10x10 - Taming Heterogeneity for General-purpose Computing

As Moore Law’s scaling slows in power and speed, energy has become the critical challenge.  The emergence of multi-core, GPGPU computing, and other accelerator-based computing approaches are all signs if the imminent crisis.  Heterogeneous computing approaches promise of 100-fold, but the ad hoc, irregular, and revocable introduction of large-scale heterogeneity, poses major challenges for software and computer science generally.  We are developing the 10x10 paradigm, a principled, systematic approach to heterogeneity in computer architecture. 10x10 exploits deep application knowledge to tame the power of VLSI customized design and create new architectures that are both programmable and low-power, appropriate for future data centers of millions of processors and future billions of mobile devices.   These new architectures will represent stable targets for new software stacks that deliver both energy-efficient performance and robust, portable software for future generations of systems.   Current efforts include: 10x10 Clusters: Analysis and Clustering of Computational Structures for Heterogeneity Benchmark suites (UHPC, Parsec, ...) DOE Science applications and Problem Solving Suites/Libraries 10x10 Benchmarking - Current and Future Heterogeneous Machines A Portable Benchmark for Heterogeneous CPU's, GPU's, and Accelerators Benchmark results for Heterogeneous Systems Publications: BorkarChien2011,10x10May2010,ICCS2011. People: Apala Guha, Yao Zhang, Dominick LoBraico, Andrew A. Chien (UChicago), Mark Gahagan, Pietro Cicotti,  Allan Snavely (UCSD/SDSC) The LSSG is part of the Systems Group in the University of Chicago's   Department of Computer Science, and also affiliated with Chicago's Computation Institute, and Argonne National Laboratory's Mathematics and Computer Science Division.