Unity 5.3 Performance issues

Users are noting higher CPU usage on builds in the 5.3 versus 5.2.3. Here are some screens from the profiler:
5.2.3

5.3


Same scene from the exact same camera angle. (Lights were rebaked for 5.3)

5.2

5.3

 

Okay set Unity 5.3 back to OpenGL 2.1 and now the profiler looks like this:



This is on OSX, but users on Windows are reporting worse performance overall. Their defaults seem to be Direct3d9.

 

I'm getting a very bad performance as well with OpenGL 4.1 profile on the latest El Capitan:





I'm running this on a late 2012 quad core iMac with GeForce GTX 680MX.

This gets fixed if I switch back to the old OpenGL profile.

 

Hmm this is really interesting! Thank you so much for posting! I'm going to send this forum thread over to the WebGL team

 

Note: These are native OSX64 and Windows 64 version that are receiving worse performance. Not sure if WebGL affects?

 

Oh gosh! I totally mis-read! Thank you so much! I've posted to the graphics team

 

On iOS, the same game scene now has doubled CPU time per frame after updating to 5.3, which caused frame rate to drop by 1~4 fps. I used dozens of particle systems but it was 60 fps smooth in 5.2.3p2.

 

It could be one of the new particle features that create the issue, i hope this will be addressed swiftly.

I read particles have some new shadowing features too, maybe these are heavy and are enabled by default ?

 

Same problem, changing the editor back to opengl 2.1 solves the issue for me.

 

On a 2015 Macbook Pro (dual core 3,1 GHz Intel Core i7 + Intel Iris Graphics 6100) this problem does not manifest itself. What's going on?

 

I tested a bit more, some times the particles work ok, others they have some very bad spikes and hangs that never appeared before in previous versions. Another guess is that the new threading system bottlenecks and maybe it increases speed temporarily, but create a hang at random (which is what i experience)

 

That's probably the cause. Looking forward to p1.

 

I actually found some of the profiler differences in the original post to be of interest:
Used Textures
5.2 - 109 - 238.4 MB
5.3 - 0 - 0 B

VRAM usage
5.2 - 48.2 MB to 328.6 MB
5.3 - 48.2 MB to 90.1 MB

There were slight differences in draw calls and batching and a fairly significant difference in SetPass calls (going from 1015 to 731). It's possible that the profiler isn't returning the proper information for OpenGL 4.1 in Unity 5.3. It does look like these differences should be favorable for 5.3, yet it performs slower on the CPU side.

 

I'm also getting awful performance on OSX builds (about -50%) with OpenGL 4.1. Anyone from Unity know why?

 

I have a project that I'm using to test vertex/poly limits that was running on a particular Android device at 30fps using Unity 5.2.3f1. After upgrading to Unity 5.3, that frame rate has dropped to 7fps. I cannot attach the profiler on Unity 5.3 as doing so immediately crashes the application for memory (this is not the case on 5.2.3f1). Whatever is going on seems to be across multiple platforms.

 

I've got also awful performances on Android builds.
I tested with a scene with a bunch of meshes.
Some of them usign a particule additive shadder
near 200ms just for the transparent rendering in profiler ! 4fps !!
The same scene was 60fps stable on 5.2 on the same targeted device just before upgrading.

Any idea why ? I am really interested by some features of 5.3, but for now, I will roll back

 

Using a 27" 2013 iMac 2.5ghz i7 8Gig ram. OSX 10.10.5.

In editor and builds the FPS is capped to 30. It used to be between 30-60 in editor and a rock solid 60FPS in builds.

There must be a bottleneck somewhere that's causing this 30FPS limit.

Because of the low FPS, if I apply any kind of post-process/image effect it drops to 15FPS. Regardless of which one I pick. I appreciate post process is expensive, but it shouldn't be a universal 50% hit.

@holliebuckets or someone in Unity : can you please confirm this bug and/or suggest further tests?
*Edit: not certain if this helps :
- I had to disable AA for 'Beautiful' and 'Fantastic' as it made everything washed out. So something is going wrong there
- Builds run just fine on Windows
- Seems to run just fine on a newer OSX 10.11 5k iMac (2014, i7, 24 gig of ram).
- Shrinking the editor window to something tiny (200x200 ish) makes the FPS to go over 45. So probably not a hard cap... Just awful performance on this machine.

