Reflect on back faces issue that I thought was fixed by brdf change....

[fixeighted]

There still exists this problem; in a glass shader with reflect on back faces checked, when reflection glossiness is anything but 100 then the segments of e.g. a window (thin box) are shown.
If the geometry is curved, then depending on the (arbitrary) amount of segments of the box, this goes away, sort of...moreso it seems lessened.

This was discussed a while ago and the workaround that was given was to swap the default microfacet to Blinn, as Phong is even worse.
Unfortunately, as these images show, none of the available options works.

[kosso_olli]

This is a problem here as well. Headlight covers, windshields etc. Really annoying.

[fixeighted]

It is very weird as I was sure that everyone agreed that switching to Blinn solved it; at least in a previous version.
I will install one and see...

EDIT: Nope, same issue with Next 3.1

[fixeighted]

Well, um...wow...wasn't expecting this result....

Using GPU that particular issue is completely gone, with any option of brdf, regardless of segment amount, curved or flat geometry.

[hermit.crab]

I guess the workaround was scene-specific. The issue is still open (internal bug-tracker id: VMAX-9978 ); I'll notify the developers about the matter. Maybe you could send (or attach) this scene as well (mention me and include a link to this thread)?

[fixeighted]

It's just a box with a dome but anyway, the link in the below still is valid

https://forums.chaosgroup.com/forum/...ext-update-3-2

[^Lele^]

It seems it's Phong not having the issue.

[fixeighted]

Well sort of...scene here https://www.dropbox.com/s/o2cjf4zx7t...20ODD.max?dl=0

The problem is that by using Phong there are such weird refractions and artifacts that it doesn't really seem to fix it, just adds a different weird thing

As you can see from these 2 images, that odd curvature created by using phong, plus the doubled...reflection...refraction, don't know what it is....
The one with the 2 glass planes was at 1 min and takes very much longer to complete without being correct.
The other image has a dark grey material, so better to show up the other strange boxy artifacts that also appear from somewhere (as well as the unwanted secondary reflect/refract).

I still find it odd that gpu renders this all fine, regardless of which I choose...all are as expected.
Now I profess to not knowing the maths behind the different approach to cpu/cpu, but I imagined that refraction would be some sort of constant between them, as well as the brdf model,
so it's all a bit confusing and without a solution under cpu.

[fixeighted]

Here is a BF+BF version of the same, with no boxy stuff going on, leading me to suppose the LC is doing this.
I mean, it is still wrong but at least indicates something afoot

[kosso_olli]

Same problem described here: https://forums.chaosgroup.com/forum/...siness-below-1

[^Lele^]

The devs have been reminded of this.

[fixeighted]

As an update to this topic, I have just discovered by accident that this behaviour does not always occur and seems specifically linked to only local x and z rotations, with y rotation being totally fine
in my example of a thin box window with x amount of segments. It also only occurs in this example when the rotation is 0.09 or above.
Maybe that will help diagnose the problem