清华大学 | 信息学院 | 国家实验室 | English Version

重要公告

【2012学术报告4】GPU Accelerated High-Performance Computing

2012年清华大学信息技术研究院系列学术报告4

Title: GPU Accelerated High-Performance Computing - Challenges in Algorithm Design and Application Development

Speaker: Bo Hong   Assistant professor

Time: May 17, 2012, 9:00-11:00 A.M

Place: 1-315, FIT Building

Organizer: Research Institute of Information Technology (RIIT), Tsinghua University

Biography:

Dr. Bo Hong is currently an assistant professor with the school of Electrical and Computer Engineering at Georgia Institute of Technology, Atlanta, USA. He received his PhD degree in Computer Engineering from the University of Southern California in 2005. He received his Master and Bachelor's degree from Tsinghua University in 2000 and 1997, respectively. Dr. Hong's research interests include high performance computing, parallel and distributed computing. He has published extensively in these areas and served on the organization/program committees of many conferences such as IEEE IPDPS, HiPC, and CSE. Dr. Hong is a recipient of the US National Science Foundation CAREER award.

Abstract:

GPU has recently become an essential high performance computing device and has been widely adopted in many HPC systems. Efficient utilization of GPUs is challenging though - the peculiarities of GPU architectures, especially its memory and the thread scheduling subsystems, have made it challenging for general applications to achieve efficiency on GPU platforms. In this talk, Dr. Hong will discuss algorithm design and application development for GPU platforms. In particular, he will talk about (1) efficient GPU algorithms for the widely known maximum network flow problem, which will demonstrate the importance of thread synchronization, and (2) efficient GPU algorithms for the protein-DNA docking problem, which will demonstrate the effectiveness of coalescing memory accesses and minimizing execution path divergence. Dr. Hong will analyze how such algorithmic techniques can be utilized to achieve 2x to 30x speedup against state-of-the-art CPU-based algorithms for these problems.

【发布时间:2012-05-21】【浏览次数: