Emerging Architecture Research & Development at Future Technologies Group

Emerging Architectures [ Personnel \| Projects \| Architectures \| Sponsors]
Heterogeneous computer systems will play an important role in future high performance computing systems, as underscored by recent announcements from major computing vendors such as AMD, Intel, Cray, and IBM. The Future Technologies Group at ORNL is studying emerging architectures for use in applications from real-time low-power processing to scientific computing on petascale systems. Field Programmable Gate Arrays (FPGAs), Graphics Processing Units (GPUs), and the Cell BE Processor are among the architectures we investigate with respect to performance and productivity for future computing platforms.
Personnel Team: Jeffrey Vetter (Leader) Sadaf Alam Jeremy Meredith Olaf Storaasli Collaborators: NSF Center for High-Performance Reconfigurable Computing Xilinx SGI
Projects We are pursuing several directions with our research related to emerging architectures. For some projects, we have publications and even code downloads; for other projects that aren't yet that far along, we encourage you to check back for project updates. If possible, publications are available from the Future Technologies publications database. Contact any team member with questions or comments. Exploring Reconfigurable Computing Programming Models to Accelerate High-Performance Computing Applications Smith-Waterman DNA sequencing shows a 100x speedup on Field Programmable Gate Arrays (FPGAs), as seen in the paper "Performance Evaluation of FPGA-Based Biological Applications" from the 2007 Cray User Group technical conference (CUG). Matrix multiplication and equation solvers are explored in the paper "Mapping Sparse Matrix-Vector Multipication on FPGAs" presented at RSSI (2007 Reconfigurable Systems Summer Institute). In "Achieving Supercomputer Performance for Biomolecular Simulations on Reconfigurable Systems" from IEEE Computer 40(2) (2007), time-to-solution was reduced by a factor of three while using only high-level languages. Productivity The IEEE Cluster 2007 conference paper "Balancing Productivity and Performance on the Cell Broadband Engine" investigates a variety of scientific computing and real-time data processing application kernels. The paper "Exploring Accelerating Science Applications with FPGAs" from RSSI 2007 analyzes several methods of FPGA programming. Workload and Performance Characterization "Characterization of Scientific Workloads on Systems with Multi-Core Processors" in IISWC (2006 International Symposium on Workload Characterization) explores state-of-the-art commodity CPUs. At CUG 2006, the paper "Characterizing Applications on the Cray MTA-2 Multithreading Architecture" investigated several applications such as molecular dynamics, finite difference methods, a fast multipole method, and a discrete event simulation engine. Performance Evaluation and Analysis The paper "Analysis of a Computational Biology Simulation Technique on Emerging Processing Architectures" in HiCOMB 2007 (6th International Workshop on High Performance Computational Biology) compares performance across a diverse set of emerging architectures such as theCell BE processor, Graphics Processing Units (GPUs), Multithreaded Array systems (MTA), and multicore commodity CPUs. Future directions: Lhotse - System Software for Tightly Coupled Heterogeneous Computing Programming Models and Languages for Computation Accelerators
Architectures Hardware Cell Broadband Engine Processor IBM Cell BladeCenter, 2xCell BE per node Graphics Processing Units (GPUs) NVIDIA G71 (7900GTX) NVIDIA G80 (8800GTS) AMD R580 (ATI FireStream) Field-Programmable Gate Arrays (FPGAs) Cray XD1 SGI/RASH SRC-6 Digilent XUP Programming Models CUDA PeakStream Mitrion CHiMPS (beta) Viva VHDL
Sponsors Department of Energy, Office of Science National Center for Computational Sciences (NCCS) DARPA High Productivity Computing Systems (HPCS) UT Science Alliance AlphaStar Corporation

ORNL | Directorate | CSM | NCCS | Disclaimer