understanding non progressive cycles rendering

Hi all,
I’m learning blender&cycles and I’m studying how to use the non progressive render to cut rendertimes.
Reading the DingTo post in his website I’ve understood that AA samples multiply the “channels” samples (diffuse, glossy, ecc).

So if I set in PROGRESSIVE mode 2000 samples and I’m happy with the noise level, if I want to recreate the same quality in NO PROGRESSIVE I’ve to set AA & channels samples in a way to have 2000 samples as result. Is it correct??

The light samples are multiplied by the AA samples??

2000 AA samples would give the same result, but you aren’t getting any use out of the NP integrator at that point. The theory behind the NP integrator is that things like GI, reflections, AO, and so forth require more samples than is need to actually get smooth edges. So instead of sending a single ray to sample these over and over again, you send a handful of primary rays out for each pixel, and once these hit something they split into multiple rays to help average out that sample. This reduces the number of times you need to (unnecessarily) calculate the camera > first-hit path.

What you want to do instead is set a basic number of AA samples so you get clean (not noisy or jagged) edges on things. This will probably be somewhere between 16-100 depending on the scene. Then look at where you still have noise, and up the per-ray samples for that type. For example, if you get clean edges with 30 AA samples, but notice you still have really grainy reflections, you’d increase glossy samples.

The wiki for Arnold has a really handy flowchart for this, the settings in Cycles are relatively similar for the most part: https://support.solidangle.com/display/ARP/Removing+Noise+Workflow

thanks J_the_Ninja, tomorrow I’ll try and I’ll post some images.
I have a scene where the ceiling of a room receive only indirect lighting and I have difficult to clean the GI noise.
And in addition of this, I have different behaviour from the spot lights when I use PROGRESSIVE and NO PROGRESSIVE.
Anyway with images it’s simpler to speak about these problems.

As I wrote I’m showing you my problems.

https://drive.google.com/folderview?id=0B5J7MJ7idXJgUDRHeUtRcm5UQ2s&usp=sharing

I explain the different images.
The images with “01” prefix show the problems I have and in the “02” images you can see in detail the different brightness in the non-progressive version (probably the reason I have higher rendertime and noise level).
In those images I used 127 spotlights with a size of 15cm (angle 60° or 95° for different lamps with a 0.6-0.8 of blend value)

The images with “03” prefix show a test I made to check if the problem was in the spotlights I used or not.
As you can see the problems of the noise level and high rendertime still remain.
In those images I used 62 area lights with a size of 35cm

For both versions, in the “film” panel, I have an exposure value of 7 to adjust the general brightness.

I hope it’s clear…and I hope to find a solution because the opportunity to optimize rendertime give me more flexibility with the deadline.

I’m trying to achieve better rendertimes with branched path tracing (ex non-progressive) but I’m always slower to path tracing with a medium heavy and complex scene.
It could be really good to have a scene to test all together to go deeper with this mode because at the moment for my needs I think it’s very un-usefull.
If someone want to share some tests can use this thread to show us the progresses… :slight_smile:

Thumbs up!

shameless self-promotion

you are right but I’m not a BlenderCookie citizen so i can’t see your video.

is it possible to show an example of a “non balls/suzannes scene” with the path tracing version and the branched one?

and in particular if you have an interior test.
I’m using the country kitchen scene from blendswap to make some test and , at moment, I can achieve visually better results with path tracing…

thanks in advance.

As written above, I’ve done some tests with a “non spheres/suzannes scene” and if you are interested here a zip file with some results.
The filename are self-explanary.

https://docs.google.com/file/d/0B5J7MJ7idXJgNDlHR1QzRVJ0VnM/edit?usp=sharing

The most interesting is AA350_D16_G8_T2_AO1_ML16_S1_Sobol_Bounces_4_0__D16_G1_T4_time_1h_31min_37sec

The “progressive” version with 5600 samples took 1h 56min (the glasses are black due to a too low transmission bounces) and is better as noise both in diffuse areas and glossy areas but the “non-progressive” one is good and acceptable.

What do you think?
The rendertime gain is good.
Do you think it’s possible to do better with the same quality??

if I modify my thread you probably will never notice the change (at least, I see this post as an old thread).