Search Issue Tracker
Won't Fix
Votes
0
Found in
2018.2.13f1
Issue ID
1096407
Regression
No
Multithread performance is lower than Single thread performance on all iOS devices, some Android devices, macOS and Windows
using new Thread(()
Steps:
1. Build and Run attached project on device
Expected: Multicore performance is higher than single core.
Actual: Multicore performance is lower than single core.
Reproduced with: 2017.3.1p4, 2018.2.17f1, 2019.1.0a9.
Not reproduced with:
TEST RESULTS:
Note:
-higher score is better
-works as expected on Exynos Samsung Galaxy S9+ and macOS Editor.
[iOS]
iPhone X (iOS11.3.1): single: 59838; multi: 51974.
iPhone Xs Max (iOS12.0): single: 72159; multi: 56012.
[Android]
Samsung Galaxy S9+ (SM-G965F), Android 8.0.0, CPU: Exynos 9 Series 9810, GPU: Mali-G72: single: 9693; multi: 16774.
Razer Phone 2 (Phone 2), Android 8.1.0, CPU: Snapdragon 845 SDM845, GPU: Adreno (TM) 630: single: 21741; multi: 17835.
[Editor]
macOS: MacBook Pro 2015 (macOS10.14.1): single: 28182; multi: 36095.
Windows: Intel Xeon @3.7GHz (16 cores): single: 55651; multi: 37454.
[Standalone]
macOS: MacBook Pro 2015 (macOS10.14.1): single: 60001; multi: 52217.
Windows: Intel Xeon @3.7GHz (16 cores): single: 68530; multi: 33765.
Add comment
All about bugs
View bugs we have successfully reproduced, and vote for the bugs you want to see fixed most urgently.
Latest issues
- Crash on RenderBufferManager::Textures::GetTempBuffer when rendering a specific scene
- The Inspector header is incorrectly displayed when scripts are recompiled while a Scriptable Object instance is selected
- NullReferenceException is thrown when overriding the UxmlObjectReferences in template instances
- Shaders created with Shader Graph generate a large amount of shader variants and cause slow build times when using a baked lightmap in BIRP
- Shader Graph Matrix property produces random output values when SRP Batcher is enabled
Resolution Note:
The GC has a lock for allocations leading to contention when allocation heavy code is run in parallel. Reducing the allocations will allow much better scaling for parallel code.