Bo Wu, assistant professor of computer science at Colorado School of Mines, has received a NSF CAREER Award to develop techniques to support multitasking in graphics processing unit (GPU) computing.
Wu’s project, “Compiler and Runtime Support for Multi-Tasking on Commodity GPUs,” will receive $501,546 over five years starting in May.
Designed to accelerate parts of a single application and work as co-processors with general-purpose central processing units (CPU), GPU computing has become mainstream in recent years, used in machine learning, graph analytics and scientific simulation, as well as data centers and cloud computing infrastructures where users increasingly demand accelerated applications.
“When users connect to services on the internet like Google’s search or Apple’s Siri, their request is handled on the same server as many other users’ requests. To handle this traffic, many companies are now leveraging the computing power of the graphics cards (GPUs) present in many modern computers. However, GPUs lack necessary hardware support to guarantee quality of service (QoS) in such scenarios,” Wu said. “This CAREER project aims to tackle this problem by developing software which slightly changes existing applications to allow them to coordinate their execution for both performance and fairness.”
Wu’s team will develop both compiler and runtime techniques to help accomplish transparent, efficient multitasking. Compiler techniques will help circumvent the hardware limitations of GPUs and allow a set of additional features, such as preemption, while runtime systems will permit the scheduling of applications to best utilize the potential of the GPU and guarantee quality of service.
As part of the project, Wu will work to advance GPU education at Mines for both computer science and non-computer science majors.
Two courses open to students of all majors, Introduction to Parallel Computing (CSCI-440) and Advanced High Performance Computing (CSCI-580), will cover the fundamentals of GPU computing and cutting-edge optimization techniques for GPU applications. Wu also plans to offer GPU workshops for faculty and students to get hands-on experience writing GPU code in CUDA, a parallel programming language.
Wu joined Mines in 2014 after earning a PhD in computer science from The College of William and Mary. He holds a master’s degree in computer science and a bachelor’s degree in mathematics from Central South University in Hunan, China.
Emilie Rusch, Public Information Specialist, Communications and Marketing | 303-273-3361 | firstname.lastname@example.org
Mark Ramirez, Managing Editor, Communications and Marketing | 303-273-3088 | email@example.com