END OF AN ERA, FRACTALFORUMS.COM IS CONTINUED ON FRACTALFORUMS.ORG

Basically I work in the area of generative art. I use the GPU for realtime rendering and very rarely use fractals, they struck me as pretty boring. But two weeks ago I found information about mandelbox... Wow!  I think it is a very interesting approach... So I applied this algorithm in my existing engine and .. look at the result:



The speed is not too impressive - about 7fps on 1920x1080 GTX480 but I'm waiting for a new generation of videocards.

www.sanbasestudio.com

Yes, I want to have 30fps

This is demo:
<a href="http://www.youtube.com/v/z18omLWzChE&rel=1&fs=1&hd=1" target="_blank">http://www.youtube.com/v/z18omLWzChE&rel=1&fs=1&hd=1</a>

Depends on the context, but for smooth animation in realtime around 25fps is a generally accepted minimum, though a lot of games in the past used to get away with as low as 10fps and IMHO anything less than that is not "realtime" but just "interactive"

Very nice images - I like the final one in particular.

Excellent work ! Any chance of releasing the binary ? (or even the source

I have an awful lot of catching up to do with respect to 3D hardware, I'm working on it though.
At the moment I'm just trying a combined diamond-square/perlin noise algorithm in the hope that once I've got it fast enough simply rendering in 2D to a texture I can then use an adapted form of DE to give a realtime 3D landscape renderer without any polygons at all, though if that's asking a little too much I could easily convert to using a multi-resolution mesh method instead
Am still working on the design for the initial format of a UF-like GPU program - I'm deliberately sticking to a format that will work on iPad/iPhone (and other mobile devices if anyone wants to convert from iOS) as well as OSX (and Win7/Linux if anyone wants to convert).

Unfortunatelly it requires VERY powerfull system...
take a look: http://www.sanbasestudio.com/system.html

I can show the formula (first image):

vec4 sphere(vec4 z)
{
   float r2 = dot(z.xyz,z.xyz);
   if(r2<2.0)
      z*=(1.0/r2);
   else   z*=0.5;

   return z;

}
vec3 box(vec3 z)
{
   return clamp(z, -1.0, 1.0) * 2.0 - z;
}

float DE0(vec3 pos)
{
   vec3 z=pos-from;
   float r=dot(pos-from,pos-from)*pow(length(z),2.0);
   return (1.0-smoothstep(0.0,0.01,r))*0.01;
}

float DE2(vec3 pos)
{

   vec4 scale = -20*0.272321;
     vec4 p = vec4(pos,1.0), p0 = p;  
   vec4 c=vec4(param[31].w,param[32].w,param[33].w,0.5)-0.5; // param = 0..1

     for (float i=0;i<10; i++)
   {
      p.xyz=box(p.xyz);
      p=sphere(p);
      p=p*scale+c;
     }

   return length(p.xyz)/p.w;
}

float DE(vec3 pos)
{

   float d0=DE0(pos);   
   float d2=DE2(pos);

   return max(d0,d2);
}

 
Record:
  <a href="http://www.youtube.com/v/RSjWB21PJfs&rel=1&fs=1&hd=1" target="_blank">http://www.youtube.com/v/RSjWB21PJfs&rel=1&fs=1&hd=1</a>


Concerning:

Take a look at some of my works:

<a href="http://www.youtube.com/v/lkTJL7PZdNY&rel=1&fs=1&hd=1" target="_blank">http://www.youtube.com/v/lkTJL7PZdNY&rel=1&fs=1&hd=1</a>

<a href="http://www.youtube.com/v/M-zxQ7tFncU&rel=1&fs=1&hd=1" target="_blank">http://www.youtube.com/v/M-zxQ7tFncU&rel=1&fs=1&hd=1</a>

<a href="http://www.youtube.com/v/F7LE7W3n5BI&rel=1&fs=1&hd=1" target="_blank">http://www.youtube.com/v/F7LE7W3n5BI&rel=1&fs=1&hd=1</a>

OK, I'm new to shader code but so far I've found nothing to indicate that the OpenCL compilers do auto-inlining - am I missing something ?

Thanks, given that info I will be doing a test shader fragment to check if inlining is done automatically on the devices I'm interested in (and to what extent) - to be honest I never even trust the documentation with respect to this sort of thing because it's often out of date or even completely wrong - in the past that even goes for documentation from Intel, and for many others it's nothing unusual