Since upgrading our project from 2020.1.15f1 to 2020.3.3f1, we have noticed a significant regression of in-editor performance during play mode. We have our build target set to PS5 (I already posted this on DevNet but as it doesn’t seem to affect performance on the console, they suggested I post here as well).
I have already submitted a bug report with the ID [ 1331167 ], but I thought it would be worth posting in the forum as well in case anyone else is experiencing a similar regression.
Note: I’m currently looking for either a solution, or for anyone else that can confirm and also submit a bug-report so we can reach a solution faster.
In summary:
The packges in use did not change much when upgrading, apart from the HDRP. 2020.1.15f1 is using 8.2.0, and 2020.3.3f1 is using 10.4.0. In a minimal-repro project, the regression is still the same even when ensuring that all other packages are the same version.
I tried lowering the HDRP version in the 2020.3.3f1 LTS project, but it looks like it’s not possible to automagically downgrade the ShaderGraph shaders so I gave up - I can try again if required but honestly I’m a but pushed for time at the moment - sorry! (I’m not sure if 2020.3.3f1 even supports HDRP 8.2.0, or if 2020.1.15f1 supports 10.4.0 and vice versa).
We are experiencing around a 25-40% drop in in-editor performance during play mode, depending on what’s happening in the scene. It’s usually around 25% but it looks like once the CPU load increases, it gets worse.
This may also be an issue outside of play mode, but due to the reduced frame rate and lower CPU load, it’s not so much of an issue.
Performance doesn’t seem to have changed for builds run on a PS5 Dev Kit.
We don’t really have a workaround for this at the moment. It’s possible to double the FPS for one view by hiding the other with a different window, but realistically we need to be able to see both in play-mode.
At this point it looks like it’s this regression was introduced either by the upgraded HDRP package, or changes to the editor itself.
While investigating, I made sure to use exactly the same scene, with exactly the same in-game camera position, and exactly the same scene-view camera position.
I have attached some general notes and stats taking during this investigation, as the following attachment:
version-up-performance-notes.txt
The bug report mentioned above includes profiler data dumps, and I have also shared minimum-repro projects via the blind FTP as detailed in the bug-report.
If you have any other questions, or need any other information, please let me know and I’ll get on it ASAP.
Thanks for your post. I have already looked through that thread and it looks unrelated to the issue I’m facing here. I have not seen any increase in the prevalence of “Busy… Hold on…” type notifications.
same here in a 2D game, the editor was super slow but on play mode everything was fine, the problem was in the tilemap, basically it was merging colliders and could not handle the calculations, lowering the number of components to calculate fixed the problem, it is something that causes no issue whatsoever in 2020 but makes the editor ridiculously slow in 2021, I found my issue in the old way, removing the gameobjects by groups and seeing which one would make a difference, then narrowing down from there, eventually I found exactly which object and which component in that object were causing the issue, I know it is of little help but for what I read looking for my issue people is having problems with very different stuff.
I suspect that your issue is unrelated to mine, as I am only really seeing the performance regression during play mode, and the editor version I’m using is currently 2020.X, not 2021.X.
There was a change somewhat recently to the game view’s “stat” tab. In essence, Unity discovered that the fps readout there was poorly implemented bogus, and changed it to give a more accurate result. There’s a thread about it … somewhere.
If you’re using that, try comparing profiler results instead. If you’re comparing profiler results, you’ve probably hit a performance regression bug, and should try the bug reporter!
I’m measuring performance using the profiler. I have also already submitted a bug report with the ID 1331167.
The FPS readings I gathered where taken using a custom script that manually tracks the real-time between frames, so it should be accurate. As you mentioned, the “Stats” tab does not give accurate data, or at least it doesn’t feel like it does, so I have been doing it like this for a while.
What is your exact question? I find it hard to understand the reason for bump. You didn’t ask for anything. You submitted your bug report. What can we do for you?