Due: Thursday February 29, 1996
The Matrix Multiply Contest
The purpose of this assignment is
not so much parallel performance, but to
hit home what it takes to get good performance
out of a machine. The point is that writing
correct code just is not good enough.
What we want to do is write fast code!
Let's see how far we can go by using the
Sivan Toledo's Performance Monitor Page
We are going to do this on nodes of the sp2.
I suggest you pick a node such as sp2n3 or sp2n11
that nobody else is using. There is no parallelism
just ordinary f77 or C on one processor is fine.
The Berkeley crowd ran on individual RS6000's,
but it hardly matters.
Here is some of Berkeley's reference material:
Back to the class home page.