Is this expected behavior? The tile-looking pattern does not represent the geometry’s makup.
Hard to know what information to supply in this case other than a 2gb file + assets, so let me know what other info might help diagnostics…
Hi, thanks for posting. I am afraid there is no other way but reviewing the scene. Please try to simplify it and send it. You can use our support contact form for that. Please include a link to this thread in you e-mail.
Done! .
.
.
.
.
Overlapping polygons?
No not from what I can tell.
Ok! Thanks for looking into it ![]()
So what the heck would have caused that, though? Can you do a diff on the settings @vladimir_krastev ?
Here it is.
Looks like it’s to do with retrace threshold. I generally turn this off as render times explode when it’s on.
In the case of this small region it’s almost 30% slower with it switched on.
With retrace OFF (blocky artefacts present) and switching ‘store direct light’ → Off, it fixes the issue on the column as you can see here:
But the ceiling’s issue is still present.
Looks like it’s to do with retrace threshold.
Definitely. Now that you bring it up I recall having similar issues when rendering interiors in the past. I used to lower it as well, but not turn it off completely.
In my experience it is mostly needed with bad geometry. I used to set it to 0.5 I think but this value does not work in every scene. You can test the lowest value that works.
Retrace should always be on for the LC for final rendering.
It’s a toggle, one may want the LC as primary engine too, for purposes other than photoreal rendering (f.e. as a painterly pass, or a super quick preview of light distribution. Both would be defeated by retracing.).
The painterly effect however worked better for the old LC as it had rounder samples, and a different screen distribution.
The quick preview may work as well now, with some tweaking to the LC settings.
Retrace is the part of the rendering algo that decides an LC sample is not in fact expressive of the shading needed, and per-pixel sampling should instead be used: turning it off produces precisely the results you see, as any LC sample is considered fine and rendered as is, light leaks included.
Primary samples are generally fine, but anything through reflection and refraction is often problematic, and needs retracing (hence the issues seen in that RE).
Analogously, you should not turn off “Store direct light” as you’d lose a big chunk of LC performance.
You can check in the Log what percentage of the LC has been used -directly- in the final render: it’s often around 20 to 30%, the rest is being retraced.
TL;DR: Trust the defaults, they have been chosen for non arbitrary reasons.
I know that the defaults will generally give the ‘best results’, but at some point a render that can be 30-50% faster with retrace turned off is worth it, despite the artefacts it might bring. Some of us don’t have the luxury to wait anything longer than 6 hours for a render nor do we have the money required to overcome this with raw computing power.
Well, if it was you wouldn’t be here. ![]()
As it is, it costed you more time than rendering at defaults, with slightly higher noise threshold, and then denoise the result.
Never have i suggested that one should wait forever for renders to complete…
hey I can live with a blocky reflection pass, I’m just trying to help with reporting bugs.
Can I give you a file for you to have a play with? It’s a fierce monster that I think is probably a challenge for any rendering engine…I can’t show it here, but I’d be really interested to see if you can get a reasonable render time with retrace switched on (or even with default settings).
Sure, send it over, i’ll be too happy to give it a go!
At the very least i should be able to mention to you what are the main performance killers.
But what i’m trying to say is that it’s not a bug, it’s the expected behaviour for the LC in a set of situations, when retrace is off.
I’m actually unsure if we left the toggle in the UI for other specific cases that deal with final production rendering, or if it is a leftover.
I’ll investigate and report back.
Understood ![]()
I will send this file over sooner or later this week.
I asked Vlado for confirmation (he’s the author), it should indeed be always left on: the current implementation of the LC requires it so.
Then it would be best to get rid of the checkbox, as it evidently creates artifacts and confusion.
Or perhaps hide it with the basic/advanced UI modes.
Agreed on both points. ![]()