Shading issue with Vray GPU / Normal map creates weird results

[alex_jantsch]

I often get some weird shading issues when using Vray GPU. I'm creating a basic flakes layer onto my geometry with a flakes normal map that is put into the bump slot. There is some crazy stretching going on in the mesh and the flakes only appear at certain areas. Does anyone here know kow to solve this issue? The normal map is using gamma 1 with no filtering.

[alex_jantsch]

[^Lele^]

What does it look with a checkerboard texture in the diffuse?
I ask as it looks like UV stretching, of course barring a GPU bug.

[alex_jantsch]

Oh that looks horrible, but how can I fix this? What do you mean with GPU bug?

[^Lele^]

Pipe the normal map through a triplanar, or use the stochastic flakes in a blend with mapping set to object XYZ (fairly more expensive to render, though, if likely much prettier.).
From your OP i had assumed the stretching was *due* to the GPU engine, hence the mentioning of a possible bug.

[alex_jantsch]

Using a triplanar doesn't solve the issue, the left hood with the flakes above with the checker map on the right hood uses a triplanar node, using stochastic flakes isn't an option as well. The weird thing is that it's not only happening with the imported car model, it is basically happening with every imported geometry like megascans for instance. If I render the same geometry with corona render or with fstorm everything works perfectly. This post here from adan martin basically describes the excact same issue that I am facing: https://forums.chaos.com/forum/v-ray...-5#post1124665 .Basically right now you can`t use normal maps for Vray GPU which is pretty crazy if you ask me.

[^Lele^]

This post here from adan martin basically describes the excact same issue that I am facing:

Not at all, they are two different things: there is no sign of stretching in that case (as it's got to do with the space the normals are calculated in, for cpu and gpu under max).
In yours, it looks much as the UVs are calculated wrong, for whatever reason: if you could provide a file and the texture for investigation, it'd be good for us.

EDIT: attached, a substance painter starter model, with normal mapping rendered under GPU (beauty + bumpNormals). CUDA or RTX makes no difference to me.

[alex_jantsch]

I uploaded the scene now with the following link: https://we.tl/t-cg0O0nvueJ
I tried to apply a vray scanned material to see if it works but it tells me that there is some bad mapping with the model as well.

[fixeighted]

I remade your material and it works now. It is different in that the triplanar is used as the mapping source for all axis, not just the one that your material uses, which is probably the problem.
Here is the scene to compare the two

[alex_jantsch]

Interesting solution! But why are you not using a VrayNormalMap to pipe between the bump slot and the texture? If I do so, the problem occurs again. I just don't get why every geometry I import into 3ds max now has bad mapping?

[fixeighted]

Oops, sorry about that. I updated it and tested again to find some odd things and a solution.

Yes the object has messed up mapping, which shouldn't matter when using triplanar.
It works fine with cpu, but with gpu I had to apply a UVWMapping Clear to get it to not distort.
Even then it does not work, so gpu seems to demand some uv mapping as default, in order to work.
So the solution is to apply some 1/1/1 box mapping just to satisfy that demand, which can be collapsed.

[^Lele^]

What he said.

[alex_jantsch]

So it is a bug with GPU?

[^Lele^]

Bug, or limitation, the devs will know (VGPU-5673).
For now, you should fix the UVs as fixeighted mentioned (or simply with a UVW Clear mod): without willing to speak for the devs, given the case and easy workaround this isn't likely to shoot up their priority list (i could be wrong, mind you.).

[alex_jantsch]

Alright, thanks a bunch guys!