Parallel Computer Architecture

11.10 Exercises 927

sistency models, and prefetching are commonly encountered in microprocessor ar-

chitectures today. The most general latency hiding technique—multithreading—is

not yet popular commercially, largely because it is unproven for uniprocessors. Re-

cent directions in integrating multithreading with dynamically scheduled superscalar

processors appear promising, but they bear comparison with multiple simpler proces-

sors on a chip. An interesting general question is how well the provisions made for

hiding uniprocessor latencies will succeed in hiding multiprocessor latencies.

Despite the rich space of issues and alternatives in hardwar ...