Does adaptive sampling happen in the viewport?

I have a scene that I was troubleshooting for months in the first half of 2020. I have several essays worth of notes about this file and I’m revisiting it to see if I still have the same problem in newer versions of Blender.

In my notes at one point I say it seems the problem is caused by adaptive sampling. With viewport denoising I have no ugly areas but in the final render there are several areas that are turned into absolute trash by the denoiser (at render time) but not by the denoiser (in compositing).

Does anyone know if adaptive sampling happens in the viewport render and does adaptive sampling info get used as part of the at-render-time denoising process?

Adaptive sampling does not influence viewport samples, if you turn it on with noise threshold at 1 and min samples at 1 you can see in viewport render it will go through all the samples you have set for viewport if 2000 all of them, but in final render it will cut them dramatically.

Adaptive sampling simply turns down the number of ray samples in areas where it considers they are not necessary and can potentially result in more noise but not ugly artefacts.

I can confirm that the default denoise option in rendering tab can make some horrible looking artefacts, I never use it. I either render out the noise with more samples (the best quality option) or use post pro in gimp to have more control on how I denoise.

I don’t do professional level things so the OIDN denoiser has always worked well enough for me. I just have this one scene where activating adaptive sampling causes horrible denoising issues in an area that is actually fairly well lit.

The one intentionally odd thing about this file is it follows Jonathan Lampel’s advice of using real world strength lights and then adjusting exposure. In the previous version of the file where I left exposure at default and adjusted light strength around that I never got any rendering or denoise issues.

I like the simplicity of his idea “just use real lights then adjust exposure like a real camera” but I’ve been trying FOREVER to use that concept in this scene but it seems using that approach is incompatible with adaptive sampling + denoising which are necessities for me because I’m working on a weak laptop and don’t have much free time or patience.

I do not do professional things either! The denoiser can work ok under some circumstances but not always, and it is possible that he combination of the 2 options can influence the result, adaptive sampling can cause more noise and thus force the denoiser to work “harder”

Might want to look into how it is well lit. Mixed size emission geometry is a no-go for me due the the noise it generates (at least used to, haven’t tried in a while), and also re-transmitted light (i.e. translucent materials) can get noisy real fast, same with light transport through caustic features. Sometimes I have materials I don’t want to denoise much as it removes the dense texture information in them, in which case I use cryptomatte to do selective denoising - might be something to look into for your case.

No emission geometry actually doing lighting. They are just bright enough to be seen and then there is a point light above them. The translucent texture on the windows is using light path node to let the full strengh of the outside light come through as if the window pane isn’t really there.

I think my problem is just that real world strength lights are not handled good enough and I have no choice but to work with overall brighter lights. The advice I was trying to follow is this: https://www.youtube.com/watch?v=ipqyVWm5JmY

The scene is the restaurant in the first post of my sketchbook: Thinsoldier sketchbook This is the normal blender lights version. The Extra Lights addon version is the one with the noise problem. I don’t think I have an image of it online anywhere. The counter top to the right of candle would be absolutely horrible while the rest of the image was fine.