Disappointed with performance

[Joachim_Ante]

If you have any performance regressions with 3.5 we would like to know about it.
In all our tests we ran internally and for beta testers as well performance is somewhere between 10% to 100% faster depending on what you are doing.

It is always possible that some use cases are slower. We are eager to fix them if we can reproduce them.
For us to fix them you have to file a bug with the full 3.4.2 based project folder attached in a state where it can be opened on both 3.4.2 and 3.5. So that we can do comparisons and then optimize.


So far we haven't yet gotten an actual project folder from this thread, that lets us profile and optimize. This is unfortunate.

[sawfish]

If you have any performance regressions with 3.5 we would like to know about it.
In all our tests we ran internally and for beta testers as well performance is somewhere between 10% to 100% faster depending on what you are doing.

It is always possible that some use cases are slower. We are eager to fix them if we can reproduce them.
For us to fix them you have to file a bug with the full 3.4.2 based project folder attached in a state where it can be opened on both 3.4.2 and 3.5. So that we can do comparisons and then optimize.


So far we haven't yet gotten an actual project folder from this thread, that lets us profile and optimize. This is unfortunate.

I will see if I have a recent backup (before upgrading to 3.5) of the project I'm having trouble with.

I seem to have narrowed it down to being transparency related.

I am building to Android. The FPS dropped from a constant 55fps+ to an unplayably stuttering 25-30fps.

Changing some of the transparent built-in shaders to diffuse brought the FPS back up close to normal.

If I can get the pre-3.5 copy of the project I will send it in.

[Demostenes]

I am experiencing terrible random lags with 3.5f5. Framerate is OK, after while drop to 0 FPS and than OK again. After while again drop....it was OK in 3.5f1

[rortos]

We also reported very poor performance on all our test android devices after upgrade to 3.5. Frame rate falls at least 30% down.
This performance are not accetable and we are goind to re-install 3.4.2.
We are very disappointed..

Rortos

[n0mad]

Hey, instead of just saying "I am disappoint", maybe could people start investigating a bit better in the potential reasons why they encounter performance drops ? (= when, how, which conditions, which type of component used, shaders, etc)

That would speed up the rate of a fix release by helping everyone see some patterns between different cases.

[rbisso]

Joachim - posted a bug report with our project 3.4 and 3.5 versions and let our account manager know. Our issue is here: case 444031.

[kenshin]

Hey, instead of just saying "I am disappoint", maybe could people start investigating a bit better in the potential reasons why they encounter performance drops ? (= when, how, which conditions, which type of component used, shaders, etc)

That would speed up the rate of a fix release by helping everyone see some patterns between different cases.

I totally agree with you!

We all must fill a bug report to help unity tech guys to fix these problem and release soon a fixed release.

[Joachim_Ante]

Joachim - posted a bug report with our project 3.4 and 3.5 versions and let our account manager know. Our issue is here: case 444031

Thank you so much. You are the first person on this thread who replies with something actionable.
We will look into your project folder as soon as possible.

I would appreciate if others will follow this example.

[Joachim_Ante]

We investigated the issue and this is a result of fixing a bug.
In Unity 3.4 realtime lights were not affecting static objects.
In Unity 3.5 realtime lights always affect static and dynamic objects.

The result of this is that if you had previously setup your lighting to work around this bug, then you will now get worse performance as you will receive the lighting twice. It will also look different since your lights will be added twice to the scene.

In this specific case:



Baked only lighting:


The same realtime only lighting:



You have two lights. One marked as baked only. And another one marked as realtime only.
Both lights will affect static geometry. Thus you get another set of draw calls.

The correct way to use these lights is to use a single auto directional light and remove the realtime only light.


If you wish to have different realtime lighting than static lighting, you should using light culling masks to affect only dynamic objects.

[jlevel]

Getting bad performance after upgrade from 3.4 to 3.5 on Android.

What I discovered is the following:

- Galaxy Nexus (Android 4.0, PowerVR) -> kill framerate, poor performance
- Galaxy Tab 10.1 (Android 3.1, Tegra2) -> better performance than Unity 3.4

The same build on 3.4 run fine on both devices

[Joachim_Ante]

I am very curious to see a project folder with the bad android PowerVR performance.
Please file a bug with a project folder attached that has bad powerVR performance on 3.5.

Thanks.

[jlevel]

I am very curious to see a project folder with the bad android PowerVR performance.
Please file a bug with a project folder attached that has bad powerVR performance on 3.5.

Thanks.

I tried to sent my project, but I had some problems with the bug report, was a lot of GB and was stuck for some reason. I'll try to narrow down more and send it soon.

Anyway, I discovered another behaviour:

I switched "Graphics level" from OpenGL 2.0 to OpenGL 1.x and I have a huge performance boost than OpenGL 2.0 on Unity 3.5.

It's curious because on Unity 3.4 I received better performance with OpenGL 2.0 than 1.x

Not sure if it will help

[unity_sg]

Same problem here, a decrease of performance about 25% on a big project (windows)
Unfortunately I can't send the project to report a bug

[Lypheus]

Upgraded from 3.5b -> 3.5f and I'm seeing a huge increase in performance, using stats window i'll get around 120-400fps vs 18-70fps from the previous build (in worst case, 1772 draw calls @ 120fps with 3.5f vs 18fps with 3.5b). For comparison, the wiki fps script reports about ~60fps with new build vs ~18fps with beta.

Demo is at: http://forum.unity3d.com/threads/123...l-Sandbox-Game

No complaints here.

[a1s2d3f4]

Im pretty sure i dropped 30 FPS from 3.5b to 3.5f

[Tiles]

Lightmapping issue does not solve my problem with more drawcalls. I have removed all lights, and have still much more drawcalls than in 3.42. I have stripped it down to just one object. See here: http://forum.unity3d.com/threads/124...gher-drawcalls

[rbisso]

@Joachim - Two fixes have brought us back up to approximately our 3.4 framerate: First, we implemented the culling masks on our realtime and baked lights, as you suggested (Saved about 300-400 draw calls in our worst-case view). Second, as suggested in another thread, we temporarily disabled our blob shadow projector (which saved us another 275-300 draw calls). Thanks so much for your help and we'll be looking out for the projector fix.

[TheGeoff]

One of the issues with sending a 3.4.2 project, at least for us, is the the 3.4->3.5 project is pretty hap-hazard. We had to do it a couple of times to get all of our tags, layers, etc to import properly. And even then some of the physics settings are off.
Did anybody with the performance problems have 3.4->3.5 Upgrade issues? Maybe the upgrade process is where the bug it?

[Paulo Henrique]

~30% Performance boost in a huge scene without realtime lights.

[tarragon]

Sorry, your tests are wrong. QED: 99% of forums.