Reduce noise in single material vs entire scene

V-Ray V-Ray for 3ds Max General
1

I have an arch viz scene where everything renders (cpu) pretty noise free with modest render settings max subdivs = 8 noise thresh = 0.01. (using vray de-noiser mild)

The wood lined ceiling however is not so noise free. So increasing overall setting to macx subdivs =10 and noise thresh = 0.005 helps a lot but almost double render time.

My question - is there a way and is it advisable to adjust that one ceiling material so it renders more noise free versus cranking up the overall render settings quality?

2

Try increasing the Subdivision multiplier in the V-Ray object properties (RMB on object > V-Ray properties) for the particular shaded objects.

3

Thanks, that’s just the info I was looking for!

4

The best way possible to get quicker renders is to leave max AA at defaults, and raise N.T., not to clamp max AA.
In your case, if you add a SampleRate RE, there should be large areas of red, that mean the max AA has been reached, but the pixel hasn’t reached the set noise threshold.
In fact, the doubled rendertime you experience is a sign of it, as you went from 64 (max) samples to 100, about double, *if* no adaptivity was into play, which seems to be the case. :slight_smile:

Raising subdivs multiplier will only work for secondary rays, so for a flat diffuse area it’ll do the trick, but where fine geometric detail is concerned, the max AA of 8 (64 rays) will still be the limiter.
Further, changing subdivs multiplier will leave the image with an uneven noise level (which you already had, anyway, because of the limited max AA.), which is generally undesireable (the denoiser won’t hide it all the time, and at some point it will show the difference where the noise levels change.).
Lastly, it’s seriously non-obvious to debug a scene where those multipliers have been altered, if one has forgotten or doesn’t know they have been changed (f.e. a rerender to be made two years later.).

If you really want to render the ceiling with different settings than the rest of the scene, it’s much safer to render it as an object mask with lower NT (assuming defaults everywhere else.).

5

Thanks for your comments Lele. Always valuable and educational for me. You may need to dumb it down a bit, or a lot, in my case <g>. To be sure, for all of this discussion we are talking about the Bucket sampler.?

The best way possible to get quicker renders is to leave max AA at defaults, and raise N.T., not to clamp max AA.

In the Image Sampler (Antialiasing) section I leave the Min Shading rate at default 6. In the Bucket Image sampler Section are you saying Max subdivs = max amount of AA?

So I changed the Max subdivs from default 24 to 8. Then I tried it at 10 and the render time about doubled. If I had left it at the default 24 then the render time would be… a very big increase (or so it seems to me).

Raising the N.T. would give faster render but image with more noise. I was trying to reduce noise, if it takes longer to render so be it.

In fact, the doubled rendertime you experience is a sign of it, as you went from 64 (max) samples to 100, about double, *if* no adaptivity was into play, which seems to be the case.

I don’t know or understand why there would be no adaptivity in effect?

To summarize and if I understand correctly, It is best to leave Max subdivs at default 24 and then raise the noise threshold to reduce render times and lower noise threshold to reduce noise at the cost of longer render times?

Full disclosure - I have completed a lot of Arch Viz renderings using Max sub divs = 6, N.T. = 0.01. they render reasonable fast and seem pretty noise free for the most part. For interiors or dusk shots, scenes with lower light, I would use max subdivs = 8 and generally leave noise threshold at default 0.01. I’m getting the sense that I am not following recommended best practice in this regard.

I did not end up using different settings for the ceiling. The project came out well and is completed and archived. I will use all this info on the next one.

Please forgive my incomplete understanding of this and thanks very much for your helpful explanation!

6

Thanks for taking the post graciously! :slight_smile:

It works more or less the same regardless of sampler.

are you saying Max subdivs = max amount of AA?
Yes, exactly.
Max subdivs is the square root of the maximum number of camera rays to shoot.
So, for 8 subdivs, you’d be casting 64 camera rays (also called primary), and assuming you have the rest at defaults, 6 secondary (call them shading) rays as per the MSR, for each camera ray.
This is a guideline, as V-Ray will also obey a number of other conditions in order to decide how many rays to actually cast.

Ignoring secondary rays, as we won’t change MSR at all, and concentrating on primary rays, it’s apparent why a small change from 8 to 10 produces such a big change in rendertimes: rays went from 8x8 to 10x10, so from 64 to 100.

I don’t know or understand why there would be no adaptivity in effect?
Adaptivity will be free to decide where to sample more until it has primary rays to cast.
With a max of 64 rays, you’re forcing it to stop early, likely a lot earlier before any image part has converged to the 0.01 noise level you set.
Mine is a guess, however, as i have no access to your image, and most importantly to the vraySampleRate and vrayNoiseLevel render elements.
The first should have a lot of red and orange areas, that mean max AA has been reached for those pixels.
The second, that should ideally contain pixels only with the value you set for N.T., so 0.01 (hardly visible.), should instead show brighter parts.
If my analysis is right, the parts in red in the sampleRate RE and the visible pixels in the noiseLevel RE should correspond (more or less. some pixels may well have reached the 0.01 noise threshold.).

