Cycles - Render Pass contains black pixels with no values #90423

New Issue

Marcin Różewski · 2021-08-04T10:43:47+02:00

Marcin Różewski commented

2021-08-04 10:43:47 +02:00

System Information
Windows 10 Pro Build 19043.1110
GeForce GTX 1080
GeForce GTX 1080
NVIDIA-SMI 461.92
Driver Version: 461.92
CUDA Version: 11.2

Broken Blender Version: 3.0.0 Alpha -> 3.0.15 bpy.app.version_file
Broken Blender Version: 3.0.0 Alpha -> 3.0.7 bpy.app.version_file
Worked Blender Version: 2.93.1 -> 2.93.20 bpy.app.version_file

Getting NANs since 9c6a382f95.

Short description of error
In the almost all pass there are black pixel with no value. In compositing this pixel make artifacts.
The Pass is ok in Blender 2.93.1 with the same geometry.

This is a good render pass AO

This is a broken render pass AO

All other pass have the same problem.

**System Information** Windows 10 Pro Build 19043.1110 GeForce GTX 1080 GeForce GTX 1080 NVIDIA-SMI 461.92 Driver Version: 461.92 CUDA Version: 11.2 **Broken** Blender Version: 3.0.0 Alpha -> 3.0.15 `bpy.app.version_file` **Broken** Blender Version: 3.0.0 Alpha -> 3.0.7 `bpy.app.version_file` **Worked** Blender Version: 2.93.1 -> 2.93.20 `bpy.app.version_file` Getting NANs since 9c6a382f95. **Short description of error** In the almost all pass there are black pixel with no value. In compositing this pixel make artifacts. The Pass is ok in Blender 2.93.1 with the same geometry. This is a good render pass AO ![image.png](https://archive.blender.org/developer/F10264555/image.png) This is a broken render pass AO ![image.png](https://archive.blender.org/developer/F10264563/image.png) All other pass have the same problem.

Marcin Różewski commented

2021-08-04 10:43:47 +02:00

Added subscriber: @mdsanima

blender-admin commented

2021-08-04 10:43:47 +02:00

#90861 was marked as duplicate of this issue

Philipp Oeser commented

2021-08-04 10:54:53 +02:00

Added subscriber: @lichtwerk

Philipp Oeser commented

2021-08-04 10:54:53 +02:00

Changed status from 'Needs Triage' to: 'Needs User Info'

Philipp Oeser commented

2021-08-04 10:54:53 +02:00

Could you share an example file where this happens?

Marcin Różewski commented

2021-08-04 12:44:30 +02:00

This is a file prepare file thats give me a error on 3.0.0
render_pass_error.blend

In the 2.93.1 version is ok.

But now I noticed that the material is badly connected, the normal map was accidentally connected to the alpha slot and the opacity to the emission.
After switching the slots to the correct ones, the rendering no longer shows errors. But if the same file with the wrong connection is made, version 2.93.1 renders without error.

I hope it will help and it is nothing serious. Good luck.

This is a file prepare file thats give me a error on 3.0.0 [render_pass_error.blend](https://archive.blender.org/developer/F10264905/render_pass_error.blend) In the 2.93.1 version is ok. But now I noticed that the material is badly connected, the normal map was accidentally connected to the alpha slot and the opacity to the emission. After switching the slots to the correct ones, the rendering no longer shows errors. But if the same file with the wrong connection is made, version 2.93.1 renders without error. I hope it will help and it is nothing serious. Good luck.

Marcin Różewski commented

2021-08-05 01:30:55 +02:00

Changed status from 'Needs User Info' to: 'Needs Developer To Reproduce'

Marcin Różewski commented

2021-08-05 01:30:55 +02:00

Please let me know if you want more information.

Philipp Oeser commented

2021-08-05 05:32:21 +02:00

Changed status from 'Needs Developer To Reproduce' to: 'Needs Triage'

Pratik Borhade commented

2021-08-05 05:37:50 +02:00

Added subscriber: @PratikPB2123

Philipp Oeser commented

2021-08-05 10:33:00 +02:00

Bisecting note:

This already has NANs in b73dc36859
7223a0348f crashes
No NANs in 6afafc46f6

Bisecting note: - This already has NANs in b73dc36859 - 7223a0348f crashes - No NANs in 6afafc46f6

Philipp Oeser commented

2021-08-05 13:38:36 +02:00

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

Philipp Oeser commented

2021-08-05 13:38:36 +02:00

Added subscribers: @ktdfly, @brecht

Philipp Oeser commented

2021-08-05 13:38:36 +02:00

Freeze / crash was due to 5f9677fe0c, was fixed in 2dbb492268

So getting NANs since 9c6a382f95.
Strange thing: no matter what I set the Shadow Terminator Geometry Offset to (0.0 should be disabled), those NANs wills always be there.

Think rendering devs would need to look at this (but dont think this is one of the limitations mentioned in the commit?), will confirm for now.

CC @ktdfly
CC @brecht

Freeze / crash was due to 5f9677fe0c, was fixed in 2dbb492268 So getting NANs since 9c6a382f95. Strange thing: no matter what I set the Shadow Terminator Geometry Offset to (0.0 should be disabled), those NANs wills always be there. Think rendering devs would need to look at this (but dont think this is one of the limitations mentioned in the commit?), will confirm for now. CC @ktdfly CC @brecht

Mikhail Matrosov commented

2021-08-09 13:44:18 +02:00

I've refined the example. I think the problem is somehow connected to Alpha input of Principled BSDF and how it mixes in Diffuse component.

render_pass_error_refined.blend

I've refined the example. I think the problem is somehow connected to Alpha input of Principled BSDF and how it mixes in Diffuse component. |![Screenshot from 2021-08-09 14-38-48.png](https://archive.blender.org/developer/F10275118/Screenshot_from_2021-08-09_14-38-48.png)|![Screenshot from 2021-08-09 14-38-35.png](https://archive.blender.org/developer/F10275120/Screenshot_from_2021-08-09_14-38-35.png)| | -- | -- | [render_pass_error_refined.blend](https://archive.blender.org/developer/F10275122/render_pass_error_refined.blend)

Brecht Van Lommel commented

2021-08-19 19:57:32 +02:00

@ktdfly, if the geometry offset commit introduced this, can you figure out how to avoid that?

There may be NaNs or invalid values generated earlier, but I think we should both fix that and make the geometry offset robust to such things.

@ktdfly, if the geometry offset commit introduced this, can you figure out how to avoid that? There may be NaNs or invalid values generated earlier, but I think we should both fix that and make the geometry offset robust to such things.

Mikhail Matrosov commented

2021-08-19 20:33:57 +02:00

I'll do my best @brecht