Comfortably fit matrices of order 125000 into 4GB of memory. (126491)

Naive method can't even fit 25k (22360)

32 bit ints: (31622)

8 bit shorts:  (63245)

We need to build blocks of size: 


addin efficiency gain: 74.999475836
smulin efficiency gain: 133.385306893
axpyin efficiency gain: 63.547325425