CIS Seminar - Deploying Simultaneous Multi-Parallelism for Performance Enhancement

CIS Seminar (PhD Comprehensive Exam Series)
225 (conference room) Weir Hall

Speaker: Kyoshin "Joel" Choo

Title: Deploying Multi-dimensional Parallelisms for Performance Enhancement

Committee: Dr. Rhodes, Dr. Wang, and Dr. Jang (committee chair)

Abstract: Computer architects are exploring ways of increasing throughput as increasing processor speed reaches to the limit. One of the trend is to exploit different types of parallelism in instruction, thread and data level. In this talk, several examples to increase parallelism are presented. Those include 1) relaxing control flows to execute several independent blocks simultaneously enhances instruction level parallelism, 2) combining advantages of superscalar and multithreaded architecture gives more parallelism in instruction and thread level, 3) adding instruction and thread level parallelism to existing streaming processors can exploit all levels of parallelism, 4) lastly, designing polymorphous architecture that configures its parallelism to suit the characteristic of application provides configurable parallelism. Adding parallelism gives great benefit under certain assumptions, however, it also has limitations and increases hardware complexity.

Bio: Kyoshin “Joel" Choo is currently a Ph.D student in the Department of Computer and Information Science at the University of Mississippi, working with Dr. Jang as a graduate research assistant. Kyoshin received his BA degree in Electrical Engineering and Computer Science from Handong Global University in South Korea in 2000, and his Master of Science in Electrical Engineering-Systems from the University of Michigan, Ann Arbor in 2001. Before his doctoral study, he worked for Samsung Electronics for 10 years where he developed and mass-produced demodulator chips for world broadcasting standards. His research interests include GPU hardware architecture, CPU-GPU heterogeneous computing and system architecture, optimization of GPGPU applications, and modeling and analyzing the performance of GPU and heterogeneous systems.