Consider the attached contact sheet, rendering a random crop (it was there when i opened the file.) of some evermotion interior scene:

In the top row, your original settings: 1-8, and a N.T. of 0.01, in the bottom row, the defaults for AA (1-24) and a slightly higher N.T. at 0.025.
The first renders in 103 seconds, the second in 110, so about equivalent (use as vague yardstick, my CPU wasn’t set up for benchmarking and may have declocked in either render.).
In order, you’ll find the untouched render, the sampleRate RE, the noise level RE for which i measured min, max and average value (this maps directly to the N.T. reached by the render.).
The next image is a mix between the samplerate Re and the noiselevel RE, showing the correlation.
Then come the denoised result (same settings for both, ofc.) and a measurement of the min, max and avg noise level across the pixels of the denoised image, as it has the nice side effect of showing the contrast edges the denoiser managed to preserve (so, more readable detail, better a denoising job, less readable detail, more blurry a denoising job.).

clamped_AA_CS.jpg
clamped_AA_CS.jpg1404Ă—702 289 KB

As i was saying, the top row shows those qualities i mentioned: the sampleRate RE is red in large areas, the noiselevel RE shows visible parts, and while some areas do reach values below the set N.T. (0.007 for min values), most of the rest is miles off the needed noise level, up to twenty times higher, and the image has an average noise level of 0.024, or 2.4 times higher than the one you set. It’s a sure sign of the AA sampler unable to do its job of converging the image.
This will also impact the denoiser job across the image: it will clean the uneventful pillows (they are bland with any metric: geometry, textures, lighting), it will eat into the detailed parts of the bricks (this is evident in the rightmost image of the set.)
The bottom render is -for an analogous rendertime- better under all considered aspects, it has uniform noise, as the sampler has been allowed to clean up the image to the set N.T., and this is verifiable by the noiseLevel RE’s measurements: no pixel is above the set N.T. as max and avg values are both below it.
As the sampling on the image has been done properly, the denoiser results will retain more detail, even if the set noise threshold is higher, as the threshold has been reached everywhere.

tl;dr: it’s better to not second guess the sampler by lowering Max AA, but rather raise or lower N.T. as needed. Ideally, always rendering with a sampleRate and a noiseLevel RE for guidance.

7

I fetched a bit more info i can share on why that V-Ray property exists at all.
Originally, it was indeed so that it could be raised for a node (and all its associated shaders) at once.
Since V-Ray 3.3, however, that part of its usefulness has ceased.
Raising the value isn’t guaranteed to do anything specific to the noise level of the node: if the adaptive sampler is active, tracing twice the secondary rays will mean it’ll stop earlier -if primary rays are not needed, f.e. on a flat, even surface-, but it may well mean it’ll render for longer if camera rays are needed, without any appreciable change in noise levels (tracing more secondary rays, f.e. for sub-pixel detail, or for DoFed/Moblurred parts.).

So, why would it stay?
So that it can be lowered.
If one knows a node is moving very fast across the screen, leaving but a smear, then one knows that primary (camera) rays will be needed for its noise level to converge to the set N.T., rather than the secondary ones.
In this case (and those of DoF, and sub-pixel details), lowering the value will speed the rendering up as V-Ray will trace f.e. 1 primary + 1 secondary ray for each pass, instead of 1 camera + 6 secondary by default.
1+1 is also how V-Ray GPU works, as it currently lacks an editable MSR value, and it’s the reason why it’s hard to compare noise levels across the two engines, as it’ll vary greatly depending on what is being rendered, and which part is favoured by one approach or the other.

There are plans to automate this behaviour (it is already automatic for Hair and some fPro items), but so long as they aren’t in place, the parameter has to stay.
Just don’t raise it above 1.0! :smile:

8

tl;dr: it’s better to not second guess the sampler by lowering Max AA, but rather raise or lower N.T. as needed. Ideally, always rendering with a sampleRate and a noiseLevel RE for guidance

wow and thanks to you Lele. Really appreciate your time in setting down that detailed and comprehensive explanation. Kind of a mini master class.

9

Something odd is happening with this post. It only saves the first part of what I posted and won’t let me save any edit/add.

Trying a separate post.

I am chagrined to think of the number of renderings I have done while not using these settings correctly.

10

Something odd is happening with this post. It only saves the first part of what I posted and won’t let me save any edit/add.

Trying a separate post.

I am chagrined to think of the number of renderings I have done while not using these settings correctly.

11

Sorry, this forum is acting very strange will not accept full post. May have to do w inserting emoticons.

Something odd is happening with this post. It only saves the first part of what I posted and won’t let me save any edit/add.

Trying a separate post.

I am chagrined to think of the number of renderings I have done while not using these settings correctly.

Going forward I will leave max subdivs at default 24, raise/lower noise threshold as needed. I will learn touse and interpret the sampleRate and noiseLevel render elements.

Thank you x10!

​

12

You’re very welcome, and notice that mileage may vary: very strong highlights may well take the sampler to max AA, at nigh any noise level, so be wary, and use the sampleRate RE! :slight_smile:

Yeah, the forums have been occasionally acting up a bit, in the past few days.
It comes and goes, i know it’s being worked on, sorry for the bother.

13

Especially with motion blur or DOF. But we know what to do in those cases.

14

And you can rest assured we’ve not stopped thinking of ways around the issues.