Maya: Limits to the number of colorComposite and floatComposite nodes on GPU?

[stehrani3d]

We have a problem where our material networks can only accommodate so many color or float Composite nodes in GPU mode before stuff stops working as expected (compared to CPU mode). This is a constant issue that we are running into any time we try to build more complex materials. Is there any way these limitations relating to the Composite nodes can be addressed?

Thank you

[hermit.crab]

Could you share some screencaps (the scene itself would be very helpful as well) of the simplest material setup, where this issue is present so we can take a look? You can either attach it here or send it via the contact form. Address it to me in the subject and include a link to this thread in the message.

[stehrani3d]

Hi Aleksandar, thank you for reaching out and apologies for the late reply. I am attaching an example scene to illustrate the issue. Essentially we are using Maya's colorComposite or VRayLayeredTextures nodes to build up complexity by compositing patterns and textures together, pretty standard stuff. The problem appears once you reach a certain threshold of composite operations and the math stops working. The attached render is from CPU mode, in GPU mode it renders as white. Removing one Composite node (or VRayLayeredTex, or whatever other node that does compositing/blending) anywhere in the material network gets you back to expected behavior. Essentially, we need to be able to do more of these operations without the shader breaking.

FloatComposite seems to be broken altogether in GPU mode actually, but that is not a main concern.

[hermit.crab]

Thank you for the provided scene and information. There does seem to be something the matter with the shader network, a lot of variables seem to be connected for the issue's reproduction. It may be some internal GPU limit or a mistake in the math, however, this needs additional testing. I've logged the issue (internal bug-tracker id: VGPU-5277) for developer analysis. As an improvised workaround, the least destructive change that seems to help out is to plug solidFractal1/solidFractal2 directly as the composite's factor or switching the DistanceTex composite's operation to Mix (this may not give an identical result, though).

[stehrani3d]

Thank you Aleksandar. We did notice that some blending operations react differently, but unfortunately the results are not interchangeable most of the time. We look forward to hearing back on this issue.