Hello, I recently updated my build and started gaming at 4K, and LG performance is terrible. Used to work fine at 1080p.
I am running a 7950X with iGPU for the host, Asus ProArt B650, 32GB 6000 CL30 dual channel RAM and an AMD 7900XTX GPU for the guest using Windows 10.
With the guest GPU either plugged into my monitor via DP or to an HDMI dummy and using LG, I get ~40fps (confirmed by Afterburner and LG FPS and UPS) at a 100% GPU utilization.
By directly using the monitor via DP, I get 140+fps.
I am on Arch Linux 6.2.11, LG B6 both host and guest, using KDE with X11 with “VariableRefresh” and “TearFree” options enabled and confirmed to be working. shmem is set at 128M.
I am worried that I might have set the wrong setting somewhere and I feel like the Linux host is actually causing the issues, because driving the display directly from the guest GPU works as expected.
Is there something I can do to troubleshoot or try?
Your bottleneck is probably your iGPU. iGPUs lack VRAM and use the system RAM as a “poor man’s” VRAM. Using an iGPU has a dramatic performance hit with LookingGlass. Once your expectations go beyond the basic 1080p@60, you should seriously consider following our recommended requirements.
Also provide at the very minimum your LG client log (terminal output) so we can see how you have your system configured and if there are any glaring issues.
are you certain?
i am on x670e aorus master with a 7800x3d and 32gb@6k with a 4080.
Using a 1440p ultrawide i can not yet see any bottleneck even up to 165hz/fps
Yes I’m certain. iGPUs do not use proprietary VRAM. They share system RAM with the CPU. Looking-Glass is a massive data mover application. Large data blocks need to be moved around very fast. The problem arises when your iGPU needs to move data into system RAM, it will pause any CPU governed memory transfer. Basically your CPU will now wait during those transfers. As Looking-Glass is moving those large blocks at the same time, you’ll get at least 50% bandwidth limitation compared to a dGPU that can execute those transfers independently of the CPU RAM moves. Performance is dependent on a myriad of parameters and modern iGPUs can perform pretty OK, but that minimum 50% bandwidth limit is a killer on many system setups.
I got that iGPU/IDG does not have VRAM (:
but i have to say that thoe bandwith limit so far did not affect me.
And as i wrote i push 4k60Hz 4:4:4 RGB or 1440uw at RGB4:4:4.
Or i just do no notice the difference? (i had the same setup on AM4 with a wx3200 for the host and a 6900xt for the guest) But lookingglass performance overlay always shows the same FPS as my RTSS
therefor i would wager that your point is technically correct but not affecting the small workload of lookingglass
We have gone to great lengths to get iGPU performance reasonable. On more modern hardware the performance isn’t as bad as it was but it’s still far from optimal.
Great, you’re one of the lucky few. There are hundreds of reports of poor performance when using iGPUs.
ok havent thought that.
the theroy looked great for that. And since ive recently sold all my gpus which are not i use i am happy that i feel no bottleneck ^^
7800x3d is as new as it get though.
@gnif is there a way to measure fps/performance of LG aside the build in fps counter? to check that i really am not missing frames?
I wonder if the 96MB of cache is what makes the difference there? It’s about 4x the size of the frame to be copied, which is a nice amount of room to operate in.
I would think, though, that PCIE bandwidth would be a problem faster than memory bandwidth for something like lookingglass. It should only be a couple gigs/second?
You can use a high-speed camera to record the screen if you have one. Afaik, most high end phones have a 120fps camera option. Just scrub through for doubled up frames. I did that to sanity check Gsync frame pacing problems in newer nvidia drivers on Windows.
havent thought that.