This shows you the differences between two versions of the page.

Link to this comparison view

seminars:seminar_9_11_15 [2017/09/20 22:02] (current)
Line 1: Line 1:
 +====== Spatial Computing: Programming our way into the post-Moore era ======
 +Friday Sept. 11, 2015\\
 +Location: TBD\\
 +Time: 2:30PM\\
 +**Joel Emer**\\
 +The historical improvements in the performance of general-purpose
 +processors have long provided opportunities for application
 +innovation. Word processing, spreadsheets,​ desktop publishing,
 +networking and various game genres are just some of the many
 +applications that have arisen because of the increasing capabilities
 +and the versatility of general-purpose processors. Key to these
 +innovations is the fact that general-purpose processors do not
 +predefine the applications that they are going to run.
 +The constantly improving performance of conventional single core
 +processors, aided significantly by Moore'​s law, has fueled these
 +application innovations. However, as we start to see a slowing of
 +Moore'​s law we are seeing alternative approaches for improving
 +performance. This has included using programable accelerators,​ such as
 +GPUs. It has also included increasing use of dedicated accelerators.
 +Unfortunately,​ while this improves performance it sacrifices
 +generality. More specifically,​ the time, difficulty and cost of
 +special purpose design preclude dedicated logic from serving as a
 +viable avenue for application innovation.
 +There recently has been interest in addressing this dilemma between
 +providing programmability and higher performance via an interesting
 +middle ground between fully general-purpose computing and dedicated
 +logic. In specific, spatial computing uses arrays of small
 +programmable processing elements (PEs) that operate in dataflow
 +fashion to provide a very efficient execution engine. We have been
 +exploring the possibilities for spatial computing as an ingredient of
 +general-purpose computation,​ and we will discuss the instrisic
 +operational characteristics of such an approach.
 +In this talk, we will also describe a specific architecture that uses
 +the twin ideas of triggered instructions and latency-insensitive
 +channels to create an efficient control and communication environment
 +for a spatial processor. Triggered instructions completely eliminate
 +the program counter and allow programs to transition concisely between
 +states without explicit branch instructions. Latency-insensitive
 +channels allow efficient communication of inter-PE control
 +information,​ while simultaneously enabling flexible code placement and
 +improving tolerance for variable events, such as cache accesses.
 +Our analysis shows that a spatial accelerator using triggered
 +instructions can achieve over 8X greater area-normalized performance
 +than a traditional general-purpose processor.
 +Dr. Joel S. Emer is a Senior Distinguished Research Scientist in
 +Nvidia'​s Architecture Research group. He is responsible for
 +exploration of future architectures as well as modeling and analysis
 +methodologies. In his spare time, he is a Professor of the Practice at
 +MIT, where he teaches computer architecture and supervises graduate
 +students. Prior to joining Nvidia he worked at Intel where he was an
 +Intel Fellow and Director of Microarchitecture Research. ​ Even
 +earlier, he worked at Compaq and Digital Equipment Corporation.
 +Dr. Emer has held various research and advanced development positions
 +investigating processor microarchitecture and developing performance
 +modeling and evaluation techniques. He has made architectural
 +contributions to a number of VAX, Alpha and X86 processors and is
 +recognized as one of the developers of the widely employed
 +quantitative approach to processor performance evaluation. More
 +recently, he has been recognized for his contributions in the
 +advancement of simultaneous multithreading technology, processor
 +reliability analysis, cache organization and spatial architectures.
 +Dr. Emer received a bachelor'​s degree with highest honors in electrical
 +engineering in 1974, and his master'​s degree in 1975 -- both from
 +Purdue University. He earned a doctorate in electrical engineering
 +from the University of Illinois in 1979. He has received numerous
 +public recognitions,​ including being named a Fellow of both the ACM
 +and IEEE, and he was the 2009 recipient of the Eckert-Mauchly award
 +for lifetime contributions in computer architecture.
 +**[[seminars| Back to the seminar page]]**