 

Let me know when i can upgrade from 5.1.2f (stable and fast) to new that are infact better and not just new functions .

5.2 and 5.3 seams like beta, and not very well tested. UI low performace...both this version.. Fix it please..

 

Submit bug reports, please...

 

Are you serious? It's hard to create a simple default scene to test the engine?

Check that http://forum.unity3d.com/threads/performance-issue-after-5-1-3.373038/

 

There are many factors that can lead to issues. When you submit a bug report, Unity will know exactly which operating system was used, which graphics drivers are involved, what exact settings were used, ... .
There are tons of factors that have an impact to the performance and the combination of them matters. I have comparable setups to some of the people who are posting they are having performance issues, but I don't. In order to react, it is necessary for Unity to be able to reproduce the issues and for that, this kind of information is invaluable.

 

When you have between 1/5% of people crying about performance issue, it's understandable, but it's not the case since unity 5. Every update,something is broken, it's why some user stay on the 4.x.x cycle.

When you see that we have a big problem with the character controller since unity 5 and that the answer from the unity staff is that we need to wait for the 5.1.4 (for march) because they can't fixe it before ..... one year after the release of unity 5.

 

Happy to take some time out to help track down the issue, however I can't send the entire project over (commercial reasons) - so some kind of guidance from a Unity member of staff on what we should be testing/looking out for would be greatly appreciated.

I would find this less irksome if Unity's marketing team didn't bang on about their wonderful testing systems/methodologies. For me this is the second half baked release in the 5.x cycle. (But that's probably just me being a grumpy git.)

 

I don't know how many people have problems with the performance. What is clear for me is that some of those people are very loud, especially the ones who don't submit bug reports. That's not productive at all. Many of the people who took the time to submit good bug reports later on replied that their issues were resolved.
There are also a lot of loud people who started new projects in Unity 5 and are using all kinds of functionality on mobiles that is not supposed to be used or is known to be slow. Those ones should be taken out of the equation.

All in all, bug reporting is worth it in my opinion.

Can't say anything about that.

 

Dantus, I think the issue here is people running the exact same project and scene files and getting worse results.

 

Yes, that's possible. But if that is clearly the case, submitting a bug report would kind of make sense, such that Unity can analyze exactly what is going on.

If you check the posts in this thread, it is not that unlikely that different there can be different causes leading to worse performance. Editor, iOS, particles are mentioned. Which iOS version is involved, what are the settings? When it comes to the particle system, there are also tons of settings and the device has a huge impact.
With an actual bug report, Unity can have a look at the actual issue and doesn't need to guess where it might be. I am sure they don't make this guessing game at all, because it is a waste of time.

 

Exactly. Each time I upgrade a project I get a downgrade in performance. I jumped to 5.3 because I thought the new GL core would speed things up. (very wrong)

Submitted bug report. As I can't attach the project I'm guessing it'll be ignored

 

Especially for performance regressions, a reproduction is unavoidable. It can be graphics, physics, multi-threading issue and many more. Those are just some of the areas in which a lot of modifications were done. How is Unity supposed to find out where the bottleneck in your particular case is if they can't try it out?

 

You're right. I would hope they'd have some in editor analytics that are sent over with bug reports ? Assuming it's a bit more than a system info dump?

Perhaps we should start posting specs here?

(IMHO asking users to attach and send commercial projects (often large and confidential) is ridiculous. Also, from a purely business POV, time spent helping Unity fix borked code is time/money down the drain. )

 

Maybe because we have a big project in production? Do you have one yet?
My game is not based on mobile and was running perfectly fine in the 4.x.x cycle (and now on 5.1.3 after a lot of headache).

Since unity 5 a lot of people like me have a problem with the character controller, the issue is here http://forum.unity3d.com/threads/character-controller-problem-since-unity-5.370766/ with a nice answer : wait 4/5 months, funny .....

