Scientific computing works with ever more complex models and seeks ever more
detailed simulations, analyses and results. The advent of multicore,
many-core and other new computing platforms offer the prospect of greatly
increased computing power that will make these computations feasible. However,
in order to realize this potential, application developers will need the help
of computer architects, software engineers, programming language
and compiler researchers and parallel algorithms designers. This talk is a call
for closer collaboration between computer scientists and domain scientists that
will lead to a principled approach to tap into the power of tomorrow's high-performance
computers to solve the scientific challenges of the future.