LSM, GI/AA and noise questions

[subpixelsk]

I´m trying to optimize my interior scene to achieve noise free result in shortest possible time and have searched throughout this forum for info and answers but I still haven´t found answer that would satisfy my needs

Let´s say I have an interior and am approaching it usually to determine whether noise is coming from direct or indirect lighting - tests showed much more noise in direct lighting (indirect pass is noise free very quickly) and I ended up with LSM raised to 16 - and here is the question - how high can I go with this number and why is it recommended to keep it at 4 maximum when at 4 I´m still getting much more noise in direct lighting than indirect (when observing stats I also get much more actual rays when raising LSM to 16 - almost doubled - but I don´t know if this number is something I have to keep my eye onto).

So what is the reason to keep GI/AA between 2-64 and LSM up to 4?

Thanks

[Ludvik Koutny]

Hi,

first of all, where did you find out that you should not go higher than 4 with LSM?

Then also, if most of the scene noise is in direct light, and it's light entering through some opening, then that's ideal scenario to use portals. Or, if it's CoronaLight, there could be some bug causing excessive direct light noise. There is one know bug causing excessive direct light noise when using disc light shape.

[subpixelsk]

It is recommendation when hovering over with mouse in render settings so I thought it´s for some reason not to go higher than that. I just wanted to be sure I´m on the right track

I´m getting a lot of direct light noise from Corona LightMtl used as led light stripes but I think solution is just to let it render longer isn´t it?

[Ludvik Koutny]

I think we should update the tooltip to mention not to go higher than 8, instead of 4. But if values higher than 8 still improve noise, then it's probably overcomplicated scene lighting and something should be done about that. Maybe if you posted images of your scene setup, that could help a bit. Also make sure you did not exceed 250 light groups limit. You can find that in CoronaVFB stats.

[subpixelsk]

Sure I will upload it when I will be back on my computer, generally it is very simple scene - only hdri with some correctly set up backplate -  and some led lights created with lightmtl - these are causing the most noisy areas . For sure I have not more than 250 light groups. So more than 8 LSM brings no noise reduction improvement when scene is set up correctly, right?

Thanks for any advice

[Ludvik Koutny]

So more than 8 LSM brings no noise reduction improvement when scene is set up correctly, right?

Yes... Well, it would still improve quality of direct light, but at the significant expense of indirect light for example. So your GI would get more noisy. If LSM at extreme values improves direct light quality and does not worsen anything else, then that's the situation to be worried about.

[subpixelsk]

I see.. and is there any difference regarding noise between corona light and lightMTL? or are they on par with each other?

[Ludvik Koutny]

Shouldn't be as long as you don't use some high poly or crazy meshes. Like several light meshes each in different room of an appartment, yet attached toghether into one object. So if you have 5 light meshes in 5 different rooms attached in one, they will be sampled as one light. So in each room, light mesh will get only 1/5th of the sampling it should get. You should attach them together, but smartly, if you have row of 50 LEDs in same room next to each other, you should attach those together for example.

[subpixelsk]

I´ve attached my scene with region test highlighted with red -  first one LSM 8 and second one 16 - not really much difference in those two however noise level still high after 15 min. of small region which gave me 250-300 passes. Problematic are also the wall with led lighting on the right etc. Any help appreciated (portals slow down rendering significantly)

[subpixelsk]

actually I did one more test to check possible cause of noise - replacing corona light mtl geometry with regular corona light - after 3 minutes you can see noticable less noise with corona light - is this a bug? ( I know I did not match the intensity quite well but I don´t know if that would make a difference)

[maru]

replacing corona light mtl geometry with regular corona light - after 3 minutes you can see noticable less noise with corona light - is this a bug?

This would be a bug unless you are using light mtl on geometry that is totally different than the Corona light object. I would also suspect reflective caustics in this case. You would be able to reduce them by decreasing MSI value or using rayswitch material on reflective objects (like the floor). Could you post _direct and _indirect passes?

[subpixelsk]

the geometry was actually the same as corona light - only a plane with no thickness