Want another one? check that http://forum.unity3d.com/threads/5-2-caret-missing-in-inputfield-even-in-the-default-one.353817/ and tell me if it's not a joke, they have released the 5.2 without testing a simple input caret ! wtf seriously ...

About making a bug report, it's ok for me when it's really needed. In my last thread here http://forum.unity3d.com/threads/performance-issue-after-5-1-3.373038/ i have created a simple scene with 4 planes on it. yeah really hard to reproduce .....

It's maybe not your case, be we and i have paid for a final produce , not a broken one.
I can understand your point of view for a beta version, not for a final release.

I'm not a loud guy, but i know how unity work and the performance that i can have with it. But since unity 5 it's the descent to the hell.

 

Even if it is a little more information, it is by far not enough to identify issues. And without knowing what the actual issue is, nothing can be done.

Many projects are large, many projects are confidential. There is still the possibility to create a reproduction.

From a business point of view, it is always a risk to update any software package, because something may be broken. That's why many either don't update, or consciously allocate some time for it, always with the option to go back. If you spend the time to upgrade your large project, you should also be able to identify the bottleneck and to create a reproduction for it, if you can't send the actual project. If you don't do that, you wasted time for absolutely nothing. That's a lot harder to justify from a business point of view!

 

Yes, there have been serious issues in Unity 5. But that doesn't prevent anyone from submitting bug reports. When you just report about an issue here in the forum that is very easy to reproduce, why don't you submit an actual bug report? And if you submit a bug report, you should make it public, such that others with the same issue can vote for it in the issue tracker to show to Unity that this is an important issue that should get a high priority.
Just posting in the forum won't make anything better. Maybe it helps to let off a bit of steam, but other than that, no actual progress can be expected.

 

Well hopefully Holliebuckets submit to the graphics team helps

 

Has anyone sent a bug report with an example project so graphics can take a closer look? If you have please post the bug number(s) here. Seems like there may be a few things going on here.

 

Hello,

I think that this thread http://forum.unity3d.com/threads/performance-issue-after-5-1-3.373038/ is explicit already.
Simple scene with 4 planes.

Thank you.

 

According to my understanding the profiler pictures show that both the CPU and GPU usage were considerably improved since Unity 5.1.3.

 

Yes indeed. Our projects are very big, and very confidential. And when needed, we provide example projects to show a bug. We literally have to. Our projects simply don't run self contained, or outside our building.

You don't (and shouldn't) be submitting your whole project, it will just make the search more complex and doesn't really benefit Unity or you. You make a small project that isolates the exact bug you believe you are encountering. Honestly, you should be doing this anyway (separate from filing a bug report), it is how you trouble-shoot. Create a new project copy out enough to isolate the problem and remove elements until you discover the core/repeatable issue. In most cases, you will discover that it is not an engine bug, but one with your code. If it is not, then you can file a bug report and attach the example project.

Very true. And should be pretty obvious, clearly things have changed. We vet most new versions of Unity from time to time, but only upgrade if there is a clear reason to. Unity is a big tool, games tend to be complex. You really have to take care to avoid problems.

Absolutely not. Attaching repro projects is the best way for you to get your bug attention and to help Unity resolve it quickly. It is hardly a waste of time/money, as it improves the chance of getting it solved, and more importantly, it doesn't suck up valuable dev time on Unity's side pouring over every random "it don't work" noob bug. They have reported before that only a small fraction of bug reports are actually bugs. Putting less attention on bugs with no repo example or clearly defined issue is benefical to ALL developers.

This is exactly why support has to put some sort of reasonable restrictions on what they investigate and spend time on. This isn't a bug.

Not really a huge issue. Those demo/standard assets are a starting point. Sure, it would be nice if they fixed it at some point, but as far as priorities go, it should be pretty low. Mainly because it is pretty simple to fix or change it to fit your own needs, or as most developers do, create your own. Editor bugs that are internal are not ones that the end-user can correct or deal with directly, those are much more critical. Really, it's just a script, why have
you been waiting for 4-5 months? Just fix it or write your own.

---

Ultimately, Unity has bugs, and they are always updating and fixing them. This will always be the case. (as with any large 3rd party tool). You can help yourself (and the entire community), by submitting possible bugs in the proper way, so they can be isolated and more quickly be addressed. Or you can complain about it on the forums, and demand that Unity read your mind and trobleshoot your game? Which do you think will help you ship your game?

