Mandelbuld 3D v1.8.6 beta test

[lenord]

Oh Boy, doin the Happy Dance now

[bib]

Test MC render in 600x450 on a corei5 laptop. Left image: normal render, 17 seconds. Right image: MC, 45 minutes.

[cKleinhuis]

that mc rendering combined with the 16 bit heightmaps will keep us busy for all over the year
woot, i wonder what amazing piccies will be created ... !!!! crazy stuff!

[taurus]

Impressive demo scene bib! really looks like antialiased

[bib]

Thanks taurus. Even without MC, the softer HS is a fantastic addition.

ck: I agree that the potential for creativity and ultra realistic pictures will be huge by combining MC with soft HS, reflections/transparency, heightmaps,...

[lycium]

Seeing these super slow renders reminds me of the brute force monte carlo GI renders I did of the original mandelbulb, on a dualcore AMD back in 2009.

I really need to get some distance estimators into Indigo, hopefully with Knighty's help!

[Jesse]

Just wanted to mention that i have updated to beta-3 with most fixes...
please still report all issues with it so i have not to change the final release that often. 
 

[knighty]

Jesse: Just curious, what random number generator are you using?

I really need to get some distance estimators into Indigo, hopefully with Knighty's help!

I'll try my best. That would be cool.

[Jesse]

Jesse: Just curious, what random number generator are you using?

The fastest i found so far, dunno if its the best one:

Code:

function TMCCalcThread.GetRand: Double;
const dm: Double = 1 / $7FFFFF;
{  seed := 214013 * seed + 2531011;
    Result := ((seed shr 8) and $7FFFFF) * dm; }
asm
  imul edx, [eax + seed], $343FD
  add  edx, $269EC3
  mov  [eax + seed], edx
  shr  edx, 8
  and  edx, $7FFFFF
  push edx
  fild dword [esp]
  fmul dm
  pop  edx
end;

Seed is initialized for each thread independent with its Thread ID and the built in random generator.
.. with Seed as Integer.

[Teena]

Absolutely great render comparison, bib!

Jesse, I can't change the term 'Author' to my name. I started a new fractal from scratch just to test this new option. I clicked the Author button and replaced 'Author' with my forum name. When I clicked the Insert button it inserted 'Author' instead of my name. Am I doing something wrong or is it a bug?

[Jesse]

Absolutely great render comparison, bib!

Jesse, I can't change the term 'Author' to my name. I started a new fractal from scratch just to test this new option. I clicked the Author button and replaced 'Author' with my forum name. When I clicked the Insert button it inserted 'Author' instead of my name. Am I doing something wrong or is it a bug?

Hmm, seems there is a bug, maybe you must restart the program when changed, i will fix it...

i think the only way by now is to edit the Mandelbulb3D.ini textfile and add 2 spaces plus your name behind "Author"

[Teena]

Hmm, seems there is a bug, maybe you must restart the program when changed, i will fix it...

i think the only way by now is to edit the Mandelbulb3D.ini textfile and add 2 spaces plus your name behind "Author"

That did the trick. Thank you very much! However, when you've fixed it I will check it again to confirm it.

[knighty]

The fastest i found so far, dunno if its the best one:

You maybe need to try a better RNG In order to make the rendering converge / reduce noise level. The RNG quality (periodicity, uniformity and such) can have a huge impact. I'm not an expert though, maybe Lycium could give us better insight.

[Jesse]

You maybe need to try a better RNG In order to make the rendering converge / reduce noise level. The RNG quality (periodicity, uniformity and such) can have a huge impact. I'm not an expert though, maybe Lycium could give us better insight.

I will give it a test in comparison with the built in RNG, rendering something up to same average raycount.

Btw, the code can be simplified further by omitting the shr edx, 8 and adjusting the norm factor.
Guess this was done because the lower bits are not that uniform... but 23 bits or so would be more than enough i guess.

Of course the magicians mathematicians decisions should be preferred...

...ähh, did some testing and can't see any differences (beside the speed advantage) in the results of both RNG's.
Tested with a 6000x3000 pixel background map with a one-pixel white spot on dark background and performed a DoF on it.
The result is a noisy disc bokeh that resamples the randomness quite well...
but as the raycount is limited to 65535, i don't think that a very high quality for the RNG is neccessary at all.
Those special cases where this would have been taken into account are not realistic for a CPU renderer, not if you want to wait a month for the result...

[Alef]

When trying to do buddhabrots I searched number of articles about random numbers and found some article claiming that best algorithm is quasi random number generator, cos it don't pretends of being random but instead concentrates on the most equal distribution possible, and someone proved, that quasi random number generator could be expanded to 360 dimensions for doing some sort of financial analysis.
Throught I newer looked deeper.


Doing google search I found few links. Knighty's quasi-random buddhabrots indeed looks smoother than random.

I'm not an expert though, ...

Here is thread by Knighty:
http://www.fractalforums.com/programming/buddhabrot-montecarlo-and-quasi-montecarlo/?PHPSESSID=b140a5bcfb93304fed816b4a0de8d24b
And then this article:
http://en.wikipedia.org/wiki/Quasi-Monte_Carlo_method


Still, I wount be soon, when I 'll see monte carlo in action, cos programm don't works on Windows 2000, and probably a compiler is a reason why programm don't works on Win2K