Rice University’s Dr. Jan E. Odegard recently added his voice to a growing chorus of HPC experts weighing in on the changing HPC landscape. As the featured speaker at the first Lunch’n Learn event put on by the Norwegian Consulate General in Houston, Odegard spoke about the inevitable death of Moore’s law and what it will mean to no longer have this exponential driving chip performance.
Since the invention of the integrated circuit in 1958, the computing industry has been making “diamonds” out of sand, observed Odegard, referring of course to silicon-based microchips. Moore’s law and Dennard scaling fueled five decades of smaller, faster and cheaper microelectronics, and delivered thousand-fold performance increases roughly every ten years, setting the stage for the information age and everything that goes along with it, from supercomputers to iPhones.
But Moore’s law won’t last forever. Feature sizes are hitting the limits of feasibility from a physics, power and cost perspective. Dr. Odegard predicted that Moore’s law will only hold out for another three or four generations (6-8 years) using current fab techniques. Currently no replacement for silicon-based CMOS exists, but researchers are hot on the trail of candidates, including spintronics, nanotubes, graphene, and other exotic technologies.
Moore’s law created a “free lunch” situation, where performance increases were a matter of waiting for the next generation of chips. But chips aren’t the only place to exploit performance. Dr. Odegard reports that at Rice, where he is the executive director of the Ken Kennedy Institute for Information Technology, there is increased focus on better software and tools, and on system level optimizations. This is highly-skilled work, however, which explains why institutions like Rice are doubling-down on outreach and training efforts.
HPC is facing the backend of an exponential that has propelled five decades of progress. That’s a pretty big loss to overcome, but there’s also cause for optimism because innovation flourishes under adversity.
Dr. Odegard will also be speaking at the Rice University Oil and Gas High Performance Computing (OG HPC) Workshop, taking place March 4-5, 2015.