 

@Dantus : people are not loud here. Just kind of disapointed
It can takes a lot of time to point the problem... and submitting a bug report with the scene can be hard with assets...
Here, I reinstall 5.2 and try to show the strange differences between 5.2.3p3 and 5.3.0f3 strange diff...

On 5.2, transparent rendering takes between 4 and 5 ms > 60fps approx


On 5.3, same scene, same project, same device, the transparent rendering takes 130 - 150 ms > 6fps

I keep both builded apk if it can be usefull for unity staff.
I will try to submit a bug report but It will takes a lot of time to remove all scripts and things to focus on this problem

 

Not everyone is able to/willing to attached their project to a bug report. In fact, I'd say the majority of people can't.

"Submit a bug report" should never override Unity releasing a stable and well-tested version of the engine. It's understandable that all software has bugs, but more often than not new releases seem to bring these MAJOR bugs that I don't think should have got past QA in the first place.

 

@Aurore Just submitted a case : 752924 with a scene to build on android on 5.2 then on 5.3

 

I experience a severe slow down with my particle systems as well, there are huge spikes that were not present before and everything seems to get stuck at random. I hope the next update just restores the previous particle system as the new one is unusable right now and release it as beta.

I do not get the same issue with InfiniGRASS with transparent grass though, this seems unaffected, so my guess is that this is an issue with the new particle system.

The above observations are on PC.

 

Hi,

I can send my Particle Dynamic Magic pack, the performance was smooth in v.5.1.1f1 and now it is getting some major spikes and hangs at random in many of the effects. Or you can directly download it from the store and test with it i guess.

I have not build a demo with the system in v5.3, so i cannot say whether the issue is happening only in the editor due to some new editor related overhead or is happening in general. You may also use my store webplayer demos for reference on the previous particle / decal systems speed. These are also included in the asset.

 

I've posted one in a different thread:

(Case 752933) Massive CPU spikes under OS X and Open GL 4.1 in Standard Assets Example Project

 

Still using 5.1.2 too: http://i.imgur.com/uKELSpy.png

 

I get the same big random spikes in Windows OS and DirectX9 as well, so it is probably a general issue.

I wish there is an instant update that restores the previous particle system as a first step and a beta testing for the new one for a lot more time and with more users. This can cause a massive downrating in my particle asset (Particle Dynamic Magic) that will now seem like not working while it worked fine in v5.1.1.f1.

I hope i wont have to remove the asset, it would be amazing to have the previous particle system and its speed restored as it was and the new one be a separate system (like the legacy particle system and the current one)

 

@nasos_333 : IMHO the performance issue concerns shaders.
Even without any particle system, if you try to use particle>additive, or mobile>particle>additive the rendering results are awfull. unlit>transparent seems to be same. InfiniGRASS use perhaps a custom shader ?

Please note that even opaque rendering is takes 3 times more time each frame but for this point, an issue already exists and seems to be fixed on 5.3.1.

 

InfiniGRASS does use a custom shader i created by altering a base shader i created with ShaderForge, maybe that is why it is unaffected.

If the standard shaders are the issue, then this is a relief as changing the shader handling is probably much easier than the whole particle system.

I do not see a difference in the volumetric clouds in Sky Master either, these also use a custom shader in v5.3 and the older legacy particle system.

It would be great if the new versions beta was extended to all Unity users, so there is a massive testing phase before each major release that does radical changes like v5.3.

 

Dear all,
I have a really big project (a game almost finished) and I decided to update unity from 5.1.2 to 5.3 because of the new particle features. Without doing an extensive profiling, I agree is noticeable slower overall. My builds are for Windows, and ceteris paribus, I get random framerate drops that are quite akward.
I appreciate the new features in the field of the shuriken particle system, but this performance issue is a bit annoying.

 

Good work!

 

Its very simple, performance is not a bug, its the canvas (UI elements) to world matrix that uses allof singe thread CPU power.

When over 50 of ppl on the forum says same thing about the UI, maybe worth looking in to?