Near-Threshold Computing: How Close Should We Get?
Alaa Alameldeen,
Research Scientist, Intel Labs
Near-threshold computing research has been gaining traction recently due to the increased focus on energy efficiency across market segments, from ultra-mobile to high-performance computing. Prior work has focused on trying to run systems as close to threshold voltage as possible without hurting correctness. Some research went even further by investigating "approximately" running systems at near-threshold voltages, aka "Approximate Computing". In this talk, I would argue that running "as close as possible" to threshold voltage hurts the cause of energy efficiency in many cases. The talk advocates for another approach where near-threshold computing could be done adaptively, and only when it improves energy efficiency.
Alaa R. Alameldeen received BSc and MSc degrees from Alexandria University, Egypt, in 1996 and 1999, respectively, and MSc and PhD degrees from the University of Wisconsin-Madison, in 2000 and 2006, respectively, all in computer science. He is a research scientist at Intel Labs, in Hillsboro, Oregon. His current research focuses on energy-efficient memory and cache design.
