Announcements

The Autodesk Community Forums has a new look. Read more about what's changed on the Community Announcements board.

Arnold AA settings yield diminishing returns

Anonymous

Arnold AA settings yield diminishing returns

Anonymous
Not applicable

After watching a wealth of tutorials and reading documentation I've got (what I feel is) a decent foundation in arnold render settings. There is one area however where what I see in the tutorials is not matching what happens in reality when I apply it in my arnold render. Assuming there isn't a major change from Arnold settings 2021 to the most current 2022 version, I don't understand why adding samples in the Camera AA settings actually makes my render worse than when I don't use them at all. I'm currently using my CPU as my primary render source as the gpu is too much for my current workstation. I would attempt gpu, but this is the machine I have to work with for the moment- and even then I couldn't control the individual samples in AA.


Before adding the camera sampling, I have a simple set up with an area light at a resolution of 2048 and acceptable results when I sample around 8. Beautiful results at 16, but of course, render time is a factor.

If I add just camera aa samples, the images is fine - but the moment I add either specular or diffuse samples, my image erupts in noise. Using AOV's ive determined that the problem is definitely sampling in the Camera AA diffuse/spec settings. I know that it is easy to oversample either of these and you don't need a large number considering how the samples are calculated at render time. Further, I thought that using these properties was supposed to allow you to dial down a high sample rate in your light settings to save render time and still have decent quality. I know you can use the denoiser to aid in the image output but right now I believe I can still get better reduction in noise before using that.


Given all that, IF I am not mistaken, I should be set a small to moderate value of AA samples (including diffuse and spec where applicable) and have a moderate amount of light samples to produce an attractive result. Why, then am I still consistently getting significantly noisy renders? Is this one of those issues where it could be a hundred reasons outside of my render settings? Is there something super basic that I am overlooking? Surely I don't need to drive up my diffuse samples to 5+ on top of camera aa in order to get a good render





0 Likes
Reply
662 Views
10 Replies
Replies (10)

lee_griggs
Autodesk
Autodesk

An image is worth a thousand words.

I am guessing the reason that you don't have noise is because with diffuse_samples: 0 you are effectively disabling gi bounces in the scene (hence no noise). Introducing diffuse_samples will of course then look noisy in comparison.

Lee Griggs
Arnold rendering specialist
AUTODESK
0 Likes

Anonymous
Not applicable

following you will find some images describing the situation perhaps better than the text. I'm aware, or so I thought, of the effect of the diffuse sampling on the image - but if the natural/only result of adding samples is going to be unavoidable/unfixable noise, then what is the point of the option? is it now deprecated? Is not the objective of these additional samples to allow the camera to gather more info on light and reflections relative to objects in order to improve the image - And then pair that with good direct light sampling in order to achieve a happy medium? The majority of answers across the internet is to increase samples until an acceptable result is achieved- but not oversample. And in this example, I'm using a single light and material override , so there aren't complex shaders and multiple lights to calculate. I've played with adding another light and adjusting it's distance but to no real avail. While this may all be obvious to most others, I just don't see it.

0 Likes

Anonymous
Not applicable

Here is example onep1.png

0 Likes

Anonymous
Not applicable

Here is example 2 p2.png

0 Likes

Anonymous
Not applicable

Here is example 3 p3.png

0 Likes

Anonymous
Not applicable

Here is example 4 p4.png

0 Likes

Anonymous
Not applicable

Here is the actual file for reference. I am continuing to research answers. https://drive.google.com/file/d/19Prfmf4mztJF4hUTVI_LNxNgIVvSDWZo/view?usp=sharing

0 Likes

Anonymous
Not applicable

I've attempted to use the insert link for everyone's convenience but the form submission isn't working, so please copy and paste if you have the time and patience.

0 Likes

lee_griggs
Autodesk
Autodesk

There are two of the same lights overlapping each other. There are also two pieces of coincidental geometry. Lights samples shouldn't need to go above 3 or 4 in most situations. Increasing the ray_depth settings will not affect the sample quality (just ray bounces) and will increase render times unnecessarily in this case.

I suggest you either clean up the scene or try a different scene.

You shouldn't need to increase the resolution of the light unless you have a texture map connected to its color.

1646806904331.png

Here is an updated scene that takes 1 min 28 secs to render for me. I pasted your geo and light into a new scene and applied a standard_surface shader to them.

sampling and gpu-updated.mb

1min28.jpg

Lee Griggs
Arnold rendering specialist
AUTODESK
0 Likes

Anonymous
Not applicable

Thank you for the file and catching the flaws in the scene. A new clean scene was very helpful. my appreciated.

0 Likes

Type a product name