Maniphest T101983

Cycles render defect (with GN use)
Closed, Archived
Actions

Assigned To
None
Authored By
Alexey (pseudoua)
Oct 21 2022, 9:40 PM
Tags
  • BF Blender
Subscribers
Alaska (Alaska)
Alexey (pseudoua)
Germano Cavalcante (mano-wii)

Description

System Information
Operating system: Windows-10-10.0.19044-SP0 64 Bits
Graphics card: NVIDIA GeForce RTX 3080 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 522.30

Blender Version
Broken: version: 3.4.0 Alpha, branch: master, commit date: 2022-10-14 22:00, hash: rBbc06fcca47ca
Worked: (newest version of Blender that worked as expected)

Short description of error

Exact steps for others to reproduce the error

Cube.blend6 MBDownload

Event Timeline

Alexey (pseudoua) renamed this task from Cycles render defect to Cycles render defect (with GN use).Oct 21 2022, 9:40 PM
Alexey (pseudoua) created this task.
Germano Cavalcante (mano-wii) changed the task status from Needs Triage to Needs Information from User.Oct 22 2022, 12:03 AM
Germano Cavalcante (mano-wii) added a subscriber: Germano Cavalcante (mano-wii).

Thanks for the report. Unfortunately the scenario described is too time consuming for us to track down, we require the bug reporter to narrow down the problem.

Normally .blend files can be simplified by removing most objects and disabling settings, until the problem reveals itself more clearly.

Alaska (Alaska) added a subscriber: Alaska (Alaska).EditedOct 22 2022, 12:25 AM

I decided to look into the issue. This is what I found:

The issue with the white bit showing up is caused by this link here in your Geometry node setup:

You have the same piece of geometry used in the output twice. One is shade flat and white, while the other is shade smooth and green. And the differences between how they are shaded results in them not perfectly overlapping and one appearing on top of the other at different points.

As for why the result appears different between EEVEE and Cycles, that's due to them using different systems for positioning and precision.

I should note however that the result observed with Cycles changes based on the BVH used. You get the result seen in the bug report if you use the OptiX BVH. And you get this result if you use the BVH2 BVH (used with Metal, CUDA, HIP, oneAPI, and with the right debugging settings, the CPU):

I'm not sure if this is a bug or not.

Alexey (pseudoua) added a comment.Oct 22 2022, 9:07 AM

Thank you! Usually I see overlays immediately in solid mode - artifacts, this time I did not notice them. But I did not notice because the object is closed and transparent.
Yes, this is my oversight.
I'm sorry!

Germano Cavalcante (mano-wii) closed this task as Archived.Oct 22 2022, 9:44 PM
In T101983#1435722, @Alaska (Alaska) wrote:

(...)
I should note however that the result observed with Cycles changes based on the BVH used. You get the result seen in the bug report if you use the OptiX BVH. And you get this result if you use the BVH2 BVH (used with Metal, CUDA, HIP, oneAPI, and with the right debugging settings, the CPU):

I'm not sure if this is a bug or not.

This deserves its own report.
But I wouldn't say that the undefined behavior of the engine when the geometry has very close positions should be considered a bug.
Such overlapping should be avoided in any case.
Perhaps it could be classified as a Known Issue.