NAN pixels only with GPU

[fixeighted]

I have this scene that I just noticed is creating nan pixels on a matted ground object, making it unusable.
It is perfect when using cpu.

I have 18 shots to do on only this one scene, with plenty more to follow, so I need this solved asap. Naturally gpu is quicker.

Question is, before I start trying to solve this over potentially days through support, is there anything I can
do to troubleshoot this more immediately myself...common reasons it may occur etc.?

Thanks

[vladimir_krastev]

Hi, the issue is reproducing in a simple scene.
NAN pixels appear in the Alpha channel when using VRayWrapperMtl with refractive VRayMtl (refraction colour also works, no opacity) and VRay2SidedMtl with completely black Translucency colour (0,0,0).
The issue is reproducing in CUDA and not reproducing in RTX and V-Ray CPU.
I logged it in our bug tracking system.

[vladimir_krastev]

Hi, thanks for posting. Cоuld you send us your scene here or using the support contact form. If you use the contact form please include a link to this tread.
As any issue you have to break it to smaller parts an to try and find out what is causing it. Could be a material, a light, etc. or it can be a combination. I am not 100% sure about these NAN pixels, are they be completely white pixel? Maybe you are dealing with fire flies?

[fixeighted]

ok thanks....I'll send the scene to support.

I've actually never seen this happen in any scene before now.
It's extremely simple as you'll see, so my thinking is that it's a gpu issue, as cpu causes no problems.

[fixeighted]

Update: I have discovered the fault and it is a Vray 2-sided material, specifically a bird's wing feathers with opacity but no translucency.

The docs say only ''(without Multiply By Front diffuse)'', though even without this it does not work, so
shouldn't that be 'not supported' at all?, or is there some way this works that I am missing?
If it is somewhat supported then in what situation does it work?

[vladimir_krastev]

Hi, opacity is supported only in clip mode on the GPU. What version of V-Ray are you using? In the latest V-Ray 6 the options that are not supported are greyed out.



Here is a test render from my simple scene but everything seems normal. I have VRay2SidedMtl with opacity and translucency.

[fixeighted]

In my scene, the nan pixels show on a matte ground object.
The feathers shader is a 2-sided with a map for each side, no translucency
and a b/w opacity map.

I'm on Vray version just before the latest a few days ago, so the options are the same as yours.

[fixeighted]

Here is a simplified scene using a stripped material with the op map https://www.dropbox.com/s/r9q3wooj1u...rs_op.png?dl=0.

[vladimir_krastev]

Hi, thanks for the scene but in the archive I do not have any assets? In the current configuration the scene renders a black image, which is expect as I noticed you are using VRayMtlWrapper for the ground plane. The plane with the feather material is missing maps and is not rendering.

[fixeighted]

I provided a link to an opacity map, which is all that is needed. It's just a shadow catcher and the plane with opacity and the lights.

Oops, sorry, my mistake, here are the other 2 materials, although anything would have done. https://www.dropbox.com/s/k3usdbid9m...ROBIN.rar?dl=0

[vladimir_krastev]

Hi, the issue is reproducing in a simple scene.
NAN pixels appear in the Alpha channel when using VRayWrapperMtl with refractive VRayMtl (refraction colour also works, no opacity) and VRay2SidedMtl with completely black Translucency colour (0,0,0).
The issue is reproducing in CUDA and not reproducing in RTX and V-Ray CPU.
I logged it in our bug tracking system.

[fixeighted]

It occurs without using a wrapper also, just fyi...I just tried the wrapper to see if it made any difference.
Also, with my 2080ti switching to rtx doesn't make any difference.

[vladimir_krastev]

Not reproducing without VRayWrapperMtl and not in RTX mode on my RTX4070ti.

[fixeighted]

Weird stuff indeed...I'm on driver 516.94 so I'll change it to the newest recommended and see what happens.
The reason I had this driver is that Vantage required it, so there's always a mismatch between that and the 'recommended' for gpu, or at least
this was the case late last year.

[fixeighted]

Ok, so updating the driver makes the rtx approach work, which I can use as a backup, as long as the horribly long compilation time doesn't interfere.
Oddly the newer driver also significantly reduces the error under CUDA, if that's useful.
I was seeing only 2 or three nan pixels so maybe that's what is happening with the wrapper for you...they are hard to spot