The sync overhead having to copy data between graphics cards is likely not worth it if you can avoid it -- and I'd hazard a guess that the integrated GPU of that new of a processor is likely going to be close/on par in performance to that old of an nvidia gpu. But generally the setup process with prime is simply installing the nvidia-prime package and running applications that should land on the dGPU with prime-run. For gaming proton and the used abstraction layers have code that will automatically pick a dedicated card over integrated, native ones you still need to provide the prime-run parameter: https://wiki.archlinux.org/title/PRIME# … er_offload