Hi there.
I recently upgraded a project from 5.5 to 5.7 and then 5.8 and have since started experiencing full windows lockups when using multiplayer PIE with VR preview and SteamVR as the OpenXR runtime.
Spinning up the multiplayer instances (either as single or standalone processes) fully locks my computer to the point where mouse inputs only update every couple of seconds. With 1-2 instances, it’ll hang for 30-60s and then come good - but with lower on-device framerate than the 5.5 project. With 3+ instances, it’ll hard lock the PC until I power cycle it.
Switching to MetaXR as the primary OpenXR runtime completely resolves the issue (but comes with it’s own problems due to Meta’s intermittent driver issues)
I’ve tried resetting all configs to default, testing with an empty project and with the VR template to no aval. I’ve also found a few forum posts from folks experiencing similar issues in 5.7.
Happy to capture some logs on Monday from editor and SteamVR, but couldn’t see anything out of the ordinary in output log. Otherwise, what is my best path forward in resolving this issue? - ideally without rolling back to 5.5 or a custom engine build.
[Attachment Removed]
Attaching logs from Unreal and SteamVR. Initial PIE w/ 3 clients was mostly successful, but slightly less performant than 5.5. Second launch w/ 5 clients struggled. Restarted editor and then launched PIE again w/ 3 clients which froze on HMD and ran single digit framerates on desktop window.
[Attachment Removed]
Hi Trent,
We’ve observed the same PC slowdown/hangs on SteamVR that you’re describing internally and found that they generally occur when VRAM usage is maxed out. Could you try reducing the resolution (or other quality settings) on your multiplayer PIE session and see if there’s a point where the issue no longer occurs?
VRAM usage tends to present no issues until it’s maxed out, at which point performance degrades significantly, and SteamVR in particular seems to lock up the entire PC when over budget. We’ve also seen that Quest + SteamVR has noticably higher VRAM usage than Quest + Horizon Link (unfortunately I don’t have comparative stats on Vive + SteamVR).
Let me know if you see any improvement from that and we can go from there.
Christopher Fiala
[Attachment Removed]
Hi Christopher
Thanks for getting back to me. Its reassuring to know it’s not just on my end.
You’re definitely correct that it’s VRAM related. PIE runs okay until it goes over my GPU limit and then locks up.
I grabbed a 5.5.4 version of the project of version control (which is a completely identical scene and pawn/pc) and tested it against the current 5.8 and noticed that it’s VRAM requirements are far lower per PIE instance.
The below measurements were taken on a 5080 laptop GPU using task manager’s resource viewer.
5.8 (High Scalability)
- Editor 5/16GB VRAM
- 1x Instance = 12/16GB
- 3x Instances = 14/16GB - Rough framerate on HMD, but okay in desktop preview
- 4x Instances = 17/16GB - Hard lock on start/end, awful framerate
5.5.4 (High Scalability)
- Editor = 4.2/16GB VRAM
- 1x Instance = 6.8/16GB
- 5x Instances = 11/16GB
- 8x Instances = 15/16GB - Awful framerate as expected, but no lock on start/end of PIE
I diffed the two configuration files and noted that Substrate and Virtual Textures were on by default in 5.8, but disabling them and relaunching made no difference to my VRAM numbers.
I’ve attached the two DefaultEngine.ini’s below in case there is something that I’ve missed, but it seems like 5.8 is demanding far more VRAM per-instance than 5.5 for the same scene.
How can I go about exploring this further to identify what is hammering my VRAM?
Thanks again for your help!
Trent
[Attachment Removed]
Hi Trent, when trying to reproduce your issue I came across the opposite results where 5.8 is using ~25% less VRAM (9885mb) than 5.5 (13310mb) with 3 PIE Clients (Blank template w. OpenXR/ISR enabled, SteamVR/Index). The only main difference between our setups is that I’m on a 5080 desktop GPU, and you’re on a laptop. That unfortunately doesn’t leave us with anything to follow up on, but as a note we don’t expect anything other than 2-3 clients to work well w. VR Preview even on a desktop GPU.
[Attachment Removed]
Hi Trent,
Following up on Victor’s comment, if you want to do any further investigation on your end, you may be able to use the Render Resource Viewer to inspect what’s taking up the extra VRAM between versions on your machine.
[Attachment Removed]