Multicore: No Silver Bullet for Lousy Code
February 17, 2012
At one time engineers thought that throwing more cores at a bottleneck would solve the problem. They were wrong. ExtremeTech examines “The Death of CPU Scaling: from One Core to Many—and Why We’re Still Stuck.”
The very informative article traces computer processor development from the time CPU scaling ceased to be feasible and manufacturers began multiplying their cores in 2004. Writer Joel Hruska does a good job explaining Moore’s law, Dennard scaling, and Amdahl’s law and how each impacts the growth of processing power. It also details the reasons multiple cores cannot revive the exponential improvements processors enjoyed during the 1990s. (I like the term “dark silicon.” Sounds like a super villain.)
The write up quotes a recent report on dark silicon and multi-core devices:
Regardless of chip organization and topology, multicore scaling is power limited to a degree not widely appreciated by the computing community…. Given the low performance returns…adding more cores will not provide sufficient benefit to justify continued process scaling…. A new driver of transistor utility must be found, or the economics of process scaling will break and Moore’s Law will end well before we hit final manufacturing limits.
That sounds pretty dire, though Hruska predicts some slow scaling progress will continue for several years. After that, who knows? To run fast, do we now think assembler?
Cynthia Murrell, February 17, 2012
Sponsored by Pandia.com
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
