Light cache settings for stills

[coolhand78]

when looking through various tutorials, bought scenes etc everyone's light cache settings vary massively, i've seen some at the default 1000 and others at 6000 - so what is the rule of thumb when setting the subdivs on the light cache when used as the secondary engine (BF being the primary).

[joelly3d]

Not sure if it really matters, but i like to do half of the output resolution. That may not be necessary, but usually i just let renders for print go overnight anyway.

[glorybound]

I haven't messed with V-Ray settings for a year; nothing but defaults. Isn't V-Ray handling all that for you now?

[coolhand78]

I haven't messed with V-Ray settings for a year; nothing but defaults. Isn't V-Ray handling all that for you now?

as far as i'm aware not when it comes to the light cache...?
brute force as the primary engine - yes, but the secondary is still adjustable by default.

[glorybound]

Yeah, I don't touch anything. I guess you can play with it yo see if it matterS.

[JezUK]

I don't have the fastest computer myself, but I'm the same as Bobby, these days I use the defaults and hardly need to touch a thing (just control the Noise Threshold etc) using BF rather than playing around with settings in LC. The extra time BF might take to render is better that way than for me to play endlessly with settings in LC (and never getting anything better than I would anyway with BF).

[^Lele^]

As stated by Bobby and Jez, your best bet since 3.3 is to stick with the defaults, changing just noise threshold up and down to suit your preview needs, and changing the LC settings in the way the tooltips suggest for animation.
It's not only safe for results' quality, but it will more likely than not give you the very best in terms of performance towards a given target (ie. noise threshold).
You could think of the algorithms chosen by default as an early form of "AI", in that the intelligence put into them was wholly human, courtesy of the Chaos people, but also in the knowledge that they have been made adaptive and clever, above and beyond what they were pre-3.3 (premultiplied LC, adaptive lights, etc. etc.).
Any other approach, for GI engines, would have you at a clear speed disadvantage towards a given quality (ie. no comparing blurry irmap with multisampled BF.), as you'd likely end up parting from features, or combination thereof, which would be beneficial, well past the GI calculation phase.

[coolhand78]

As stated by Bobby and Jez, your best bet since 3.3 is to stick with the defaults, changing just noise threshold up and down to suit your preview needs, and changing the LC settings in the way the tooltips suggest for animation.
It's not only safe for results' quality, but it will more likely than not give you the very best in terms of performance towards a given target (ie. noise threshold).
You could think of the algorithms chosen by default as an early form of "AI", in that the intelligence put into them was wholly human, courtesy of the Chaos people, but also in the knowledge that they have been made adaptive and clever, above and beyond what they were pre-3.3 (premultiplied LC, adaptive lights, etc. etc.).
Any other approach, for GI engines, would have you at a clear speed disadvantage towards a given quality (ie. no comparing blurry irmap with multisampled BF.), as you'd likely end up parting from features, or combination thereof, which would be beneficial, well past the GI calculation phase.

Thanks for the reply mate, I was hoping you'd chime in.

Do you know of any instance where adjusting the light cache value would be beneficial? for stills that is...

[^Lele^]

Thanks for the reply mate, I was hoping you'd chime in.

I was notified you "mentioned me" in your post reply, so i guess typing coolhand78 may do the trick for you? If it does, feel free to "call" me.

Do you know of any instance where adjusting the light cache value would be beneficial? for stills that is...

If for whatever reason you couldn't use leak prevention, then maybe, yes.
Otherwise, LP will take care of increasing the number of samples cast based on, in essence, your geometric detail.
Feel free to try, maybe on a production-grade scene, if raising LC subdivs to 2, 3, 5k makes any difference overall: i'd wager any potential gains (now that retrace fixes issues) in LC sampling density would be offset by the LC calculation time, without much speeding the render up in the "final" tracing phase (i'm not a betting man, we checked. But do try! :P).
The raising of both subdivs and retrace amount for animations is meant as a safety net, should havoc happen, and to stabilise the LC solution over multiple frames (oversampling there would only avoid bubbling, or temporal incoherence, not raise spatial accuracy.).