Hi

>>Btw if you mention Purcells Paper it is
>>only fair to mention Carr et als, too they have a paper at the
>>same conference with the same topic (GI strategies) on GPUs.

>A minor correction:

>Nathan A. Carr, Jesse D. Hall and John C. Hart. The Ray Engine. Proc.
>Graphics Hardware 2002, Sep. 2002 (can be found at
>http://graphics.cs.uiuc.edu/~jch/papers/)

>was at Proc. Graphics Hardware, not SIGGRAPH 2002 (where Purcell's paper
>was presented). The thing I like about Purcell et al. was that they put
>just about everything they could on the GPU-side; Carr's approach is more
>limited.

That's not the one I ment, I ment the one you mention below (which is
at the same conference, Purcell's Photon mapping is GH and Carr is GH this
year also). Altough I wonder what both papers actually have to do with new
graphics hardware. ;)

Well I have some major issues with Purcell's Siggraph Paper 2002 he does
not really mention how things actually should work or can be
implemented on the GPU. Best example how does he distinguish secondary rays and
primary
rays. Does he trace first primary rays till the end and make another setup
for secondary rays? Then what about the early stencil
test, to rely on this is not a good option. Or what about his datastructures
on GPU? How many memory does he need to get a decent scene onto the GPU?

Carr's approach is maybe limited but it should actually work, but suffers
extremly from many gpu->cpu transfers (which are costly).
Frankly I wouldn't try to implement either Purcell's nor Carr's
GPU raytracing approach.

Best regards,
J.D.

--
Explore SRT with the help of Java3D
(http://wwwvis.informatik.uni-stuttgart.de/relativity/minkowski)
(http://www.antiflash.net/java3d/relativity (mirror)