Fast parallel computing with Intel Phi coprocessors by Andrew Ekstrom.
Andrew tells a tale of going from more than a week processing a 10,000×10,000 matrix raised to 10^17 to 6-8 hours and then substantially shorter times. Sigh, using Windows but still an impressive feat! As you might expect, using Revolution Analytics RRO, Intel’s Math Kernel Library (MKL), Intel Phi coprocessors, etc.
There’s enough detail (I suspect) for you to duplicate this feat on your own Windows box, or perhaps more easily on Linux.
Enjoy!
I first saw this in a tweet by David Smith.