Noise limit without adaptivity

[romullus]

I hope this request make sense. I would love to see noise level and noise level limit available even without adaptivity. Is this possible?

[pokoy]

+1

[Ondra]

well... it would make our code simpler actually, but may I still ask... why? Do you really disable adaptivity for some serious renders? and if yes, why?

[romullus]

Nope, can't remember the last time i turned adaptivity off for other but testing purposes. Maybe my request makes little sense, but if it's possible and won't take much effort to implement, then why not? :] Besides, i find noise level limit very useful and convenient and if for some reason i feel need to disable adaptivity, i'd like to retain ability to end render process by noise amount.

[Frood]

Reasons for my +1:

• Only when noise level (NL) is reported also without adaptivity it can be used as stop condition for reproducable results using DR vs. no DR. May get obsolet when slaves get back their adaptivity if ever (don´t know what´s projected here). And I may be also wrong with this because NL calculation turns out to be very different using DR? To be tested.
• I like all kinds of render statistics :)

then why not?

Hmm... NL calculation still eats a significant amout of performance so it should really only be reported when using noise level as stop condition. That´s the only "why not" I can think of.

Good Luck

[Ondra]

Hmm... NL calculation still eats a significant amout of performance

Did you measure it? can you give us some numbers? We would be interested in it, because it seems the overhead is really small here

[Frood]

Last time I clearly noticed it (and tested it) was when default recalc interval was discussed. That time my brows went to the ceiling and I knew why it´s not calculated every pass because the performance loss was obvious. Not seriously testet since then, but I can do some tests on the side.

Good Luck

[Ondra]

still it would be interesting to get some numbers - how much time at what resolution does it approximately takes.

In other news, calculating the noise level even without adaptivity would take same extra memory as adaptivity takes (1 extra pass), so we wont be doing it

[Frood]

(...) would take same extra memory (...)

yepp, but when in the long run #passes will/should/could be replaced by image quality (=noise level) then calculating noise level at least when it´s a stop condition still makes sense - even without adaptivity in use.

Good Luck

[Ondra]

that would make sense, but the central argument is that you should just not disable adaptivity ;)

[Frood]

Then reactivate that adaptivity on slaves again and make it work ;)

Good Luck