Parallel Computing

SEP students have always pushed the limits of their available computing resources. From 1991-95 the ``limit'' was our 32-node CM5 Connection Machine. Since November 1995, the limit has been set by a 16-node SGI Power Challenge. The CM5 parallelization was most easily utilized with CM Fortran. For the SGI parallelization is possible in C, and therefore C++, but automatic parallelization is extremely limited due to the way pointers are handled in each language.

As a result even the most enthusiastic supporters of C++ abandoned it when it came to serious processing of real data. With SEP's shift into 3-D, the importance of doing high performance parallel computing is even greater.