How does Throughput Computing work?
Moore’s law has traditionally seen processor performance to double every two years as a result of the doubling of the number of the transistors that can be crammed onto a slice of silicon. Unfortunately, memory speeds have only been doubling every six years, creating an ever-increasing gap. As a result, today’s processors are stalled as much as 75 percent of the time while they wait for memory to fetch data. With Throughput Computing, when a thread must wait for the memory, the affected core simply starts processing another thread. CMT processors transcend the clock-speed focus of traditional chip design to deliver systems with significant price/performance improvements and higher reliability while lowering the cost of computing by allowing fewer systems to handle the same workload.
