Do I need to adjust GI/AA balance when rendering elements only?

[pokoy]

I have a shot with stronger DOF, for the beauty render I needed to increase the GI/AA balance a bit. Since rendering elements only won't need any GI I'm not sure if I have to set the value to something lower (so DOF resolves faster) or if the balance doesn't play any role since GI rays are skipped anyways?

[pokoy]

Strange. Lowering the balance to 8 reports a final rendertime of 9 hours, increasing it to 64 reports 5 hours, and it looks much better with even 3 passes.
I thought balance should have no impact because there are no GI calculations, or if anything it should be faster with lower balance so more rays are spent on AA. But it's strange that it's the other way around... So where am I thinking wrong here?

[maru]

It would help if you could tell us what elements you are rendering.

[pokoy]

Ah sorry, of course. Masking only (Object IDs), nothing else. I have added ZDepth at some point but I could see this with Masking only already.

[pokoy]

Some more info: Corona 1.2.1 (but I don't expect this to have change, though who knows), and Masking element in RGB mode.

[FrostKiwi]

If you render using the render elements only dialog, then only AmbientOcclusion will be affected by the AA/GI slider.
All others ignore it, as they don't sample GI, but only sample the edges, so AA only.

Some elements cannot be rendered using that dialog, these are all the passes "extracted" from a full render, meaning reflection, indirect lighting element, refraction, emission etc.
These are affected by the GI/AA slider in the same way a proper render would.

[pokoy]

If you render using the render elements only dialog, then only AmbientOcclusion will be affected by the AA/GI slider.
All others ignore it, as they don't sample GI, but only sample the edges, so AA only.

That's what I thought. But it seems that even with pure AA sampling (as in Masking elements, in this case with DOF) the GI/AA balance has an impact, and it affects sampling in an 'inverted' way it seems.

[FrostKiwi]

That's what I thought. But it seems that even with pure AA sampling (as in Masking elements, in this case with DOF) the GI/AA balance has an impact, and it affects sampling in an 'inverted' way it seems.

How do you determine that? Don't just read the Rays/s number, it is not correctly interpreted in certain scenarios.

[pokoy]

If no GI contributes to the time a pass takes, I should get the same time estimate with a balance of 8 and 64... it shouldn't play a role whatsoever, at least that's what I thought.

I should do some tests but unfortunately I need to get the job out of the door, I may have the time to dig into it later. As for now I just wanted a hint whether I need to touch the balance setting in this situation.

[maru]

You are using pass limit, right?
I tried reproducing it but a simple scene took exactly the same time when using "render only elements" with 4 an 64 GIvsAA.
It would be great if you could reproduce it and share the scene, or share the original scene (you can upload it somewhere and send link to support@corona-renderer.com).

[pokoy]

Thanks for testing Maru. So it's something on my end but I really can't think of anything other than Corona version. Have you tested this on 1.2.1?
I am using a pass limit, yes.

I'll try to find some time to reproduce this, maybe it will show the same behavior on a stripped-down scene.