Benefits:
New libraries speed transfer of recent algorithmic technology in linear algebra, hierarchical methods, and other areas to DOE users.
A new division of labor between compiler writers, library writers and perhaps architects will simply the problems of all of them.
New architectures will make old solutions much less attractive (i.e. relatively slower compared to peak machine speed) unless we develop algorithms that accommodate them.
Controlling the complexity of large heterogeneous, insecure, dynamic applications is a central Computational Science theme.