Maniphest T48538

Render Fragments on CPU with Cycles
Closed, Resolved
Actions

Assigned To
Thomas Dinges (dingto)
Authored By
R. E. (rehr)
May 28 2016, 5:01 PM
Tags
  • BF Blender
  • Cycles
Subscribers
R. E. (rehr)
Sergey Sharybin (sergey)
Thomas Dinges (dingto)

Description

System Information
Tested on: Windows 7 64bit and Windows 8 64bit with i3, i5, i7, different NVIDIA GPUs

Blender Version
Broken: 2.76, 2.77

Short description of error
Rendering on CPU in Cycles causes "Bounding Box" to appear on some frames of animation.

Issue happens on some planes which follow a curve. No real pattern on which it happens visible to me (could be intersection/distance/scale). Only happens on CPUrenders, no visible issues on GPU. Also not visible on Dynamic Viewport.

Exact steps for others to reproduce the error
In Blendfile it starts appearing on Frame 200

  • Set to CPU rendering and either render or set viewport on static BVH

-disappears with GPU renders or dynamic BVH

Related Objects

Mentioned In
T48943: Unexpected black frames or frame regions when using holodut materials
Mentioned Here
rB438bdccff34b: Buildobt: Update master config
rBd733826708f9: Fix T48614: Blender from buildbot crash when Separate selection in this…

Event Timeline

R. E. (rehr) created this task.May 28 2016, 5:01 PM
R. E. (rehr) raised the priority of this task from to 90.
R. E. (rehr) updated the task description.
R. E. (rehr) added projects: BF Blender, Cycles.
R. E. (rehr) edited a custom field.
R. E. (rehr) added a subscriber: R. E. (rehr).
R. E. (rehr) added a comment.EditedMay 28 2016, 5:07 PM

Fails to add Blendfile to original post...

cpubug.blend4 MBDownload

*Update* Reproduced issue on 2.74 version. Also gave it a try with the simple Defaultdiffuseshader - issue stays (see image). Only seems to happen within the Animation, final frames of the Render are always fine on all tests I made so far.

Sergey Sharybin (sergey) lowered the priority of this task from 90 to 30.May 31 2016, 11:49 AM
Sergey Sharybin (sergey) added a subscriber: Sergey Sharybin (sergey).

Final renders always use Static BVH, you can't change that.

What are the exact steps reproducing the error?

R. E. (rehr) added a comment.May 31 2016, 11:21 PM

Sorry for bad phrasing. I know it only renders static. So the issue with the given file can be reproduced either by rendering frames from 200 onwards on the CPU or by setting the viewport option to static BVH. It is NOT appearing on a dynamic viewport or a GPU render.

Sergey Sharybin (sergey) added a comment.Jun 7 2016, 1:47 PM

So i've downloaded cpubug.blend a,d started viewport render and did not see any artifacts on neither of the frames.

Some more questions:

  • When you're talking about 2.77, is it 2.77a corrective release? If not, it totally worth checking 2.77a.
  • Does it happen with latest builds from builder.blender.org?
  • Does scaling objects up help the issue?
R. E. (rehr) added a comment.Jun 7 2016, 3:02 PM

Latest build is crashing if I try Renderpreiview. I'm giving it a shot with Ubuntu Studio right now using the official 2.77a download from blender.org. The issue there happens under the 64 bit release, with 32 bit everything appears to be fine so far. I'll check it under Windows again and update the results in respective to the version later.

Renderpreview Frame 212 in Viewport with Static BVH 2.77a 64bit:


Same Frame and settings in the 32 Bit Version:

Sergey Sharybin (sergey) added a comment.Jun 7 2016, 3:22 PM

Can you try following:

  • Set Debug Value to 256 (Ctrl-Alt-D)
  • Go to Render Buttons -> Debug and see if toggling SSE/AVX flags makes any difference? (you'll have to re-run viewport render (stop viewport render, start viewport render) to make it recognize new settings).
R. E. (rehr) added a comment.Jun 7 2016, 4:09 PM

Yes, AVX-2 Settings make it work.

Sergey Sharybin (sergey) raised the priority of this task from 30 to Normal.Jun 7 2016, 4:46 PM

Ok, that explains it. Currently was testing on non-AVX2 machine. Will try to find one for debugging.

Thomas Dinges (dingto) added a subscriber: Thomas Dinges (dingto).Jun 8 2016, 12:30 PM

I cannot see issues with latest master build on Linux with Intel Skylake CPU (AVX2).

Sergey Sharybin (sergey) added a comment.Jun 8 2016, 2:14 PM

@R. E. (rehr), there are new builds on builder.blender.org which should fix the crash and have all the latest fixes which might affect on this bug. Please give it a try again :)

R. E. (rehr) added a comment.Jun 9 2016, 2:37 PM

Tested on 2 Skylakes and it works on them, but test on Broadwell still throws Access violation and crashes.

Sergey Sharybin (sergey) added a comment.Jun 9 2016, 2:40 PM

@R. E. (rehr), Can you describe a bit more what "works" means? ;) Doesn't crash and render result is correct, or renders without crash but still with box artifacts?

Can you submit another bug report about crash with .blend file which crashes and instructions how to crash?

R. E. (rehr) added a comment.Jun 9 2016, 9:55 PM

No issues with todays build d733826-win64, no crashes, no artifacts on any of my available comps, Broadwell included. 438bdcc crashed when triggering Renderpreview on both Linux and Windows.

Thomas Dinges (dingto) changed the task status from Unknown Status to Resolved.Jun 9 2016, 9:57 PM
Thomas Dinges (dingto) claimed this task.

Nice, so it's solved! Thanks for testing again :)

R. E. (rehr) added a comment.Jun 9 2016, 10:08 PM

Thanks for fixing :)

Brecht Van Lommel (brecht) mentioned this in T48943: Unexpected black frames or frame regions when using holodut materials.Jul 27 2016, 12:42 AM