Keynote speaker: Doug Burger, Microsoft Research:
Transitioning from the Era of Multicore to the Era of Specialization
Challenges in silicon scaling have already limited single-thread CPU performance gains, and will soon limit multicore CPU performance gains as well. Many researchers are exploring hardware specialization as the next frontier to drive continued improvements in computing efficiency and performance. Unfortunately, there is a well-known tension between generality (which affects programmer productivity) and efficiency. In this talk, I will describe how programmable hardware based on FPGA chips can bridge this gap and permit specialization while also supporting reasonable levels of productivity. I will describe the Catapult architecture, a system-level architecture that incorporates a programmable logic fabric into a scale datacenter. We manufactured specialized FPGA boards tightly integrated into a 6x8, 2-dimensional torus, and deployed 1,632 of them in a production datacenter. We showed that this fabric can be used to improve datacenter efficiency significantly, reducing the number of servers required to run the Bing ranking service by half. Finally, I will show why tightly coupling FPGAs into an elastic fabric meets the needs of future specialization for large-scale datacenter computing
Doug Burger is Director of the Client and Cloud Applications group in Microsoft Research's Technologies division. His group's goal is to build disruptive research prototypes that provide new directions and capabilities for Microsoft's product divisions. His group currently works in the areas of new datacenter architectures, specialized silicon accelerators, new device form factors, platforms for virtual and augmented reality, advanced optics, personalization and privacy of mobile services, and new mobile/cloud system architectures. Prior to joining Microsoft in 2008, he was on the Computer Sciences faculty at the University of Texas at Austin for ten years, where he co-led the TRIPS project. He is an ACM Fellow, and IEEE Fellow, and the recipient of the 2006 ACM Maurice Wilkes Award.