Designing Scalable Algorithms on GPU-accelerated Hybrid Architectures

Wednesday, 10th February 2010, 10:00 am (PDCC Meeting Room)
Speaker: Dr. Weiguo Liu (NTU)
Title: Designing Scalable Algorithms on GPU-accelerated Hybrid Architectures

The explosive growth of data collection in different research fields has literally forced upon us the need to analyze and mine useful knowledge from them. It presents unprecedented opportunities for knowledge discovery and at the same time new challenges for computer engineering. Due to the continuing exponential growth of various kinds of data, high performance computers with ever increasing performance are highly needed for applications in computational biology, physics simulations, financial computing and scientific fields. Unfortunately, traditional means of high performance computing (HPC) involve prohibitive costs. Therefore, new architectural HPC approaches are needed as well as efficient and scalable algorithms that can take advantage of these new architectures. Hybrid architectures describe the combination of fine-grained (such as graphics processing units (GPUs) and coarse-grained (such as the Beowulf cluster and computational Grids) parallelism. Making use of the power of coarse-grained computing is currently one of the most efficient and simple ways to gain high performance for a reasonable price. Installing in addition massively parallel accelerators within PCs can further improve the cost/performance ratio significantly. This talk will focus on challenges and our methods to design efficient and scalable algorithms on this type of architectures.


Weiguo Liu received his Bachelor and Master degree from the Xi'an JiaoTong University, China in 1998 and 2002, and the Ph.D. degree from the Nanyang Technological University (NTU), Singapore, in 2006. He is currently a Research Fellow with the School of Computer Engineering at NTU. His research interests include computational biology, parallel algorithms and architectures, high-performance computing, and data mining. He has published more than 20 papers in the major conferences and journals in the relevant area including IEEE TPDS, IEEE TITB, JCB, PRL, CPC, ICPP, IPDPS, EuroPar, Cluster, and HiPC.