Yellow-Dog-Man
PhotonDust: Persistent performance issues after excessive particle induced slowdown on Linux
Describe the bug?
Test/Reproduction world here: https://go.resonite.com/record/U-qualia/R-25ba71b3-ac2f-43dc-bffc-c96633ec3653
A PhotonDust simulation with an extreme number of active particles will slow down Resonite's overall performance, as expected.
On Windows, when the particle count is reduced, either by particle lifetime or parameter adjustment, Resonite's overall framerate will return to normal.
On Linux (testers wanted!), when the particle count is reduced after an extreme-particle-event, if any particle system is still active anywhere in a world with even a single active particle, the original framerate impact felt in the initial particle excursion will still be felt, at least partially. This leads to significant persistent performance degradation which requires a client restart to resolve.
Leaving the world in which it occurred does not resolve it. Removing every active and emitting particle system from a world will restore the original performance. Particle systems that are active but are not emitting/simulating particles do not have an impact. A one-particle system completely stripped of simulation modules is sufficient to cause the issue, as long as that single particle is active in the world.
This appears to be entirely specific to Linux. I do not experience this same issue on Windows, on the same hardware.
To Reproduce
On Linux, start a new empty gridspace Create a new ParticleSystem. Note the fast simulation speed Set the emitter rate to some very high value, like 1,000,000 Set the maximum particle limit to some very high value, like 1,000,000. Note the slow simulation speed Reduce the maximum particle limit to the original 100. Simulation remains slow, with commensurate impact on framerate Disable the entire ParticleSystem and performance returns. Re-enable and framerate drops again
After this, any active ParticleSystem in any world, regardless of heaviness, will have a similarly severe performance impact, until the client is restarted.
Expected behavior
The PhotonDust engine should return to its original high performance when a poorly-performing particle system is rectified or disabled/destroyed. This slowdown should also not persist across all active instances of particle systems.
Screenshots
https://github.com/user-attachments/assets/681afa36-16ef-4d3b-9382-929c1d35612f
Resonite Version Number
2025.4.10.1305 through 2025.8.9.420 inclusive
What Platforms does this occur on?
Linux only
What headset if any do you use?
Quest 2, WiVRn
Log Files
SPINCOAT - 2025.4.10.1305 - 2025-04-17 00_43_42.log
Additional Context
14-core Xeon W 2275 (no HT), 64GB ECC DDR4, 2080 Super, GE-Proton9-22-rtsp17-1 (now GE-Proton10-10)
Experienced previously while designing complex particle effects, and also now while attempting to hunt down a potentially PhotonDust-related headless crash.
Reporters
qualia (qualium on discord)
I am still experiencing this issue on Beta 2025.9.23.1237, and all prior versions.
I made a reproduction test world and published it, in hopes of getting more feedback.
https://go.resonite.com/record/U-qualia/R-25ba71b3-ac2f-43dc-bffc-c96633ec3653
I've gotten a few Linux users to test this out, and so far only I seem to be affected by this.
Sure thing. Beta 2025.9.23.1237, no mods or command line options, technically on the Satori GC right now. If that's a confounding issue I can try and revert to the old one, but it hasn't changed this particular behavior any.
Before, 10 particles active, >>90 FPS (didn't catch exact number without dash open):
During, 1M particles active, ~8 FPS:
After, 10 particles active, ~13 FPS:
Particle system disabled, ~330 FPS:
I can repro it using the provided world here are my logs
void-live - 2025.9.23.1237 - 2025-09-27 00_13_28.log
Unfortunately, I can not reproduce this issue with the provided setup. (Which is intriguing, because I distinctly remember having encountered the issue once, during the pre-release.)
In order to try to replicate the issue, I duplicated @mxjessie ’s setup 3 times and run all of the particle systems at full blast for a few dozen seconds. And despite my framerate crashing down to 5 FPS when they were all active, it went immediately back to max FPS upon setting them back to low (not turning them off). This log file is from my desktop testing, but I also tested it in VR without encountering the issue either.
PC-Urban-Garden-Endevour - 2025.9.23.1237 - 2025-09-27 02_11_55.log
Side note: if someone wants to get a grabbable spawnable version of @mxjessie ’s setup, here is a copy of it I made for the purpose of my testing: resrec:///U-1Udo02hMUFc/R-99875046-ff2f-4f96-b5dc-e4d14cd69fc4
Can't reproduce on Linux EndeavourOS with Nvidia RTX 4070 Super and AMD Ryzen 5 7600X, using ProtonGE10-17
Tested in Screen mode
owo - 2025.9.23.1237 - 2025-09-27 13_39_34.log
I failed to replicate this on my own, but I have a feeling this is not a linux-specific issue actually, as I have previously had massive near-unexplainable lagspikes in the frog bog (a very particle heavy world) that were very similar to what is described here with lag being persistent even when you wouldn't expect to be without having to disable everything essentially.
The only problem is for me its very inconsistent so I don't know how to force it to happen
@Nytra Your log lists your GPU as "Raphael (rev c7)", can you verify that Resonite wasn't running on your iGPU? Currently trying to rule out this being an NVIDIA-specific issue.
I'm not sure how to verify that aside from the fact that I'm getting more than 100FPS which shouldn't be possible on an iGPU
On Linux, nvtop will show your NVIDIA GPU's current usage stats, including which applications are running on it. If Resonite is running on your NVIDIA GPU, that would be a separate Resonite bug
Yes it's running on the Nvidia GPU. There must be a bug in the way Resonite detects the GPU on Linux.