I don't know if this has been brought up before, but I was just perusing
www.gpgpu.org and was wondering how feasible this would be? My guess is
that it will be limited due to povray using double-precision floats rather,
but may be there's something else?

Post a reply to this message

jhu spake:

> I don't know if this has been brought up before, but I was just perusing
> www.gpgpu.org and was wondering how feasible this would be? My guess is
> that it will be limited due to povray using double-precision floats
> rather, but may be there's something else?

This has been asked before - I don't know much about the actual raytracing
algorythm (I'm just a user) but apparently that is one of the problems. The
other is (if I remember more knowlegdable guys' answers) that a GPU is
optimized for doing polygon / scanline type rendering, i. e. the scanline
type algorytms that programs like 3D studio or Blender use natively to
generate images, while raytracing itself is apparently way different, and
thus would not benefit from being implemented using a "standard" GPU. I
suspect also that, due to the GPU being optimized for "polygonal"
acceleration, it will in fact be less effecient to implement a raytracer on
a GPU than on a standard general-purpose CPU... but I may be wrong.

As far as I know, and from the little personal experience I have, to trace
faster get more CPU's, or increase floating point performance. Going faster
is mostly possible by higher CPU clockspeeds, faster busses and faster RAM.
But the most benefit is probably to be had from parallelizing a render -
divide the image into little blocks and send off a block to each CPU in a
cluster.
-- 
Stefan Viljoen
Software Support Technician / Programmer
Polar Design Solutions

Post a reply to this message