Since you're counting a streaming coprocessor as multicore, how is your offering "democratizing access to parallel computing" when the cheapest OpenCL capable GPU is half the price, and the programmer likely already owns one?
Part of democratizing is access to information, access to different choices, access to drivers.
Can GPUs do this. In my view, not a bad result considering 2014 general availability? How long (and how many $billions did it take for CUDA to catch on...)
In terms of cost, not really sure what you are referring to... supercomputer.io is free to researchers, which last time I checked is less than not zero.
