Regression: hi-poly mesh with collision in empty scene causes low fps. #84340

New Issue

Vyacheslav Kobozev · 2021-01-03T08:21:36+01:00

2021-01-03 08:21:36 +01:00

System Information
Operating system: Windows-7-6.1.7601-SP1 64 Bits
Graphics card: GeForce GTX 660 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.71

Blender Version
Broken: version: 2.83.9, branch: boneMaster_283, commit date: 2020-11-17 11:33, hash: 6bd7c877fb
Broken till 2.92 today
Worked 2.79b

Short description of error
If I use collision after subdiv (modifier or real) blender hits performance.
Compare with 2.79b:
2021-01-03_10-11-35.mp4

2021-01-03_10-14-32.mp4

**System Information** Operating system: Windows-7-6.1.7601-SP1 64 Bits Graphics card: GeForce GTX 660 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 456.71 **Blender Version** Broken: version: 2.83.9, branch: boneMaster_283, commit date: 2020-11-17 11:33, hash: `6bd7c877fb` Broken till 2.92 today Worked 2.79b **Short description of error** If I use collision after subdiv (modifier or real) blender hits performance. Compare with 2.79b: [2021-01-03_10-11-35.mp4](https://archive.blender.org/developer/F9548512/2021-01-03_10-11-35.mp4) [2021-01-03_10-14-32.mp4](https://archive.blender.org/developer/F9548511/2021-01-03_10-14-32.mp4)

Vyacheslav Kobozev commented

2021-01-03 08:21:36 +01:00

Added subscriber: @Vyach

Germano Cavalcante commented

2021-01-06 15:43:41 +01:00

Added subscriber: @mano-wii

Germano Cavalcante commented

2021-01-06 15:43:41 +01:00

Changed status from 'Needs Triage' to: 'Confirmed'

Germano Cavalcante commented

2021-01-06 15:43:41 +01:00

I can confirm.
If I'm not mistaken, a new system for the dependency graph was implemented in 2.80.
Perhaps the collision modifier now has a dependency on the object's position and therefore the entire stack of modifiers is recalculated when it is moved.

But since at 2.79 this was not necessary, it does appear to be a regression.

I can confirm. If I'm not mistaken, a new system for the dependency graph was implemented in 2.80. Perhaps the collision modifier now has a dependency on the object's position and therefore the entire stack of modifiers is recalculated when it is moved. But since at 2.79 this was not necessary, it does appear to be a regression.

Jacques Lucke commented

2021-01-07 14:09:55 +01:00

Added subscriber: @JacquesLucke

Jacques Lucke commented

2021-01-07 14:09:55 +01:00

Changed status from 'Confirmed' to: 'Archived'

Jacques Lucke closed this issue

2021-01-07 14:09:55 +01:00

Jacques Lucke commented

2021-01-07 14:09:55 +01:00

The Transform -> Collision Modifier dependency was added in c384b5ddbd. This seems correct, because the modifier is accessing ob->obmat. The collision modifier builds a bvh tree in world space. This is not ideal in this example file, but that is how the code is designed to work currently.

The performance can be improved slightly by moving the collision modifier before the subdivision modifier. This way the original mesh is used as collision shape. However, the subdivision modifier is stuff evaluated every time the position changes, so you still get quite a performance hit from that.

For the time being, the most practical solution might be to have two separate objects. One collision object that you parent to the subdivided object. The collision object can be hidden for rendering. This way, the collision modifier is evaluated on low res geometry when the object is moved. Since the subdivision modifier does not depend on the object transforms, it is not reevaluated every time you move the object.

I don't consider this to be a bug, because the commit that causes the performance regression fixes a bug. Thanks for the report anyway!

The `Transform -> Collision Modifier` dependency was added in c384b5ddbd. This seems correct, because the modifier is accessing `ob->obmat`. The collision modifier builds a bvh tree in world space. This is not ideal in this example file, but that is how the code is designed to work currently. The performance can be improved slightly by moving the collision modifier before the subdivision modifier. This way the original mesh is used as collision shape. However, the subdivision modifier is stuff evaluated every time the position changes, so you still get quite a performance hit from that. For the time being, the most practical solution might be to have two separate objects. One collision object that you parent to the subdivided object. The collision object can be hidden for rendering. This way, the collision modifier is evaluated on low res geometry when the object is moved. Since the subdivision modifier does not depend on the object transforms, it is not reevaluated every time you move the object. I don't consider this to be a bug, because the commit that causes the performance regression fixes a bug. Thanks for the report anyway!

Vyacheslav Kobozev commented

2021-01-08 11:41:57 +01:00

@JacquesLucke
Okay, I see, so it is the same global issue with unwanted recalculations.

Thanks for tip/workaround.

@JacquesLucke Okay, I see, so it is the same global issue with unwanted recalculations. Thanks for tip/workaround.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Regression: hi-poly mesh with collision in empty scene causes low fps. #84340