I don't know about ATI/others, but NVIDIA graphics cards have only recently been able to do calculations in double precision (CUDA 2.0 was released ~August 2008). The current hardware implementation doesn't even take full advantage of the architecture (good explanation here: https://www.cs.virginia.edu/~csadmin/wiki/index.php/CUDA_Support/Enabling_double-precision#Performance; that kind of accuracy seems not to have been needed before).
There are pictures in the journal article they cite: http://download.cell.com/neuron/pdf/PIIS089662731000766X.pdf?intermediate=true
I don't know about ATI/others, but NVIDIA graphics cards have only recently been able to do calculations in double precision (CUDA 2.0 was released ~August 2008). The current hardware implementation doesn't even take full advantage of the architecture (good explanation here: https://www.cs.virginia.edu/~csadmin/wiki/index.php/CUDA_Support/Enabling_double-precision#Performance; that kind of accuracy seems not to have been needed before).
I'm sure they'll keep this in mind when designing the new combination giPhone... now where should one put that?