Indigo Renderer Forum

Why does aperture diffraction fail when rendering above 2400 pixels in width? I have 32gb of ram, but it still says it runs out of buffer. Is there something that can be changed to increase the buffer?

-Lightrik

hey lightrik,

the Post Processed AD Filter is a RAM Killer for sure... Glare really NEEDS to optimize that!

So why are 32GB not enough ^^

- the PP AD takes your internal render resolution, not the output res.
So if you use a SuperSampling value of 3, it calculates that based on a 7200px width!
- In general the PP AD uses extensive amount of RAM, here a example:
a 320x240 scene using SS of 2 renders with 56MB of RAM.... enabling PP AD bumps that to 906 MB,
calculates that AD with only one core for 40 sec, and stays with that 906MB of RAM usage!
Each time the screen refreshes, the whole scene needs to be recalculated using 1 core for 40sec without rendering further doing so...
- Now I raise the SuperSampling of that 320x240 scene up to 3, eats 62MB of RAM... activating PP AD.... bad idea!!!
RAM usage raises up to 3.35GB !!!!!!!!!!!!!!! Calculation takes AGES... closed Indigo after some minutes...

Indigo has some physical AD, but afaik it is broken for BiDIr rendering... use it if you can!!

I attached the test scene so Glare has a proof...

btw I have fixed non-post pro A.D for bidir. Will be available in next release.

OnoSendai wrote:btw I have fixed non-post pro A.D for bidir. Will be available in next release.

Awesome news!!
Since it doesn't cost any extra RAM its great to work with... even if rendering is a little slower.

Any plans for the PP AD situation?
atm that feature is quite useless for most scenes and only gives (new) users the feeling that Indigo is a buggy piece of software...
Removing it would help Indigos reputation over the time more then leaving it the way it is!

Hi Zom-b,
I would like to improve it.
I'll try and take another look at it soon.

Would all developers process requests and critiques of customers that encouraging like you, the wouldn't be any bad software in this world, thank you

Thanks, that makes sense. I was using a SuperSampling value of 3, so I guess it was overloaded, but when I watch my ram usage I don't see it go much above 8 during the render startup. I did notice just before the AP calculation, it lists a buffer size:

Num buffer layers: 4
Master buffer size: 1.267 GB
Auxiliary buffer size: 693.443 MB
Creating diffraction filter image...
Done. (Elapsed: 17.39500 s)

This was a 2400 pixel width image, which was successful. Is this buffer related to the buffer that it mentions when AP fails? The buffer increases with the size of the image, but it still seems like it could be made larger.

Another question... My tests here were using AP checked in Skindigo, which activates normal AP, but it also activates Post-pro AP. Is there a benefit to manually changing Post-pro AP to false in the .igs file? Unchecking the AP box in Indigo seems to turn both off.

Just to chip in and say thanks for explaining that Zom-B, I was wondering why Aperture Diffraction kept crashing whenever I tried it... Although maybe my SuperSample level of 6 and 7MP image was to blame...

To OnoSendai, we're really grateful for the improvements....

OnoSendai wrote:I would like to improve it.
I'll try and take another look at it soon.

I'm don't really understand what that Paper is all about, but I have the "feeling" that it could help in PostPro AD too:
Adaptive Manifolds for Real-Time High-Dimensional Filtering