that there is a method to use your GPU to do all the math,
Yes but not in maya. GPU programming is hard, its expensive and it locks your users down to certain hardware. Most vendors do not like to do this. Its much harder to program, and has a smaller audience that could use the software, more cost less return. And finally only about 2% of potential customers have very good hardware to speak of, tough mayay users maybe are at 50%. It also makes running maya on render servers terribly hard.
Its not as if the gpu runs the same code as the cpu, if it did then it would be all spiffy. But it does not, you cant just take a program designed to cpus and say use gpu. It dot work that way.
Also you cant just solve a problem by throwing money at it. Similarly you cant just solve your speed issues with buying more expensive hardware. Rich would wish this was true but its not.
those classes simply do not include a method to involve gpus or even allow an alternate
and maya independent processing source
no but that does not men one could not do this, its just that there would be a big bottleneck just there in the api. Anyway its not much use to make the maya fluid container hardware accelerated. Basically you would start form scratch and write it ALL on your own, including the mentalray and maya software implementations of it. But yes that's like 5 times to work to develop it on one system alone and youd miss all the nice rendering quality stuff that maya allready has.
There just no market for this the way i see it, not atleast the required 300-3000 manhours of work. besides cpus are more flexible so once you want to extend the code then the work done for the gpu is a bit problematic.