Strange banding in Clouds

[Tangled-Universe]

I'll show you another example...it's a crop from my Dutch Landscape WIP you know about...I will render it now using 32 atmo samples and 32 cloud samples (det 0.82) with AA8 and detail 0.8 and GI 1/2/6 in the renderer. I'll compare it with RTA @ AA8 with 1 atmo sample and 4 cloud samples.

In my previous test AA4 allowed for a 2-fold reduction of cloud samples and since AA8 is 4 times "better" than AA4 ((8x8)/(4x4)=2) I estimate an 8-fold reduction of cloud samples should be ok.

[Tangled-Universe]

Here it is...
Notice the slight slight slightly more noisy atmosphere, but especially notice the quite different looking clouds!
Rendering it now with RTA and 2 atmo samples + 8 cloud samples...see what it takes to make it look as good as RTA OFF

When it comes to practical examples I think it can't get any closer than this.
It's a difficult lighting situation, requiring good sample settings and it has objects which justify using high AA levels and thus even more carefully chosen sample levels for atmosphere/clouds. As this test shows AA8 doesn't offer extreme reduction in samples as it is already slower and inferior in regards of cloud-rendering.

This fits with the previous example which showed atmosphere rendering comes cheap and clouds don't...

[Tangled-Universe]

Getting closer with 2 atmo samples and 8 cloud samples...took exactly 18 minutes:

[Tangled-Universe]

Last one...now with 12 cloud samples and still 2 atmo samples...
It turns out I have been lucky with the 1 and 2 atmo samples, since this "seed" shows I was short on atmo samples.
Clouds are nearly the same, but rendertime was 32m37s.

Logically I won't push this further...I think/hope the examples are clear enough.

Cheers,
Martin

[Oshyan]

I guess I'll have to wait until my own re-tests are finished to comment with anything useful. All I know is in the past my findings have been different than what you show, so it's very surprising to me. Definitely worth more investigation.

- Oshyan

[Tangled-Universe]

Well, I'm a bit surprised too since I too remember we tested some stuff together "in the beginning"...at least we discussed in detail how this would be benificial and how the approach should be, for hours even
Nonetheless it turns out that it somehow doesn't hold, but frankly I don't understand why this happens now.
Not that I think it has to do with changes in TG2, but the only reason I can think of is that we tested it too artificial and not enough in practical situations. Perhaps.