>Let's face it potatoswatter, the only successful software on GPU's right now has
>been programmed by NCSA type guys, be it the chinese organisation equivalent or
>the US version of it. Only those guys have lots of information that's not on the
>net, or have the tools and toys, and have them long before they ever got released.

GPGPUS are successfully used by some Finnish company you have never heard and probably never will (medical stuff).

So yes, when you have a problem which is parallel and don't require truckloads of ram, the GPGPU is a solution.
GPGPU is not a solution to *ALL* problems just hugely parallel without too large memory footprint.
Its not impossible to imagine that certain stuff the key algorithm user waits is just matrix math perfectly fitting GPGPU paradigm, and 99% of code is just UI,setup, io... stuff that user doesn't really wait the computation since computation running time is below 0.1sec and doesn't really fit GPGPU but it just doesn't matter because that is not the thing user is waiting. The method of about using parallerism is not looking where it isn't. Its recognizing where it is, and where it helps. 99% of stuff thats fast enough isn't the thing where you look if the code could parallerized, its the 1% that's running time is over 0.1sec.