Heroux / Edwards / Doerfler
Thursday, November 8th
1:15 - 2:15 pm
Chip multi-processors (CMPs) offer a new challenge and opportunity for application developers. Although using MPI to access all cores is possible in principle, it is very likely that MPI-only will not be sufficient. At the same time, no one knows what other approaches will be best, or when and how. Characterizing the effective use of CMPs is essential at all levels of computing. At the low end, exploiting increasing core counts is the only way to see performance improvement from hardware. At the high end, all paths to petascale (and exascale) performance require effective use of large-core-count CMPs in some form.
Target Audience: All HPC developers