XPU has obviouse differences to CPU render in solaris

   1429   4   1
User Avatar
Member
39 posts
Joined: June 2017
Offline
I used to put my trust to new rising star “solaris”. But after few project, I cannot achieve "clean & vivid" as other render engine does.
I compre it with XPU and CPU render on a semi-open scene.
There are four overlap dome lights and four area lights to mimic the shopping mall lighting env.
Dozen of spotlights with IES instenced in /stage for interior lighting.

These are the compare of XPU and CPU.
As the screenshot indicate:
1. XPU looks aokay in shadow but the hightlight / specular is over expose.
2. UV texture is wrong in CPU(flipped and squeezed)
3. XPU is more blurred
4. Glass on XPU is missing.

Houdini ver. 20.5.390-Py3.11
Image Not Found
Edited by digitalwu - Oct. 22, 2024 06:45:20

Attachments:
region_01_cpu.PNG (71.3 KB)
region_01_XPU.PNG (66.5 KB)
region_02_cpu.PNG (130.7 KB)
region_02_XPU.PNG (128.8 KB)
region_03_cpu.PNG (55.7 KB)
region_03_XPU.PNG (50.5 KB)
region_04_cpu.PNG (68.4 KB)
region_04_XPU.PNG (54.1 KB)
region_05_cpu.PNG (137.2 KB)
region_05_XPU.PNG (142.0 KB)
region_06_cpu.PNG (81.8 KB)
region_06_XPU.PNG (76.8 KB)
community_H205.hiplc (5.1 MB)

User Avatar
Staff
587 posts
Joined: May 2019
Offline
I'm guessing the main differences are caused by a color-limit issue, and also normal-mapping differences.
FYI, we have fixed some normal mapping issues internally (to be released in the next major version of Houdini).

In any case, it's difficult to know for sure without a repro case that we can analyze on our end. Is it possible you could post a scene to demonstrate these issues?

thanks
User Avatar
Member
8177 posts
Joined: Sept. 2011
Offline
brians
I'm guessing the main differences are caused by a color-limit issue, and also normal-mapping differences.
Do XPU and CPU have known different behavior when multiple domelights are present? (as described above)
User Avatar
Member
39 posts
Joined: June 2017
Offline
scene file is uploaded. the key shader is Greyscalegorilla's "Retroreflector 10". I modified it to be cyan color by using ColorCorrect dim the satuation , Constant(color) and Multiply nodes.

Attachments:
111111.png (810.8 KB)

User Avatar
Staff
476 posts
Joined: Sept. 2016
Offline
Hi. Part of the issue seems to be that you're applying normal maps on objects that doesn't have any UVs.
Tangent-space normal map can only be intepreted correctly if the tangent frame matches (or oriented the same way at the very least) that of which was used to create the map itself. Without UVs there are no UV tangents, so karma falls back to primtive tangents which is going to be arbitrarily oriented because they're at the whim of tessellation (and/or accelerator).
  • Quick Links