Welcome to Fractal Forums

(http://pre13.deviantart.net/0ddc/th/pre/f/2015/225/c/0/burning_teraflops_by_lyc-d95jefh.png) (http://orig10.deviantart.net/3bed/f/2015/225/c/0/burning_teraflops_by_lyc-d95jefh.png)

(Click for full 4K resolution PNG image.)


Pretty sure I'm still doing it wrong:


Isn't 0.1 "fudge factor" really bad? Anyway, that's 3 bounces diffuse GI (there is a full material system, but only using diffuse/Lambertian), 8 minutes rendering on a single GTX 970 in 34 C ambient temperature  :fiery:

At some point I wish to beg the gods of distance estimation for help figuring out how I'm doing DE so badly...

(http://img13.deviantart.net/51ff/i/2015/228/4/0/portrait_of_a_julia_set_by_lyc-d95wxho.jpg) (http://orig10.deviantart.net/e308/f/2015/228/d/c/portrait_of_a_julia_set_by_lyc-d95wxho.jpg)

(Click for full 4K resolution high quality JPG image.)


5 bounces, 1024 samples per pixel, 5 minutes rendering.

That's fast; I'm still below two frames per hour. :-)

I think on a pair of overclocked Titan X this thing would be near realtime at 1080p if using a few spp and temporal reprojection.

The last image is just (?) 5 minutes, again on a single GTX 970 at stock clocks. I can't wait for winter to overclock this GPU, but goddamn is it hot/humid in Berlin at the moment...

pics look promising, as far as i understand, the fudge factor can be used to avoid overstepping, having a concrete distance estimation allows it for beeing 1.0 e.g simple geometry like balls/squares and some fractals deliver mostly correct distances, so stepping the amoung returned by the de is just fine

there has been an improvement algorithm to take bigger steps, which has eiffie visualised in this shadertoy and gives 10 percent stepping improvement
https://www.shadertoy.com/view/lsBSDK

I tried the SOR approach from cupe/mercury, but it's sort of tuned for screenspace not indirect rays... will give that another try later (since it's something like cone tracing with analytic LOD, which is great).

What's really needed for specular reflections is root refinement, which he also covers in the paper (http://erleuchtet.org/~cupe/permanent/enhanced_sphere_tracing.pdf).

(http://img06.deviantart.net/a7b9/i/2015/228/4/2/kaleidoscopic_surface_closeup_by_lyc-d95x949.jpg) (http://orig08.deviantart.net/2486/f/2015/228/6/9/kaleidoscopic_surface_closeup_by_lyc-d95x949.jpg)

(http://pre15.deviantart.net/253e/th/pre/i/2015/228/9/0/neutron_star_by_lyc-d95yjxv.jpg) (http://orig09.deviantart.net/fac6/f/2015/228/3/5/neutron_star_by_lyc-d95yjxv.jpg)

(deleted)

(http://img06.deviantart.net/8951/i/2015/233/a/9/quaternion_gnarl_by_lyc-d96klql.jpg) (http://orig10.deviantart.net/0ebc/f/2015/233/c/6/quaternion_gnarl_by_lyc-d96klql.jpg)

(Click for full 5760 x 3240 resolution high quality JPG image.)

(http://img03.deviantart.net/aaa5/i/2015/234/0/2/radioactive_julia_by_lyc-d96p7yc.jpg) (http://orig13.deviantart.net/eecd/f/2015/234/8/f/radioactive_julia_by_lyc-d96p7yc.jpg)

Those are weirdly pretty and pretty weird :D

(http://img09.deviantart.net/421f/i/2015/353/8/a/mandelbox_by_lyc-d9kptgi.jpg) (http://orig03.deviantart.net/5ff9/f/2015/353/d/d/mandelbox_by_lyc-d9kptgi.jpg)

Cool!

For a Kaleidoscopic IFS DE_scale should be set to 1 or a slightly less than that. Could we see KaleidoIcosaDE(p)'s implementation?

