|Date||Oct. 1, 2010|
|Speaker||Andreas Olofssons (Adapteva, Inc.)|
|Topic||Compromised Computing: The Practical Approach to Energy Efficient System Design|
The last 40 years has seen incredible improvements in computing performance, power efficiency, density and cost. Unfortunately, it looks like the free ride is over and we are now faced with some serious issues to keep this trend going. Single-threaded performance stopped scaling years ago and chip designers are now having serious power issues as process technology scales to 28nm and beyond. As an industry we are now faced with a tough choice. Do we continue with the traditional programming model for high performance applications and accept the fact that the processing budget will be strictly limited by our battery/electricity budget or do we adopt a radical new programming model that would allow us to significantly reduce the hardware overhead associated with traditional Von Neumann architectures. We present a "compromised computing" approach to the problem, which combines a new C-programmable floating point math accelerator with traditional silicon architectures to enable real applications with an immediate order of magnitude better energy efficiency.