Getting up to speed :the future of supercomputing