Scalability, Speed, and Numerical Stability of Massively-threaded GPU Algorithms - A Case Study of Tri-diagonal Solvers

Colloq: Speaker: 
Wen-mei Hwu, PhD
Colloq: Speaker Institution: 
University of Illinois
Colloq: Date and Time: 
Thu, 2012-10-11 14:00
Colloq: Location: 
Building 5700, L202
Colloq: Host: 
Jeffrey Vetter
Colloq: Host Email: 
Colloq: Abstract: 
The IMPACT group at the University of Illinois has been working on the co-design of scalable algorithms and programming tools for massively threaded computing devices. A major challenge that we are addressing is to simultaneously achieve scalability, performance, and numerical stability for tri-diagonal solvers. In this talk, I will go over the major building blocks involved: memory layout, SPIKE-based problem decomposition, and dynamic tiling barrier. I will show experimental results to demonstrate how these building blocks jointly enable the first scalable, numerically stable tri-diagonal solver that matches the numerical stability of MKL and the speed of CUSP.
Colloq: Speaker Bio: 
Wen-mei W. Hwu is a Professor and holds the Sanders-AMD Endowed Chair in the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign. He is also CTO of MulticoreWare Inc., chief scientist of UIUC Parallel Computing Institute and director of the IMPACT research group ( He co-directs the UIUC CUDA Center of Excellence and serves as one of the principal investigators of the $208M NSF Blue Waters Petascale computer project. For his contributions, he received the ACM SigArch Maurice Wilkes Award, the ACM Grace Murray Hopper Award, the ISCA Influential Paper Award, and the Distinguished Alumni Award in Computer Science of the University of California, Berkeley. He is a fellow of IEEE and ACM. Dr. Hwu received his Ph.D. degree in Computer Science from the University of California, Berkeley.