nVidia releases interactive raytracing engine

But what I'm questioning is, what makes this one more "interactive" than the others demonstrated so far? (though the openrt was cpu only i think, should have double checked that before posting it)
 
You might understand if you had a 4x Tesla C1060 PSC to play with. Guessing from NVidia's ray traversal paper, that's perhaps 160M to 800M ray traversals/sec depending on ray coherency?. That's just a 4 GPU system, now imagine an entire cluster of Tesla S1070 1Us ... that would be quite interesting indeed.
 
You might understand if you had a 4x Tesla C1060 PSC to play with. Guessing from NVidia's ray traversal paper, that's perhaps 160M to 800M ray traversals/sec depending on ray coherency?. That's just a 4 GPU system, now imagine an entire cluster of Tesla S1070 1Us ... that would be quite interesting indeed.

Are talking about the paper Understanding the Efficiency of Ray Traversal on GPUs by Timo Aila and Samuli Laine at NVIDIA Research? Those are quite impressive figures.

I wonder if you, since you are at siggraph, could do some digging?

One thing that Causic seems to be doing right is trying to extract coherency to enable parallellism, and the shading which is done on CPU (or GPU) will be done on coherent data. I.e. the CPU is able to run SIMD shader code on a number of samples at the same time.
Is nvidia trying to hanlde shading in the same way, or will the user get a bunch of "random" samples to work with? If so, is that a problem or can it be worked around?

Anothing thing is a tweet from Mental Images that seems interesting.
True photo-real rendering, either Mental Ray or using RealityServer? Go to NVIDIA booth - Barry or Patrick will show you iray®...

If you don't want to or don't have time that is cool as well. Have fun!
 
You might understand if you had a 4x Tesla C1060 PSC to play with. Guessing from NVidia's ray traversal paper, that's perhaps 160M to 800M ray traversals/sec depending on ray coherency?. That's just a 4 GPU system, now imagine an entire cluster of Tesla S1070 1Us ... that would be quite interesting indeed.
Yes, they might be able to start putting shiny metal balls in a Doom3 level and render it with RT raytracing :p
 
My guess, companies doing ray tracing have big budgets and nv dont want them buying geforces instead
 
wow, those billiard balls look amazing. Not sure if it's the depth of field or something else...but getting that kind of quality at a high frame rate is nice.
 
AVP flags it as a threat saying it detected HackTool.GUN. Also, I'm getting a missing MSVCP71.dll error.

Here is the source code of binpatch.exe, sorry i forgot to include it in .zip:
Code:
#include <iostream>
#include <fstream>
using namespace std;

ifstream::pos_type size;
char * memblock;

int hex2bin(char *s,char *o) {
  int r=0;
  while(strlen(s)>=2) {
    int b;
    sscanf(s,"%02X",&b);
    o[r++]=b;
    s+=2;
  }
  return r;
}

int main (int argc, char* argv[]) {
  if (argc<4) { cout << "Usage:\n patch file.exe 7504C2 EB04C2\n";return -1; }
  fstream file (argv[1], ios::in|ios::out|ios::binary|ios::ate);
  if (!file.is_open()) { cout << "Unable to open file";return -1; }

  size = file.tellg();
  memblock = new char [size];
  file.seekg (0, ios::beg);
  file.read (memblock, size);

//  cout << "the complete file content is in memory\n";

  char *s1 = new char [strlen(argv[2])/2];
  int l1 = hex2bin(argv[2],s1);
  char *s2 = new char [strlen(argv[3])/2];
  int l2 = hex2bin(argv[3],s2);
//  cout << "searching for " << l1 << " bytes";

  int p;
  for(p=0; p+l1 < size; p++) 
    if(!memcmp(memblock+p, s1, l1)) break;
  if (p+l1 == size) { cout << "Unable to find pattern";return -1; }

  cout << "found at: " << hex << p << "\n";

  file.seekg (p, ios::beg);
  file.write (s2, l2);

  cout << "patched\n";

  file.close();
  delete[] memblock;

  return 0;
}
 
Back
Top