|
 |
Am 04.08.2013 11:30, schrieb Le_Forgeron:
> Le 04/08/2013 02:38, nemesis nous fit lire :
>> Alain <kua### [at] videotron ca> wrote:
>>>
>>> That's not realy the kind of number crunching that POV-Ray need.
>>>
>>> The GPU is good at sequemtial number crunching: Decoding a video stream,
>>> decoding an audio stream, performing some scan line rendering.
>>> POV-Ray need random access number crunching and use the FPU for that,
>>> not the CPU itself. During ray tracing, it's prety hard to anticipate
>>> what data will need to be treated next.
>>
>> http://render.otoy.com/
>
> there is one huge difference between povray and usual rendering engine:
> Usual engine only handle billions of triangles. Povray uses mathematical
> formula for most shapes.
As a matter of fact, the problem is not that POV-Ray's math doesn't lend
itself to massively parallel computations - the problem is that
POV-Ray's rendering framework wasn't designed for massively parallel
computations, and would need a major redesign. The trick would be to
shoot bundles of N nearly parallel primary rays at once, and defer
computation of any secondary rays, sorting them by origin and direction,
until you have collected enough of them to shoot bundles of nearly
parallel rays again.
Post a reply to this message
|
|
