Research Projects

Communication-Avoiding Linear Algebra

James Demmel, Katherine A. Yelick, Mark Frederick Hoemmen, Marghoob Mohiyuddin and Laura Grigori

We are currently working on modifying both sparse (iterative and direct methods) and dense (direct methods) linear algebra algorithms to exploit techniques for avoiding communication costs. Communication consists of messages over the network or bus in the parallel case, and data transfers between different levels of the memory hierarchy in the sequential case; our techniques apply to both.

More information: http://bebop.cs.berkeley.edu