Computer Engineering Cider Seminars

There is growing interest in using graphics processors (GPUs) to accelerate non-graphics, throughput-oriented software. GPUs have the potential to significantly improve performance per unit cost versus general purpose CPUs for applications containing significant parallelism. This talk will provide a brief introduction to GPU Computing and summarize some of the key challenges contemporary GPUs present to software developers beyond the need to write parallel code. It will describe in detail some recent research on microarchitecture mechanisms for improving branch handling and then summarize work on scalable memory access scheduling and on-chip interconnection networks in accelerator architectures. This talk will also briefly describe GPGPU-Sim, a simulator capable of running GPU Computing applications written in CUDA and OpenCL developed to enable architecture research on these challenges.

Tor Aamodt is an assistant professor in the ECE department at the University of British Columbia (UBC). He received his BASc, MASc and PhD degrees at the University of Toronto. His research focuses on manycore accelerators (such as GPUs), heterogenous computing architectures, and analytical performance modeling of processor architectures. Prior to joining the faculty at UBC he worked at NVIDIA on the GeForce 8 Series GPU (G80) and interned at Intel's Microarchitecture Research Lab in Santa Clara, California.