by tnt » Thu Jul 25, 2013 11:20 pm
GPU basically can execute a lot of threads in parallell (like thousands !), but they must pretty much all do the same exact task at the exact same time, just working on different data. So that's perfect for example to compute a filter on an image, because the exact same operation will produce each ouput pixel independently, only with different input pixels and you can launch thousands of // threads each processing a single output pixel.
Epiphany OTOH is much more parallell than CPUs because it has 16/64 cores, but much less than a GPU because you only get 16/64 threads. But those are not forcibly linked in anyway, each can execute completely different task.
(it's obviously a little simplified, but that's the general idea)