"Manycore processors: will evolution be good enough?"

The rapid growth in uniprocessor performance has been halted by
shifting technology characteristics and the need to limit power
consumption. Future performance gains must now come from a rapid
expansion in the number of cores integrated onto each chip. An
optimistic view is that we can simply evolve to support 100's or even
1000's of cores using existing architectural approaches. The
challenging alternative is that we will be forced to illuminate much
more of the viable design space in the search for efficient
solutions. Could this ultimately mean that computer architecture will
have to transform itself into a science!