Search Issue Tracker
Fixed in 5.3.1
Votes
2
Found in
5.3.0f3
Issue ID
749690
Regression
Yes
[Android S6] Performance drop in 5.3.0f3 compared to 5.3.0f2
android S6 performance regression from f2 to f3.
Test 1) We recreated a custom build of the Flier Project mentioned earlier that simulates a VR rendering load without the VR rendering pipeline. We added two cameras with equallay sized viewports that render onto each half of the android screen then disabled VR mode. I tested this on an S6 and found the performance to be in general worse on f3 than f2 regardless of the Dyanmic Batching setting.
Test 2) After deciding that this is a GPU bound issue we created a small but GPU heavy test scene to isolate the problem. In this project (which I will attach below) we built a stock android application and observed much better performance on 5.3.0f2 than on 5.3f3. The issue seems to be a part of the 'Clear' call in Graphics.RenderOpaqueGeometry. It takes over 20ms to complete in 5.3.0f3 . (see screen shot attached).
All about bugs
View bugs we have successfully reproduced, and vote for the bugs you want to see fixed most urgently.
Latest issues
- [Android] Audio sound is lower in the Player compared to the native music player
- Feet slide and misalign when playing retargeted animations
- IndexOutOfRangeException is thrown in NativePassCompiler when a graphics buffer is used in more than 5 render passes
- "Undo Stack Overflow" error is thrown, and Undo History is deleted when multiselected GameObjects are reparented to their GrandParents
- SearchColumn of type "ObjectReference" does not get refreshed for the custom SearchProvider when lighting generation finishes
Add comment