Octane 3.1 will include:
Cuda cross compiler, first implementation is CPU fallback mode
(Video include Octane running on AMD R9 Fury)
Octane Shader Compiler, default Open Shading Language
(API for any shader language) < Whatever that means
glTF instead of FBX
Blender for Octane 3.0 is announced too but development was always behind the big three.
There are much more info about Cloud render pipeline and so forth but these are my favorites.
I wonder if Octane’s adaptive sampling will have the same issue as the initial attempt by Lukas Stockner to code the same for Cycles.
What I’m talking about is the sampler being heavily biased towards brighter areas (so they converge much faster) while darker regions largely get ignored and converge real slowly (despite there being obvious noise). I ask because the noise pass there appears to have a rather strong correlation with how bright an area would be (as opposed to how many samples each area needs).
Then there’s the potential issue with fireflies, will their sampler get stuck in various spots as well?
I have been testing it on some of my scenes and it appears to improve how fast it converges. But I don’t have a definitive answer for you Ace. I could save noise passes at several points in the render if that would help determine if the algorithm is working or has issues? With adaptive sampling my images are much cleaner than without, at the same samples per pixel and the time it takes to get to those samples is faster.
Hi Ace, do you have an idea for a test .blend to get more information how the sampler work?
For setup you should use the border render to get how many samples you need for a clean image,
then use this sample setting for the whole image.
If I need 2 hours for 8000 samples without AS, render is finished in may 25 minutes with AS.
It does not help if the if the image is noisy all over, only some parts are noisy other not.
This seams logically for AS.