Invisible wrote:
> On 26/01/2012 04:41 PM, Saul Luizaga wrote:
>> Invisible wrote:
>>> Au contrare... Ever heard of displacement mapping? This is nothing other
>>> than running a (simplistic) ray tracer as a pixel shader. Physically
>>> correct dynamic reflections are already possible via ray tracing
>>> (although you're still tracing stupid flat polygons rather than true
>>> geometry). There have been tech demos of full ray tracers running on
>>> standard GPU hardware.
>>
>> Yes, as I have mentioned, I have seen them too, impressive, but not yet
>> in a commercial way and I doubt with 100% POV-Ray features &
>> capabilities.
>
> I think the main thing that makes POV-Ray look so damned good is that it
> doesn't use polygon meshes (it uses real curved geometry), and it
> doesn't use bitmap textures (it uses procedural texturing), and it
> doesn't fake the lighting equation quite as poorly as most game engines.
> (Although it's no unbiased renderer.)

I thought GPUs did some real curved geometry on simple primitives like 
spheres or cubes. not unbiased? how so? I thought POV-Ray accuracy came 
from accurate 64-bit floats raytracing.

>> Never hear of displacement mapping.
>
> Apparently the correct term is "relief mapping", not displacement mapping.
>
> http://en.wikipedia.org/wiki/Relief_mapping_%28computer_graphics%29
>
>> As you mention *some*
>> raytracing is possible on *some* experimental GPUs;
>
> My point is, this is *not* experimental hardware. Relief mapping can be
> done in commercial hardware. Accurate reflections can be done on
> commercial hardware. These features are in commercial games, today.

I understand. As an occasional gamer I have seen these features.

>> what I meant was a
>> 100% GPU implementation of POV-Ray giving at least 24fps at, lets say,
>> 1360x768 resolution.
>
> You're never going to get a 100% GPU implementation of "POV-Ray".
> Because "POV-Ray" is a piece of code that runs on a CPU. What you
> /might/ be able to do some day is implement the same algorithms on a
> GPU.

You are not getting the right perspective of my words: For "100% GPU 
implementation of POV-Ray" I meant "is implement the same algorithms on 
a GPU", I know a GPU can't run CPU code; otherwise I'd wrote: "100% CPU 
implementation of POV-Ray running on a GPU".

> More likely, you could move "most" of the work to the GPU; things
> like parsing the scene data and so forth will always be on the CPU.
>
> I gather that procedural texturing is quite possible on a GPU. (But
> nobody uses it, for whatever reason.) Ray tracing is certainly possible.
> Realistic real-time lighting is possible. The big thing that I haven't
> seen done is non-polygon geometry. I don't know if it's currently
> feasible to do that on a GPU. I don't see why not...

Well I wouldn't know about much GPUs and their use, but if what you say 
is true I think GPUs now are powerful enough to start using more 
raytracing features to render, but I find it harder to calculate and 
render, more transistors more & power consumption, and calculating when 
an object gets hit but another or how it gets destroyed (for games) I 
think would be another task to overcome, because rounded triangle meshes 
I think are easier to manipulate and to fracture. But I am all for 
seeing a GPU version of POV-Ray solving these problems in an efficient 
way and giving us 24 fps.

Post a reply to this message