Been having memory leaks for the past week, starting to get REALLY ANNOYING SEEING FPS DROP TO 2 FPS ON A 9800X3D AND 4090 WHEN THE GAME IS DOING ABSOLUTELY NOTHING.

PLEASE FIX THIS i gotta restart my game every 2 hours now.

I just caught another leak, and this time I paid attention to the steps I took before the leaking memory was actually released to confirm what I thought happened last time in the post below this one, and the behavior was exactly the same.

For some reason, forcing full GC in the debug menu does not actually release the leaked memory. Instead, what I have to do to reclaim it is open the debug menu, force full GC, close it, reopen it, force full GC again, close it again, and by the third or fourth time opening the debug menu the RAM usage will drop. The large initial drop seems to be triggered by opening the debug menu on that third or fourth time, and then from there I can force full GC to clean up the remaining several GB and return VRChat's memory usage to normal.

Now I don't know if the leak is causing some circular reference that doesn't let GC take care of it until it's forced a few times or what's going on, but simply pressing the force full GC button will not resolve the leak. You have to do it a few times, close the debug menu, reopen it, etc before the memory is released. I figured mentioning this is probably useful if the devs are even reading this.

The game leaks even on safe mode. It's not someone's avatar. No asset bundles were even loaded because I've had the game in safe mode to eliminate that variable. Yet I still encounter OOM crashes quite frequently. However, I managed to catch this one at 51GB and a minute or two after forcing full GC with the debug panel, the excess leaking heap was released.

I'm guessing the mono managed heap starts leaking during some allocation event and fails to release its unused heap pages during normal garbage collection. So this isn't a traditional memory leak caused by something related to Unity. Unity's memory usage didn't change. It's VRChat itself.

It is not practical to keep approximately 10 GB of cache in memory while remaining stationary in a specific, small-scale world—especially when running on a system equipped with an SSD.

I upped the limit another 10GB to 48GB because I'd be more AFK those days and wanted more time to catch it, yet the game leaked out and crashed three times in one night and it was always soon after I had to step away for a few hours every time and not during the few hours I was there to monitor it. However, after turning on safe mode and idling in the same instance after the third OOM crash, the memory leak stopped occurring. It's something on someone's avatar, not something in the world.

Next time it happens I'll use the in game debug menu to capture more info. I didn't know it existed until very recently.

I had this happen yesterday, first time ever. It was busy earlier but at the time of it happening I think there was only myself and one other random person somewhere in the instance. VRChat suddenly started eating up all my available remaining RAM (32gb) with very high CPU usage, and the only way to restore functionality to my PC was to restart VRChat. The session was earlier in my day, around 2.5 hours, and I wasn't in the world for too long when the leak started.

It's currently leaking again as I type this and I've only been on for an hour. I don't know what else I can add to this issue except to say that it happens an awful lot lately. Same world every time, but I'm not sure if it's because something in the world is causing it or just because that's where I idle most of the time.

I managed to capture definitive proof of the memory leak in action. I began logging the process as soon as I noticed the memory usage climbing and captured continuous memory statistics and smap, pmap, and lsof snapshots at several points during the leak. Hopefully you (the devs) can use this to finally pin it down. My VRChat game log is also included. You can probably line them up to see what caused the leak.

After switching to another world, RSS did not return to normal. It remained at around 30.8 GB with 30.4 GB of it dirty. The snapshot immediately before the world change was 30.0 GB RSS. I let it go for a bit longer before hopping worlds. Some virtual address space was released, but the committed anonymous memory was retained until the client was closed.

Edit: The order of the logs in pairs of three is initial -> later -> final, with the logs labeled 'final' occurring AFTER I switched to my home world and 'later' occurring a bit before that. It's also worth noting that switching worlds did stop the leak.

It happened again last night. I never switched worlds, I basically stayed in the same place for 5 hours. These are the final logs in the log file before systemd killed the process. Memory usage over time is not logged because I'm not switching worlds.

2026.07.25 22:41:50 Debug - [Behaviour] OnPlayerLeftRoom

2026.07.25 22:41:50 Debug - [Behaviour] OnPlayerLeft Murphite (edited out ID)

2026.07.25 22:41:50 Debug - Removed player Murphite

2026.07.25 22:41:50 Debug - [Behaviour] Destroying Murphite

systemd log:

Jul 25 22:42:34 nibl-linux kernel: Memory cgroup out of memory: Killed process 36602 (VRChat.exe) total-vm:85154820kB, anon-rss:39261664kB, file-rss:406132kB, shmem-rss:65144kB, UID:1000 pgtables:82144kB oom_score_adj:200

I don't think it's something I'm doing. I think it's somebody's avatar or something like that. Something I can't reproduce myself. Because I was literally AFK tabbed out in an idle game world when it triggered OOM. There's days where I can go two days there without this happening, but tonight only lasted 5 hours. I wish I could help diagnose this more but I don't have any clues. My own actions aren't causing it. I'm not doing anything different from the days where it goes 2 days straight just fine.