// http://www.fractalforums.com/3d-fractal-generation/kaleidoscopic-%28escape-time-ifs%29/

// Return reflection matrix for plane with normal 'n'
float3x3 reflectionMatrix(float3 n)
{
   const float3 n1_x = (float3)(1 - 2 * n.x * n.x, -2 * n.y * n.x, -2 * n.z * n.x);
   const float3 n1_y = (float3)(-2 * n.x * n.y, 1 - 2 * n.y * n.y, -2 * n.z * n.y);
   const float3 n1_z = (float3)(-2 * n.x * n.z, -2 * n.y * n.z, 1 - 2 * n.z * n.z);
}

float KaleidoIcosaDE(const float3 z0, float t)
{
   const float scale = 2.393331f;

   const float phi = 1.6180339887498948482045868343656f;

   const float3 offset = normalize((float3)(1, phi - 1, 0));

   //const float3 n1 = normalize((float3)(-phi, phi - 1, 1));
   //const float3 n2 = normalize((float3)(1, -phi, phi + 1));

   //const float3 offset = normalize((float3)(1, phi - 1, 0));
   //const float3 offset = normalize((float3)(1.01f, phi - 1 + 0.21f + m, -0.1f + q));

   const float3 n1 = normalize((float3)(-0.8652f, 0.752f, 1.25f));
   const float3 n2 = normalize((float3)(0.6152f, -0.23521f, 1.325f));

   //n1Mat = reflectionMatrix(n1);
   const float3 n1_x = (float3)(1 - 2 * n1.x * n1.x, -2 * n1.y * n1.x, -2 * n1.z * n1.x);
   const float3 n1_y = (float3)(-2 * n1.x * n1.y, 1 - 2 * n1.y * n1.y, -2 * n1.z * n1.y);
   const float3 n1_z = (float3)(-2 * n1.x * n1.z, -2 * n1.y * n1.z, 1 - 2 * n1.z * n1.z);
   //n2Mat = reflectionMatrix(n2);
   const float3 n2_x = (float3)(1 - 2 * n2.x * n2.x, -2 * n2.y * n2.x, -2 * n2.z * n2.x);
   const float3 n2_y = (float3)(-2 * n2.x * n2.y, 1 - 2 * n2.y * n2.y, -2 * n2.z * n2.y);
   const float3 n2_z = (float3)(-2 * n2.x * n2.z, -2 * n2.y * n2.z, 1 - 2 * n2.z * n2.z);


   const float bailout2 = 1e9f; // 1e32f
   const int max_iterations = 22; // 28

   int n = 0;
   float3 p = z0;
   while (n < max_iterations)
   {
      p.y = fabs(p.y);
      p.z = fabs(p.z);

      if (dot(p, n2) > 0)
      {
         //p *= n2Mat;
         p = (float3)(dot(p, n2_x), dot(p, n2_y), dot(p, n2_z));
      }

      p.x = fabs(p.x);
      p.z = fabs(p.z);

      if (dot(p, n1) > 0)
      {
         //p *= n1Mat;
         p = (float3)(dot(p, n1_x), dot(p, n1_y), dot(p, n1_z));
      }

      p = p * scale - offset * (scale - 1);

      if (dot(p, p) > bailout2) break;

      n++;
   }

   return length(p) * pow(scale, (float)-n);
}

You will need to do:
That's because you bail out before incrementing n.
Of course you can increment n before testing for bailout instead.

A slight scaling down (by a factor say .99) would be necessary to avoid some overstepping. Another way to avoid that slight overstepping is to subtract the radius of the bounding sphere from p:

In order to save some CPU cycles:
or
are faster than using matrices.

