Why is Vray 4 so noisy at Vray for Rhino?

[Micha]

Hi,

maybe some experts know what happens at my scenes, for example here a part of train interior - I get a lot of grainy noise. I used IM+LC and the noise threshold is 0.02. Both images based on CPU usage and the render time is approx. 30min for 3600x2400. One light is used (directional light) and a basic GI environment. (I tested it - an adaptive light dome would cause extra noise only.)

The V-Ray 3 version based on a setup with local subdivs, the V-Ray 4 image based on V-Ray for Rhino without any subdiv control anymore, since subdiv controls are not supported anymore. If I would use a stronger noise threshold at V-Ray 4, than the render times would be longer than of V-Ray 3. OK, I could use the denoiser, but this would be like to use the denoiser not for speedup, but rather as workaround.

I have the feeling I don't use V-Ray 4 in the right way or something behind in the plugin avoid to use the full potential. Maybe there is something needed, what is available at 3dsmax, but forgotten at Rhino. If I would use LC+BF the noise would be much stronger or the render time approx. 1h30min.

Maybe lele , vlado or an other expert know how to get the noise down at V-Ray 3 level before to enable the denoiser.

Best -
Micha

Comparison image are upscaled at 200% to better show the noise pattern.

[kosso_olli]

Judging by the jagged edges it seems that there aren't enough AA samples. I can not see the Max AA parameter in the screenshot, what is it set at?
Apart from that, the GI solution is full of artifacts, in both versions.

[Micha]

At VfR 4 are not more options, this is all.

There are some splotches by the low IM at -2, but most I solve this per some postwork during the next rendering is running. So, I can render a lot of images within tight deadlines.

[Micha]

I did a lot more of tests and it looks like I the raw result of V-Ray 4 can't get faster than by V-Ray 3. It looks like the new automatic subdiv doesn't work so good like the old manual setup and the speed advantages of V-Ray 4 are mostly lost. Only the denoiser helps to keep the speed or to get a faster speed. I'm curious how fast would be V-Ray 4 with the old manual control. I hope the automatic sampling can be optimized to get the same basic raw quality in the future, without the grain effect. Or is the grain at my example caused by a bug? (I posted an example scene at the VfR beta forum where I got a strong grain by a bug.)

So far the denoiser could be the solution, but that it doesn't work with glass or simple transparency is a problem at some scenes. Attached an example, an interior with a lot of glass room deviders (the glass sheets are faked glass material based on colored transparency to keep the max refl depth at 1, only the edges are refractive glass).

The impression from my tests are: LC+BF on the GPU create a good denoised quality, so I would need to invest in more GPU power. At the GPU road I strongly miss:
* missing GI AO to accentuate geometric details (not so much for photorealism, more for presentations and discussion the design).
* no denoised glass
* no GI cache (IM) for quick rerender the scene for creating design versions based on changing some colors/textures only
* no material id trough glass

[^Lele^]

I can't speak for V-Ray for Rhino, but the automatic sampling that appeared in V-Ray for max 3.3 *is* working as intended, provided defaults are respected (a couple of caveats for heavy dof/moblur and animation.).
Tampering with settings will most certainly reduce the speed more often than not (if you search on the forums you'll find numbers for all i'm talking about.), and is officially discouraged.
If you have a scene which misbehaves under default render settings, feel free to send it to support as a potential bug, rather than spend time tweaking.
The IRMap is also an avenue officially discouraged, as its development stopped with the 3.x branch (that, and that of a few other old, interpolated tech), to make room for a number of modern optimisations, and take away the guesswork inherent in the approach.
Again, search the forums to find ample debates about this.
You should keep the V-Ray settings at default, including GI, and you'll see all those issues will go away.

[Micha]

The setup was reset to default and my hope was that the noise image could tell something about the underlaying process which cause that V-Ray 4 creates more noise than V-Ray 3 and how to get more speed.

Thank you ^Lele^ for the hints and insights.

[^Lele^]

If you used default settings, then send the scene to support for analysis, as it may be a bug.
They'll know best.