V-Ray 5 GPU MAX 2021 - DoF/Bokeh noise

[Dman3d]

I am working on a scene with V-Ray 5 GPU for MAX2021 and cannot seem to eliminate noise in the DoF and Bokeh areas of the shot,
I have tried 0 (zero) noise threshold and increasing the sample limits x2 and x5. Using light cache or BF. Not too sure what to do next.

This is a lowres IPR render but represents the same areas having noise in GPU final render settings.

[kosso_olli]

You need to increase your Max AA considerably. We had shots that needed a max AA of 256 or even 512. Render times will increase quite a bit, but it's the only solution to get these clean.
Here is a thread, a result of raising the AA can be seen in post #26:
https://forums.chaosgroup.com/forum/...h-shapes/page2

[fixeighted]

Interestingly a guy was plugging his re-launch of DOF Pro yesterday, so that is an option now that it appears to work very well

[Dman3d]

I remember that thread. This was with GPU? I was hoping to do this fully with V-ray GPU. Based on that advice I will crank the samples limit to something absurd and see what happens.

[fixeighted]

DOF Pro is an AE plugin I'm afraid.

[Dman3d]

I cranked the Samples limit to 500k and it seems to clean it up. I can probably drop that to 150-250k as its half way done and clean at this point

[Dman3d]

Will run this overnight tonight so see how the whole thing turns out, some more region render dorking around looks like may need to go back up to 500k

and maybe, the bump map I have on my glass may be part of the problem, still working on it.

[^Lele^]

besides the Max AA, i would ensure to pad the render with the Min.
If you know it needs to run for long anyway, you could do worse than giving it 24 Min subdivs, or so.
This is because for these kind of low-probability effects a non-adaptive "padding" will do a better job (it's the same for thin, long moblurred highlights, particularly from small light sources).
It will then prepare the stage for the adaptive part of the sampler, which will refine what the fixed stage "found".
A low Min AA will make "filling" the bokhes very very slow.
You should then be able to use drastically less max AA as a result.

EDIT: it occurs to me that to do this you need to use standalone, as the UI doesn't quite allow for this with GPU. You can set adaptivity to 0.0, and that will sample without the adaptive routines. But it'll never stop the fixed sampling in favour of the adaptive one, later in the rendering.
I've asked if min AA (the same as CPU) could be introduced.
Lastly, it may be obvious, but these effects rely on camera rays much more than on secondary ones. So if you're going to render with many AA samples, feel free to lower the samples per pixel (make it 2, or 4. You need some for the rest of the image.), as it'll shave you some considerable time to convergence.

[Dman3d]

besides the Max AA, i would ensure to pad the render with the Min.
If you know it needs to run for long anyway, you could do worse than giving it 24 Min subdivs, or so.
This is because for these kind of low-probability effects a non-adaptive "padding" will do a better job (it's the same for thin, long moblurred highlights, particularly from small light sources).
It will then prepare the stage for the adaptive part of the sampler, which will refine what the fixed stage "found".
A low Min AA will make "filling" the bokhes very very slow.
You should then be able to use drastically less max AA as a result.

Lele
I assume you are referring to Adv CPU, I may be missing something here with V-Ray GPU
But it looks like cleaning up the bokeh map, remaking it higher res, and using online upres site for the bump map on the glass with denoising set to max seems to of helped a lot along with cranking Samples limit to 500k

Will crank it over night and see how it turns out

[^Lele^]

I edited my original post, sorry for the somewhat loose feedback.

[Joelaff]

Definitely min AA should be increased. I would say 3-4 at the least, sometimes much higher.

We used to have good luck increasing the camera subdivs, which you can still do in there Max script windows by selecting the camera and entering $.subdivs=64 (8192, etc ) or whatever. I only messed with this a little bit in vRay5, and I am not sure if it helped. Need to do some testing.

Given the number of camera rays required anyway you can likely gain a little speed boost by lowering the min shading rate to 3 or less.

A softer image filter helps as well, like Area at 1.98 or Soften at 1.98-4.5 or so.

[Dman3d]

still crunching, its about here at the moment. Will see how it looks in the morning. You can see some of the complex bokeh DoF areas need to resolve more
The scene is based on and inspired by the wonderful photgraphy of Sarvesh Chaudhari. Wanted to explore working on detail of macro shots with V-Ray 5. This is "scene 01" hope to do more macros after this.

[Joelaff]

I do like the palette.

Cheers.

[kosso_olli]

still crunching, its about here at the moment. Will see how it looks in the morning. You can see some of the complex bokeh DoF areas need to resolve more
The scene is based on and inspired by the wonderful photgraphy of Sarvesh Chaudhari. Wanted to explore working on detail of macro shots with V-Ray 5. This is "scene 01" hope to do more macros after this.

Bokeh shapes are looking really nice, well done!

[Dman3d]

Running overnight predictably cleaned up a bit more
Thanks for the input everyone

Above is 200% zoom