Mathematics Events
[PAST EVENT] Collaborative Heterogeneous Computing
Access & Features
- Open to the public
Collaborative Heterogeneous Computing
Speaker: Yifan Sun from Northeastern University
Talk Abstract:
GPUs have been used to accelerate a wide range of algorithms and applications given their massively parallel computing capabilities. Today, GPU computing faces performance-scaling challenges. On the one hand, most GPU computing applications underutilize the CPU computing capability. On the other hand, single-GPU performance cannot be further increased due to the die size limitation. To continue the historical GPU performance scaling, a GPU needs to calculate collaboratively with CPUs and other GPUs. In this talk, I will highlight my prior research on addressing the challenges of enabling Collaborative Heterogeneous Computing, from aspects including performance evaluation, performance modeling, and performance improvement. For CPU-GPU collaborative computing, I will be focusing on the design of our execution-pattern-informed benchmark suite, Hetero-Mark. For Multi-GPU collaborative execution, I will discuss the design of MGPUSim --- a high-flexibility, high-performance, parallel, multi-GPU simulator. I will conclude with my vision and future research directions on future computing systems.
Bio:
Yifan Sun is a Ph.D. candidate in the Department of Electrical and Computer Engineering at Northeastern University, and a member of the Northeastern Computer Architecture Research Group (NUCAR). He holds an MS from the Department of Electrical Engineering at University at Buffalo, as well as a BS from the Department of Electronic Science and Technology at Huazhong University of Science and Technology (HUST). His research focuses on performance evaluation and performance modeling of CPU-GPU systems, and performance improvement for CPU-GPU collaborative execution. He has published in a wide range of top-tier venues, such as ISCA, HPCA, CHI, IPDPS, IISWC, and TACO. His published papers have won the best paper awards of ICPE18 and WUWNET13, and the best paper candidate of IISWC16. He also owns three approved US patents related to GPU-in-the-Cloud management. He was a recipient of the Outstanding Graduate Student Award (4 per year) of Northeastern University in 2019 and a recipient of the Teaching Award of the College of Engineering at Northeastern University. His research has been highlighted by public media including CNN, BBC, Wired, and the HiPEAC info.
Contact
Tianran Hu