SSS Issue in GPU mode (Randomwalk V2)

SSS Issue in GPU mode (Randomwalk V2)

jacksonboss
Participant Participant
1,682 Views
4 Replies
Message 1 of 5

SSS Issue in GPU mode (Randomwalk V2)

jacksonboss
Participant
Participant

I noticed this issue quite a while ago, and although it didn’t bother me too much at first, I’d still like to report it.

 

I do a lot of work using SSS with small objects, and I’ve noticed a bug/problem when using the GPU for this.

 

In my latest project involving a protein bar, I noticed that the broken part had the SSS Scale set to 0.08, yet it still looked very strong, almost like a wax material. Even when setting the Scale to 0, there was still a minimum SSS effect on the bar.

 

I've been aware of this for over a year, but I wasn’t too concerned, thinking it would be resolved in a future update. I usually worked around it by scaling up my objects up to 10 times their size to avoid the issue, but in some cases, I can't do that.

 

Screenshot_68.png

 

The SSS in CPU mode is perfect, but obviously much slower. I did a small test with a cube, as you can see below. Please take a look and draw your own conclusions (I’m not sure why the displacement differs when switching from CPU to GPU mode—maybe that’s a bug as well).

 

I forgot to clarify that this only happens on small surfaces. I usually resolve this issue by excessively scaling up my object, but as I mentioned, there are some cases where I can’t do that.

Cube-Test.png

I always use Randomwalk V2 for SSS

0 Likes
Accepted solutions (1)
1,683 Views
4 Replies
Replies (4)
Message 2 of 5

lee_griggs
Autodesk
Autodesk

Thanks very much for reporting this issue. Would you be so kind as to upload a small scene that replicates the issue and demonstrates it? I would be interested to know if you have the same SSS issue with openpbr_surface.

 

Incidentally, are you using transmission_scatter for the caramel(?) center? There is a honey example in the openpbr_surface docs here.

 

 

 

Lee Griggs
Arnold rendering specialist
AUTODESK
0 Likes
Message 3 of 5

thiago.ize
Autodesk
Autodesk
Accepted solution

Hi, no need for a scene. This is a known GPU numerical precision bug already being tracked internally with ticket ARNOLD-11163.

0 Likes
Message 4 of 5

jacksonboss
Participant
Participant

@lee_griggs Thank you very much for responding, the caramel is just transmission and is still in the development phase, I was referring to the broken part of the chocolate.

 

@thiago.ize  I'm glad to know it's a known bug, and I hope it's resolved as soon as possible. I love using Arnold GPU for its speed and wouldn't want to have to use the CPU for this.

 

Thank you both for the quick response, and have a great weekend!

Message 5 of 5

jamie.portsmouth
Autodesk
Autodesk

Hi jacksonboss

 

Can you clarify what the length units are here? So the scale of 0.02 corresponds to 2cm (i.e. you're working in meters), or 0.02cm (i.e. you're working in cm)?

0 Likes