Seminar Abstract:

Offering massive parallelism and high energy efficiency, GPUs are now widely deployed in data centers and supercomputers to accelerate data processing. Whereas being the major horsepower for a plethora of domains, GPU-dominated heterogeneous computing is still experiencing the active migration and evolution of high performance applications. The process confronts a bunch of challenges, such as how to refactor code implementations to adapt to heterogeneous paradigms, and how to effectively manage resources to maximize utilizations. Around the challenges, this talk presents an exemplify work to port and optimize in the field of HPC represented by computational fluid dynamics (CFD), and system-level optimizations to accommodate mixed-precision computing and kernel parallelism.

About the Speaker:

Xianwei Zhang is an associate professor in the School of Computer Science & Engineering at Sun Yat-sen University. During 2017-2020, he worked in AMD Inc. on GPUs of exascale supercomputing. He completed his Ph.D (2017) in the Computer Science Department at University of Pittsburgh, and obtained his Bachelor’s (2011) degree from Northwestern Polytechnical University. Dr. Zhang’s research interests lie broadly in hardware and software co-designs to improve the performance and efficiency of computing systems, with a particular focus on compilation and GPU computing.