Slide 8 of 24
Notes:
From this little number juggle we not only learn that bad programs scale better.
Knowing that processing data in big chunks gives better speedup we conclude that the more coarse grained our parallelization is, the better performance we expect.