Corona 2/3 very long render SSS materials

[Feodor]

In the beginning I want to thank you for the wonderful render, and the work that you are doing!

I have this problem
On an example of this heavy scene a lot of SSS material. Corona 1.7 visually, the noise goes much faster, Corona 2 counts four times longer.
This scene in the format 1920x1920 when the renderer writes will be counted as 70 hours !!! Instead of 16 !!! :(

Example 640x640

[TomG]

Is it possible you can share the scene? There's the private uploader https://corona-renderer.com/upload if you can't share it with the whole world here :)

Would be interested to see what those SSS materials are set up like, and the scene would answer that and more. Cheers!

[Feodor]

No problem! I sent you this scene

[mike288]

couldn't that be explained by (https://corona-renderer.com/blog/corona-renderer-2-for-3ds-max-released/):
"We’ve made multiple changes to adaptivity (as listed in other parts of the blog) and this means that Noise Level is not the exactly same in Corona Renderer 2 as in 1.7 and earlier. We’ve balanced it as best as we can, but be aware that a Noise Level of 4% in Corona 2, for example, may well be of higher quality than the same noise level in an earlier version (and so may take a little longer to render to that Noise Level)." ?

[Feodor]

This has been checked, the noise goes away for a very long time.
In the render settings is "Noise Level Limit: 3.0"
If from each pass the noise went away faster the problem was not. The noise lasts longer. and that would wait for the noise level of 3% I need to wait 4 times longer.

I Send you File.

[romullus]

I don't think it's noise level alone, if you look at attached screenshot, you will see that rays/s differs from 600K to 200K between versions - clearly something is not right.

[maru]

Note: the screenshot is showing Corona Renderer 1.7 Hotfix 4 vs Corona Renderer 3 09.08 daily build, not Corona Renderer 2.

The user is getting:
-5 passes in 13 minutes in Corona v3
-5 passes in 3 minutes in Corona 1.7 Hf4
Visually the two images look very similar (which is expected anyway with the same number of passes), despite the render time difference (which is not expected).

We will need the scene to understand what's going on.

@Feodor - can you please clarify whether the issue appears in Corona 3 or also in Corona 2?

[Feodor]

The scene was sent to you on the Dropbox, I hope you got it.
Yes, this test did for you on Corona 3, but it was noticed by me for Corona 2. I expected similar problems for others, but they were not, today I decided to write about it.

[Feodor]

This scene I started doing in Corona 1.6, in 1.7 I continued аnd in 2, I uploaded the scene, ran the render and "turned pale" :)

[TomG]

It seems to be the Corona Scatter in the sugar bowl.

In tests, I found 1.7.4 fastest, 2 was slower, 3 daily was slower again (in terms of rays per second). Hiding the scatters in the sugar bowl caused a massive jump in rays per second in Corona 2. I wonder if related to this report about a scatter slowdown? https://forum.corona-renderer.com/index.php?topic=21394.0

EDIT or it could be the materials on those scatters, haven't looked yet, just noticed that hiding them makes a huge difference to performance.

[TomG]

So, turning off the Refraction for the sugar scatters speeds up the Corona 2 render a lot. So the issue isn't with Scatter itself.

[Feodor]

Turning off the refraction feel will affect the quality of the material loss of realism.
In the previous version it was possible, in the new one no longer? :(

[maru]

Right now my guesses are that:
1) Using environment overrides may be causing this
2) Scatter creates a different patter in 1.7 and v2, so in v2 maybe there are for example much more objects?

[TomG]

So, in the meantime, first thing would be to change that material. In the scene the sugar probably doesn't need translucency, and glossy refractions, and subsurface scattering, for such tiny objects. I am sure a good material can be obtained without such a complex material :) Though refraction seems to be the one that impacts performance (didnt do exhaustive testing). I just turned it to 0, but perhaps less glossiness would work, or removing the SSS and translucency might help, etc.

Still leaves the question as to why the same material performs differently with different Corona versions though.

Scene as sent
1.7.4:
Total 648,169
Actual 688,722

2:
Total: 294,493
Actual: 338,589

3:
Total 275, 737
Actual 239, 797


Quick test making refraction 0 for the sugar material

174:
Total 1,098,967
Actual 1,171,046

2:
Total 986,182
Actual 1,059,827

3:
Total 1,015,323
Actual 823,261

(so, even if we resolve what is happening, that is still one heckuva hit to your performance just for the sugar :) )

As a note, the scene as loaded reports some missing texture maps, and has Ornatrix too, which I don't have installed.

[TomG]

I think this will be the same issue as "blurry refractions seen in blurry reflections" that has been reported, only multiplied exponentially by lots of small objects doing that. BTW, enabling caustics for that material will raise your rays/s, though I can't speak to what it does to the look of the render.