Odd behavior with Karma CPU
3178 12 2
-
- Steven Quinones-Colon2
- Member
- 22 posts
- Joined: 6月 2014
- Offline
I have an odd situation. I have a large USD file that I'm splitting into assets and doing lookdev as I go. When I render in XPU everything looks as I would expect it, but with CPU I get odd behavior, images below ,the first image is the CPU render. I'm on Houdini 19.5.368 on an M1 mac. The setup is really simple, any ideas?
Edited by Steven Quinones-Colon2 - 2023年6月21日 13:05:49
-
- mandrake0
- Member
- 644 posts
- Joined: 6月 2006
- Offline
-
- Steven Quinones-Colon2
- Member
- 22 posts
- Joined: 6月 2014
- Offline
-
- Steven Quinones-Colon2
- Member
- 22 posts
- Joined: 6月 2014
- Offline
-
- jsmack
- Member
- 8046 posts
- Joined: 9月 2011
- Offline
possibly a problem with the normal map. XPU and CPU behave differently with bogus normals. Many USD scenes I've encountered have all kinds of non-compliant and out of date problems with how they author materials, especially normal maps. Do they look the same if you disconnect the normal map from the material?
-
- Steven Quinones-Colon2
- Member
- 22 posts
- Joined: 6月 2014
- Offline
-
- mandrake0
- Member
- 644 posts
- Joined: 6月 2006
- Offline
Steven Quinones-Colon2
Or maybe is that XPU is ignoring bad geometry and CPU is rendering it properly.
I would say maybe the Shader has functions / textures inside that XPU ignores and Karma CPU is using it. Or the Mesh has Attributes inside that gets wrong used.
What you can do:
Check the Mesh in Spreadsheet if there is something a Special Attribute or Normals(N) in Point or in Vertices you also can import in /obj and check there.
You can add for Testing a SOP Normal (Vertex or Point) and check if it is Better.
change the shader to a MTLX Standard Surface does it look same.
use a Default Mesh (Box, Sphere) and check if the shader works on both Karma(CPU, XPU)
I don't have a Apple M1 and i also don't know how XPU is Working on a M1 CPU/GPU.
When you find it could be a bug then it's required to report to Sidefx:
https://www.sidefx.com/bugs/submit/ [www.sidefx.com]
-
- Andy_23
- Member
- 918 posts
- Joined: 3月 2014
- Offline
I may add to this and am too having unexpected behavior with Normals using MaterialX standard surface.
I, too, find, that XPU gives me the expected results, while CPU there are obvious artifacts.
This happens with even the most simple geometry like grids.
Any help is much appreciated.
Running 19.5.640 — Apple Silicon M2 or Win 10
I, too, find, that XPU gives me the expected results, while CPU there are obvious artifacts.
This happens with even the most simple geometry like grids.
Any help is much appreciated.
Running 19.5.640 — Apple Silicon M2 or Win 10
Edited by Andy_23 - 2023年6月23日 03:58:17
-
- Andy_23
- Member
- 918 posts
- Joined: 3月 2014
- Offline
-
- dlee
- スタッフ
- 444 posts
- Joined: 9月 2016
- Offline
Andy_23
I may add to this and am too having unexpected behavior with Normals using MaterialX standard surface.
I, too, find, that XPU gives me the expected results, while CPU there are obvious artifacts.
This happens with even the most simple geometry like grids.
Any help is much appreciated.
Running 19.5.640 — Apple Silicon M2 or Win 10
Can you please file a bug with the scene? Thank you!
-
- Siavash Tehrani
- Member
- 729 posts
- Joined: 7月 2005
- Offline
We are seeing similar issues where Karma CPU has difficulties with normal "maps" generated from procedural patterns. Please see the example below: CAD file, normals driven by MaterialX procedural noises. Karma CPU has artifacts that line up with the model's topology, whereas XPU handles it fine. I will submit an official bug report as well.
The color difference is a separate problem that cropped up with H20, but I'll have to dig a little deeper on that.
The color difference is a separate problem that cropped up with H20, but I'll have to dig a little deeper on that.
-
- jsmack
- Member
- 8046 posts
- Joined: 9月 2011
- Offline
Siavash Tehrani
We are seeing similar issues where Karma CPU has difficulties with normal "maps" generated from procedural patterns. Please see the example below: CAD file, normals driven by MaterialX procedural noises. Karma CPU has artifacts that line up with the model's topology, whereas XPU handles it fine. I will submit an official bug report as well.
The color difference is a separate problem that cropped up with H20, but I'll have to dig a little deeper on that.
I wonder if XPU is using MikkT tangents and CPU is using ray derivatives as tangents--which will always be ugly for a triangle mesh
-
- Siavash Tehrani
- Member
- 729 posts
- Joined: 7月 2005
- Offline
-
- Quick Links