If someone is wanting the fastest processing of CUDA units, the new eVGA GTX 295 boards were released today. A dual GPU board with a whopping 480 cores processing in parallel, 2Gb of memory, and memory bandwidth of over 223Gb/sec! Heck, get two of them and hook them up in SLI.

If you want to get really extreme, install 4 of nVidia's Tesla C1060 boards into a single desktop pc and you now have a machine with 960 cores capable of an incredible 3.732 Teraflops (single precision). Yes, you'd essentially have a massively parallel supercomputer sitting on your desk.

Let's see who's really serious about cranking work units out!

How much heat does each of these setups produce?
And how much electricity do they use?

For the specs and more info, such as the power and heat you asked about, here's links to nVidia's pages

GTX 295 Video Cards
(note that many manufacturers, including EVGA, XFX, Gigabyte, and more are making them)
http://www.nvidia.com/object/product_geforce_gtx_295_us.html

Tesla (teraflop+ parallel multicore supercomputing)
http://www.nvidia.com/object/tesla_computing_solutions.html

CUDA web page
http://www.nvidia.com/object/cuda_what_is.html

Hope this helps!