High network latencies in large-scale multiprocessors can cause a significant drop in processor utilization. By maintaining multiple process contexts in hardware and switching among them in a few cycles, multithreaded processors can overlap computation with memory accesses and reduce processor idle time.
