VRAM within a qube for first person games, or other graphically "heavy" loads

I just managed to get Minecraft working earlier today, by using an external USB mouse and assigning that mouse to the qube in which Minecraft is installed, this gets around the issue of the track-pad (or dom0 associated mouse) passing through absolute mouse coordinate making first person games impossible to play.

That being said my performance is quite lacking, the qube itself is assigned 10gb of ram, and minecraft has access to 4 of those gigs, the qube had 6 vCPUS but now has 4 as it seemed to have little effect

I believe the issue is in how much vram the qube has, if I make the game full screen fps is about 15-20fps on minimum settings, but if I make the window about 1/4th of my display, I get a good 22-32 fps, I may be mistaken, but I wanted to test it and was curious if there is a way to customize how much vram a given qube has access to?

That won’t work, unfortunately…
You don’t have GPU acceleration in qubes (VMs), as they use CPU for rendering.

The thing you can do if you have a second GPU is to pass it in a specific qube and then use that for gaming.

interesting, for some reason I though the integrated graphics were used within VMs as well, makes sense they aren’t I’m pretty sure this laptop does have a weak but dedicated nvidia gpu, I may look into using that for passthough, I’ve never been able to successful do gpu passthrough on qubes though

may as well try now

You can read Create a Gaming HVM

With an NVIDIA card in an external GPU using in discrete mode (not using the display connected to the card, that’s not practical in Qubes OS), I succeeded

1 Like

Thanks yeah, I’ve looked into/made HVMs before on my desktop, but on this laptop I’m not sure how displaying a HVM with its own GPU will work, if at all

But I’ll look into it, when I tried on my desktop awhile ago, my whole system would just fully reboot anytime I tried to launch the qube with the second GPU attached to it

What you would need to use the laptop display is to attach the NVIDIA card to the qube, install the nvidia drivers and use PRIME to do offload rendering. This way, the discrete card is used for computation but the main GPU (the Qube OS one) will just display what the NVIDIA card rendered. It’s like 20% less performant but still 100000000% better than llvmpipe (GPU rendering using the CPU) we have in Qubes OS

ah nice thanks! I’ll look into it