No subject

these newer GPUs, since nvidia and ATI want to break into the HPC realm, I
wonder if the communication issues you saw in the GF7 are less of an issue
now.  The GPU as a processing aid seemed kind of tacked on to the GF7 and
ATI 1xxx series of video cards so it might be much more potent now.

Though wouldn't that be odd.  You still could end up using your video card
to render raytraced graphics but only because it could be used as a
floating point processing add-on.

I know you know of CUDA:  http://www.nvidia.com/object/cuda_home.html

But ATI has a somewhat similar thing as well:

http://sourceforge.net/projects/amdctm/
http://en.wikipedia.org/wiki/Close_to_Metal
http://ati.amd.com/technology/streamcomputing/resources.html
http://ati.amd.com/technology/streamcomputing/faq.html#10

Though ATI's stuff seems harder to find for this, not as nicely marketed
as CUDA.

Anyway, I know you're probably pressed for time with your thesis, but if
you get a chance and have access to the newer hardware, I'd be curious if
there's an actual speedup rather than a slowdown due to the PCIe
communication.

It would be somewhat ironic, I think, if that proves to speed things up. 
Intel's been pushing pretty hard for RT on multiple CPUs and it'd be funny
if the existing graphics cards can handle things without the need for
8-core CPUs or an intel-based multi-core video card.

Monk.