Page 1 of 1

Caustics cache strange variation

Posted: Mon Jul 26, 2021 9:18 am
by Sharlybg
Hi just testing cautics cache on different scene size and just catch a weird behaviour :

The pool brightness vary according to scene size.

Smaller ground plane give darker pool :
pool_1.jpg

big ground plane give brighter pool :
pool_2.jpg

Here is the blendfile: you can just increase the ground plane by 10X to see the difference:

pool.blend
(1.96 MiB) Downloaded 123 times
[/size]

Re: Caustics cache strange variation

Posted: Mon Jul 26, 2021 11:36 am
by Dade
The total size of the scene affects where the rays for sun, sky and infinite lights are emitted from (they are emitted from the bounding sphere of the scene). Also floating point numbers have less and less precision farther you are from the origin. This is likely to be the source of your problem.

Metropolis is likely to start to have numerical precision problems: it will have to hit a smaller and smaller pool, larger the scene is. Quoting myself from another post:
Something that works at a scale X or position XYZ, will not work if scale up or translated far from the origin.

This is a rendering where you can literally see the floating point bits, near the origin:

Image

far from the origin:

Image

even father away:

Image

Quite insane, isn't it ?

Images courtesy of Matt Pharr article: https://pharr.org/matt/blog/2018/03/02/ ... space.html

Re: Caustics cache strange variation

Posted: Mon Jul 26, 2021 1:18 pm
by Sharlybg
Hmm that well explain.
Thanks.

Re: Caustics cache strange variation

Posted: Wed Jul 28, 2021 7:06 pm
by FarbigeWelt
Dade wrote: Mon Jul 26, 2021 11:36 am Quoting myself from another post:
Something that works at a scale X or position XYZ, will not work if scale up or translated far from the origin.

This is a rendering where you can literally see the floating point bits, near the origin:

even father away:

Image

Quite insane, isn't it ?

Images courtesy of Matt Pharr article: https://pharr.org/matt/blog/2018/03/02/ ... space.html
That is probably mostly a not intended effect, is it?
But, I truly like it! :geek:

However, 64 bit precision would be appreciated sometimes.
If only performance costs did not more than double or quadruple at most.

Re: Caustics cache strange variation

Posted: Wed Jul 28, 2021 9:31 pm
by Dade
FarbigeWelt wrote: Wed Jul 28, 2021 7:06 pm However, 64 bit precision would be appreciated sometimes.
If only performance costs did not more than double or quadruple at most.
I think the reason we will never really have a transition to FP64 is more because of doubling memory (and bandwidth) usage is too impractical than the performance hit.

But we are literally rendering images with Lego (aka bits) at the moment.

Re: Caustics cache strange variation

Posted: Fri Jul 30, 2021 5:37 am
by FarbigeWelt
Dade wrote: Wed Jul 28, 2021 9:31 pm
But we are literally rendering images with Lego (aka bits) at the moment.
Well, this is probably closer to reality than many people are aware of.

"One can truly think our World is built on Cubes - Cubes are as a matter of mathematically proven fact the most probable shape out of all kind of fragmentation shapes - at least in our three dimensional material word.", 2021, FarbigeWelt


What? :o Why? :shock:
Please, have a look at following paper.

Plato’s cube and the natural geometry of fragmentation
Gábor Domokos,
View ORCID Profile
Douglas J. Jerolmack,
View ORCID Profile
, and
View ORCID Profile
János Török
1. aMTA-BME Morphodynamics Research Group, Budapest University of Technology and Economics, 1111 Budapest, Hungary;
2. bDepartment of Mechanics, Materials and Structure, Budapest University of Technology and Economics, 1111 Budapest, Hungary;
3. cDepartment of Earth and Environmental Science, University of Pennsylvania, Philadelphia, PA 19104;
4. dMechanical Engineering and Applied Mechanics, University of Pennsylvania, Philadelphia, PA 19104;
5. eDepartment of Theoretical Physics, University of Debrecen, 4032 Debrecen, Hungary;
6. fDepartment of Theoretical Physics, Budapest University of Technology and Economics, 1111 Budapest, Hungary

PNAS August 4, 2020 117 (31) 18178-18185; first published July 17, 2020; https://doi.org/10.1073/pnas.2001037117
Edited by David A. Weitz, Harvard University, Cambridge, MA, and approved May 27, 2020 (received for review January 17, 2020)

Re: Caustics cache strange variation

Posted: Fri Jul 30, 2021 7:38 am
by Sharlybg
But we are literally rendering images with Lego (aka bits) at the moment.
Make me wonder if cacheless caustic solution will also suffer the first post issue :?:

Re: Caustics cache strange variation

Posted: Fri Jul 30, 2021 11:29 am
by Dade
Sharlybg wrote: Fri Jul 30, 2021 7:38 am
But we are literally rendering images with Lego (aka bits) at the moment.
Make me wonder if cacheless caustic solution will also suffer the first post issue :?:
Place the camera under water, disable cache, enable light tracing and you can check (I don't know the outcome).