SmallptGPU (OpenCL) Video 2

A simple Path tracer demo written in OpenCL. Running at 640x480 on an Intel Q6600 with an ATI HD4870 as GPU. The sources are available here: http://davibu.interfree.it/opencl/smallptgpu/smallptGPU.html

The old version of this video is available here: http://www.youtube.com/watch?v=TAZsC3buDug

Category:

Science & Technology

Tags:

• OpenCL
• ray tracing
• rendering
• path tracing

[saasrus]

Can smb tell me how to increase reflection limit in small pt gpu 2? I tried to edit scenes and built parallel mirrors, but only 8 reflections render, and black abyss ... ((

[Dade916]

@saasrus: you have to edit the .cfg file and change/add the following line:

path.maxdepth = 12

to increase the max. path depth up to 12.

[saasrus]

@Dade916 there is no .cfg in folder ( maybe i watching in wrong folder? it seems to me all files are in folder i unpacked smallptgpu. list: folder "scenes" some .bat files to run smallptGPU.exe smallptGPU.exe LICENSE.txt README.txt Makefile 2 smallptGPU 94 rendering_kernel.cl 13 displayfunc.cpp 14 renderconfig.cpp 6 renderdevice.cpp 8 smallptGPU.cpp 3 camera.h 2 displayfunc.h 2 geom.h 2 renderconfig.h 7 renderdevice.h 5 vec.h 3 scene_build_complex.pl 2 numbers - Kbytes

[Dade916]

@saasrus: opss, sorry, I was thinking to SLG2. In SmallPtGPU2, you have to edit the rendering_kernel.cl and change the following line from:

if (depth > 6) {

to:

if (depth > 12) {

to increase the max. depth to 12.

[dmfaria83]

what is this noise when you move the camera around?

[Dade916]

It is a a typical behaviour of path tracing/unbiased rendering. You start with a very noise image and more you run the rendering and more the image converge to the "real" solution.

[Robosos]

@ichigoboy86 you can't write any high performance raytracer with haskell because pure haskell is slow (its in fact slow as java).

[obdeniye]

Because the lightsource does not move, you could pre-calculate all the global illumination using a method called Stochastic Ray Radiosity. SRR can also store beautiful caustics. Later, during the realtime raytracing phase, you can perform focal blur and the like. Of course, if the light moves by even an inch, pre-storing is useless.

[ichigoboy86]

Very impressive. I'm curious about what if someone make this using functional programming like Clean or Haskell. (I'm thinking of high-performance implementation of concurrency and parallelism)

[KillahMate]

@EasternMerchant: I can answer that one - yes. This is a very simple test program, designed to just work, not to look pretty. Any real, complete implementation of this algorithm would use various filtering techniques to make the noise as invisible to the naked eye as possible, so it won't take 100 fps to look good - probably anything over 20 fps (depending on the use case) will be good enough for the image to appear seamless. Meaning we're only a few generations of graphics cards away from it.

[EasternMerchant]

do you believe there will be a time when this noise will occur so fast that it will not be seen by the human eye? such as if pathtracing become 100 fps?

[XantheFIN]

Im waiting some GPU acceleration to blender. This looks nice anyway.