Arch Linux

nannerpussy

Member

Registered: 2017-02-15

Posts: 96

Anyone have notable success using PRIME w/ 2 PCI-e GPUs?

I spent a week or so learning how to get an older Radeon to work with my AMD mainboard alongside a newer nVIDIA GTX. It was complicated by the fact that I have just one monitor and have to use DVI on one card and HDMI on the other, while also dealing with the nVIDIA VRAM dying (I think). The reason I don't just use the GTX in the primary x16 port is it's posting green horizontal lines from BIOS onward, by it's hard to describe because I can "erase" the lines with my mouse, and refreshing and playing with xrandr can change the patterns, etc. That's not why I'm posting though. What I'm wondering is if anyone out there has managed a similar setup and seen meaningful performance gains in Mesa/Gallium/D3D apps/games?

As it stands, I have everything configured right, am using DRI_PRIME properly to switch to the nVIDIA for discrete GPU rendering, and can run the mesa-demos and glx tests just fine. In fact I am trying to limit the FPS on those because I'm getting something like 1265 FPS per test, which can't be right. But the point is, performance looks and acts fine UNTIL I run a 3D game - specifically Steam games with DRI_PRIME=1 pre-pended. Is there some weird 32-bit bottleneck somewhere I'm unaware of? Would setting up a VirtualBox and doing true VFIO GPU throughput with a non-Windows installation be worth trying? I don't own any Windows keys, so it's gotta be Linux.

This is more "theory-craft-fucking-and-breaking" at this point I suppose, but I legitimately can't find any forums or good info on doing Radeon/NVIDIA dual GPU setups and whether it's even worth it with glamor/32bit dependencies/Gallium9/etc. There's a few posts from Phoronix on the idea of this and him maybe testing it once with GearsOnGallium, but that's about it.

[not sure if any outputs or confs or cats would be useful here since I'm mostly looking for anyone who has experimented with this]