BTW! happy holydays.  :)

What a huge speed difference that makes :D Thanks for this, I had some fun over the holidays with it:

(http://pre11.deviantart.net/a573/th/pre/f/2016/007/7/5/kaleidoscopic_ii_by_lyc-d9n5qed.jpg) (http://orig07.deviantart.net/af94/f/2016/007/7/5/kaleidoscopic_ii_by_lyc-d9n5qed.jpg)

Happy new year to you too mate!

just a quick quaternion julia with some motion blur in a little toy gpu ray tracer, click for 4K res:

(http://pre02.deviantart.net/7985/th/pre/f/2016/200/f/e/dark_julia_by_lyc-daama04.jpg) (http://orig10.deviantart.net/520b/f/2016/200/f/e/dark_julia_by_lyc-daama04.jpg)

another simple quaternion julia, to exercise the new gtx 1070. click for 8K res:

(http://pre14.deviantart.net/e112/th/pre/f/2016/215/2/b/dark_julia_iii_by_lyc-dachada.jpg) (http://orig05.deviantart.net/01bb/f/2016/215/2/b/dark_julia_iii_by_lyc-dachada.jpg)

menger sponge, thanks Aexion for the DE formula! sorry for the noise and over-saturation. click for 4K res:

(http://pre11.deviantart.net/e8b8/th/pre/f/2016/217/9/d/menger_sponge_by_lyc-dacopkb.jpg) (http://orig14.deviantart.net/ddcb/f/2016/217/9/d/menger_sponge_by_lyc-dacopkb.jpg)

more stuff... click for high res

(http://pre13.deviantart.net/fb46/th/pre/i/2016/227/1/e/kaleidoscopic_iii_by_lyc-dae1dj5.jpg) (http://orig04.deviantart.net/0eed/f/2016/227/a/a/kaleidoscopic_iii_by_lyc-dae1dj5.jpg)
(thanks again to Knighty!)

(http://pre05.deviantart.net/037c/th/pre/f/2016/227/c/2/kleinian_stuff_by_lyc-dae1bu8.jpg) (http://orig15.deviantart.net/c4dd/f/2016/227/c/2/kleinian_stuff_by_lyc-dae1bu8.jpg)
(from the Fragmentarium examples)

(http://pre05.deviantart.net/36ac/th/pre/f/2016/227/b/b/portrait_of_a_julia_set_ii_by_lyc-dae1c9u.jpg) (http://orig06.deviantart.net/cd4f/f/2016/227/b/b/portrait_of_a_julia_set_ii_by_lyc-dae1c9u.jpg)


some experimental stuff using DEs from Aexion:

(http://pre03.deviantart.net/7535/th/pre/f/2016/227/5/7/aexion_box_thingy_by_lyc-dae1c21.jpg) (http://orig06.deviantart.net/5bc7/f/2016/227/5/7/aexion_box_thingy_by_lyc-dae1c21.jpg)

(http://pre04.deviantart.net/3936/th/pre/i/2016/227/d/8/aexion_box_thingy_2_by_lyc-dae1c5q.jpg) (http://orig14.deviantart.net/2dc5/f/2016/227/1/f/aexion_box_thingy_2_by_lyc-dae1c5q.jpg)

8 minutes rendering, 8 bounces diffuse gi, gtx 1070 @ 2ghz (click for 8K)

Edit: removed image, updated with higher quality below

At that speed, you could conceivably render animations at that same extreme image and lighting quality. Those should end up looking very real.

 :o
 :wow: :thumbsup1: :w00t:
 :worm: :worm: :worm:

glad you like it knighty, you're the hero of the show here with the awesome kaleidoscopic IFS DE :) i'd be honoured to render any other fun stuff you'd like to suggest :D

hobold, what usually happens in this situation is, i do a deeper zoom / crank up the visual quality until it needs an overnight render ;)

updated overnight render, click for 8K:

(http://pre09.deviantart.net/a11d/th/pre/f/2016/228/8/5/kaleidoscopic_iv_by_lyc-dae2843.jpg) (http://orig10.deviantart.net/df25/f/2016/228/8/5/kaleidoscopic_iv_by_lyc-dae2843.jpg)

Video plz.
 :spgloomy:

Sorry for the delay, I'm upgrading my little shader editing thing into a full animation editing studio thing :) I would love to share it with you (privately) someday!