Accelerating Computational Chemistry Applications using

Colloq: Speaker: 
Akila Gothandaraman
Colloq: Speaker Institution: 
University of Tennessee
Colloq: Date and Time: 
Thu, 2009-04-09 10:00
Colloq: Location: 
5700, L204
Colloq: Host: 
Phil Roth
Colloq: Host Email: 
rothpc@ornl.gov
Colloq: Abstract: 
Scientific computing is characterized by applications that have ever-increasing performance demands. Recent technological advances have led to a number of emerging computing platforms that provide one or more of the following benefits over their predecessors: speed, increased energy efficiency, programmability/flexibility, different granularities of parallelism and higher numerical precision. This trend can be observed in various computing fronts. For example, multi-core processors are now emerging to provide more parallelism, energy efficiency, and performance than their single-core counterparts. Traditional reconfigurable computing platforms using field-programmable gate arrays (FPGAs) are shifting from PCI-based platforms to a variety of high performance reconfigurable computing (HPRC) platforms, which, with high-performance interconnect between the FPGA co-processor and microprocessor, provide a tightly coupled system and friendly programming interfaces, to integrate FPGA acceleration into present scientific applications. Traditional graphics processing units (GPUs) have evolved from fixed-function 3D graphics pipelines to flexible general-purpose computing engines, spawning a number of research efforts that apply GPUs for general purpose computing.Emerging architectures have provided us the computing power required to accelerate scientific applications such as the simulation of large N-body systems. These systems could be quantum systems consisting of interacting atomic or subatomic particles or astrophysical N-body systems. This talk will present the architecture and implementation details of a Quantum Monte Carlo (QMC) simulation method on FPGAs and GPUs. QMC methods are widely used in physics and physical chemistry to study the groundstate properties of quantum many-body systems. For the FPGA implementation, I will present the novel techniques for implementation such as methods to approximate the kernel functions, pipelining strategies and use of a fixed-point representation that guarantees the accuracy required for our simulation along with the results obtained from targeting the implementation onto the Cray XD1 HPRC platform. I will also present the implementation details and results from targeting the QMC application onto NVIDIA GPUs using the Compute Unified Device Architecture (CUDA) paradigm. An opensource and user-friendly framework for hardware-accelerated QMC simulations is being developed to make this research available to the science community. Work in progress includes targeting the QMC application onto hybrid-computing platforms, combining FPGAs and GPUs along with general-purpose processors. To help with this, I am developing analytical performance models in order to understand how to best map the applications onto the emerging platforms.
Colloq: Speaker Bio: 
Akila Gothandaraman is a Ph. D candidate under Dr. Gregory Peterson in the Department of Electrical Engineering and Computer Science at the University of Tennessee, Knoxville. She is also working towards a minor degree in Computational Science. She received her M.S. in Electrical Engineering in 2004 from the University of Tennessee, Knoxville and B.E. in Electrical and Electronics Engineering in 2001 from PSG College of Technology, India. Her research interests are in the areas of high performance computing, high performance reconfigurable computing, computer architecture, and computational science. Her dissertation explores emerging architectures such as reconfigurable computing using field-programmable gate arrays and graphics processing units to accelerate computational chemistry applications. She is a student member of IEEE, ACM, and SWE.