Community
Arnold for Maya Forum
Rendering with Arnold in Maya using the MtoA plug-in.
cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Why eliminating noise in archviz scene take so much time than another render engine?

10 REPLIES 10
Reply
Message 1 of 11
sotomangel
389 Views, 10 Replies

Why eliminating noise in archviz scene take so much time than another render engine?

Hello.

I have time working with Arnold from it was release until now, and I have noticed that adaptive sampling is not working 100% perfectly.

For example, I have and scene AA=0 and Max AA=7, noise threshold=0.015 it last 1h35min at 720p resolution the result is the image with noise (not acceptable).

But if I configure AA=6 and Max AA=7 and all the same it last 1h56min the result is the image with acceptable noise.

I understand that Arnold works different as other render engine, but in architecture we need a system that works very well in different scenarios with acceptable time.

If I render this image in 1080p the time is huge. Between 3 to 6 hours. (My CPU Intel 5820k).

Is there any possibility adaptive sampling with noise threshold will work better? With less noise and less time, because as a Max instructor is easier teach Arnold but hard that all the students have a PC with acceptable CPU. Not all the people have good CPU.

In other render engine the time and quality is acceptable with less time. I now the engine work different but if Autodesk has a render engine, it should work for CG artists and architects, keeping in mind that not all the people has render farm or super computers.

Is there any possibility in future something happens and we (architects and instructors) could render faster and with good quality (or acceptable)?

10 REPLIES 10
Message 2 of 11
Stephen.Blair
in reply to: sotomangel

You don't say what version you're using, but assuming you're using the latest Arnold, your sampling settings could be better.

AA should be high enough to get decent starting point for adapative sampling. So AA 7 could be right. AA 0 is way too low.

Max AA should be higher. Maybe not as high as 20 or 30 (but soon as we address some issues, you'll be able to set a big Max AA)



// Stephen Blair
// Arnold Renderer Support
Message 3 of 11
sotomangel
in reply to: sotomangel

I´m using the last version of arnold. You are telling me the second experiment was good with the difference I must increase Max AA to 10-12. If I change that the render time will not increase exponentially?

Message 4 of 11
sotomangel
in reply to: sotomangel

I understand Adaptive Sampling like this:

When I press RENDER the pixel analyze with 1 AA sample, and it check for noise threshold, if the noise is upper than threshold, the engine hit 2 AA sample and check, and the algorithm continue hitting samples until Max AA sampling or reach noise threshold.

Message 5 of 11
Stephen.Blair
in reply to: sotomangel

Try something a little higher. Something like 30 or 50 will probably be too slow in the current Arnold, but as soon as we fix some issues, that will be the way to go.



// Stephen Blair
// Arnold Renderer Support
Message 6 of 11
sotomangel
in reply to: sotomangel

Ok. but this change could increase the total samples rays too much? and consequently the render time?.


Message 7 of 11
aaronfross
in reply to: sotomangel

In my tests with architectural still frames, I have found that the sweet spot for camera samples is something like:

Camera (AA) = 2 to 3

AA Samples Max = 7 to 10

Adaptive Threshold = 0.02 to 0.015

Keep in mind that with higher pixel resolutions, you can get away with lower sampling settings because the grain is less noticeable.

And the Arnold Denoiser (aka "Noice") is very effective at removing grain in a post-render process. MaxToA makes this super easy with the settings in the Denoiser tab. You render your frame(s) with "Output Denoising AOVs" enabled, then run the Denoise process when the render is finished.

I don't recommend the OptiX Denoiser because it permanently bakes the smoothing into the image at render time, you have no ability to try out different settings, and it can't work with animation. The Arnold Denoiser is really the way to go.

Message 8 of 11
aaronfross
in reply to: sotomangel

Oh, and don't forget to check in on the Samples for each light, as well as the Diffuse samples for the rendering... increasing these can reduce grain for direct and indirect light, respectively, without increasing Camera samples.

Message 9 of 11
sotomangel
in reply to: sotomangel

Aaron in other samples like diffuse, transmission and specular which are the values that works for you?

Let's say image 1080p:

Camera (AA) =5

Diffuse = 6 -Ray depth = 2

Specular = 3-4 Ray depth =1

Transmission=2-3 Ray depth = 8

AA Samples Max=6-7

Adaptive Threshold = 0.015

This is my regular settings after test with other values.

Message 10 of 11
aaronfross
in reply to: sotomangel

The answer is going to depend on the specifics of your scene.

Generally, I would set Diffuse Rays to 3 to 5 to give sufficient bounce light. This is not like Final Gather, where the optimal number of rays was only 2. You can push Arnold Diffuse rays much higher without impacting performance very much. The rays are cheap, the samples are expensive.

Specular Rays need to be 3 or 4, otherwise you will have black areas in the render where shiny surfaces are near one another.

As for sampling, I have found that I can lower the Camera (AA) setting to 2 or 3 and get substantial improvements in render times. This is especially true for images with large surfaces with little detail.

Diffuse samples can range from the default of 2 up to a max of 6, depending on how much indirect illumination there is in the scene. If the scene is an interior with natural daylight coming through a window, you'll need high Diffuse samples. If it's an interior directly lit by artificial lighting, you don't need high Diffuse samples. And again, if you increase the number of Diffuse rays, you can probably get away with lower Diffuse samples.

Remember that the Diffuse/Specular/Transmission samples are subordinate to the Camera samples. If you increase Camera samples, you're basically multiplying all of the Diffuse/Specular/Transmission samples. So in the end what's important is the RATIOS among those component samples. The Camera (AA) is the master quality setting, and the individual component sample values are tweaks or adjustments.

Here are a couple of slides from my LinkedIn course, 3ds Max Rendering with Arnold, that help explain the mathematical relationships between camera samples and component samples.

3697-01-05-01-samples.png

3698-01-05-02-ray-depth.png

Message 11 of 11
pproestos
in reply to: sotomangel

@Miguel Soto Arnold denoiser+Topaz AI gigapixel an you're done... And this is the only post production I personally do with this awesome art engine. Arnold is not a renderer. Arnold produce art, all the others produce renders.

Can't find what you're looking for? Ask the community or share your knowledge.

Post to forums  

Autodesk Design & Make Report