min samples %

[paoloverona]

I must admit that sometimes I feel really a donkey : it's easly understandable what the "noise treshold" is for, but not the same thing as for "min samples%", I tried changing this value but really not able to understand what affects.
Can someone be helpful?

[oguzbir]

noise treshold is the value you tell renderer to stop converging noise at that treshold.

Minimum samples is the percent to tell renderer when will the adaptive sampling kicks in...
So if you choose a higher value. Render will continue sampling like bruteforce... When the percentage is met.. Then it will start adaptive sampling after this value.

If you lower the value render will be more speedy but you may spot problems because early start of the adaptive sampling.
In normal cases if you set it at %5 like reasonable values. Render wll have enough information to start adaptive sampling.

[paoloverona]

Thanks a lot for the info oguzbir, I apologize for the late reply but I wanted to make some test based on what you said but now I'm more confused: here some examples made, and except the one with the shortest render time, all the others have no difference at all in noise cleaning, even changing completely the two values.
There is something wrong in my test?, too small resolution or too few samples?

[oguzbir]

I'm not sure as well. You better ask this to Karba. Dont want to mislead. And I'm curisous as well..
By the way the last 3 images looks identical.

[paoloverona]

Thanks a lot for the info oguzbir, I apologize for the late reply but I wanted to make some test based on what you said but now I'm more confused: here some examples made, and except the one with the shortest render time, all the others have no difference at all in noise cleaning, even changing completely the two values.
There is something wrong in my test?, too small resolution or too few samples?

yes, if you overlap the three photos in photoshop you can see that there is no difference at all between them. The settings was 3000samples for each and the time was
1min15sec 0,5 noise treshold and 0,1 min samples
24min31sec 0,001 noise treshold and 0,1 min samples
24min36sec 0,001 noise treshold and 100 min samples
24min42sec 0,5 noise treshold and 100 min samples

[Miguel1900]

Paolo, I'm a beginner too, but I think that you put a noise thresold soooooo low that it doesn't care if you put 0,1% samples or 100%, because the "noise points" are already too small for that resolution. Maybe those same tries but for a resolution of 10.000 pixels would show other results.

I think you should try with more visually detectable values, no so microscopic. Something like 0,01 thresold and 0,1%; 0,5 and 1%, etc.

[oguzbir]

Main determiner is the %min samples.. 0.1 kicks speed up nearly at the start of the render. While the render is very noisy.
As miguel says if you set the noise treshold at a lowest value. there is little nothing adaptive sampling can do.

As for this setup... 1min15sec 0,5 noise treshold and 0,1 min samples The result is noisy because threshold is way too much.
0.5 might mean saying "I accept noisy final render"
Do not feel free to set those values to extremes.
Maybe, theorotically, you might set between 0.01 - 0.1 as threshold and between 0.5 - 2 & min samples.
And if you want to speed up the render whilst keeping the noise low is best to buy another gpu
cheers,

[Karba]

Main determiner is the %min samples.. 0.1 kicks speed up nearly at the start of the render. While the render is very noisy.
As miguel says if you set the noise treshold at a lowest value. there is little nothing adaptive sampling can do.

As for this setup... 1min15sec 0,5 noise treshold and 0,1 min samples The result is noisy because threshold is way too much.
0.5 might mean saying "I accept noisy final render"
Do not feel free to set those values to extremes.
Maybe, theorotically, you might set between 0.01 - 0.1 as threshold and between 0.5 - 2 & min samples.
And if you want to speed up the render whilst keeping the noise low is best to buy another gpu
cheers,

This is correct.
Min samples can be set as low as possible until you start losing render details like tiny highlights.

[paoloverona]

woow!, what an intresting discussion here!
I think it's quite normal, when it's not known the mechanism of working or the behaviour of a parameter, to test it at his extreme values (because "low" variation may not be perceptible), that's why I decided to use such values.
when I'll have more time to dedicate on it, I'll do more tests: my aim is to be able to use these values consciously. As for "speeding up the render", I'm totally satisfied with what FStorm is able to do with just my GTX980Ti so for now, no need of another GPU