Dec. 15 — Steven Brandt, a computational scientist in the LSU Center for Computation Technology, delivered the first specifications of the “chess benchmark” to be used in the Defense Advanced Research Projects Agency’s (DARPA) Ubiquitous High Performance Computing (UHPC) program.Â
The UHPC Program is DARPA’s most recent and aggressive research initiative in high performance computing, to create a revolutionary new generation of computing systems that overcomes the current limitations of power consumption and programming. The UHPC objective is to deliver a single rack capable of 1 Petaflops within a power envelope of 57 Kwatts including cooling.
LSU is participating in the “TA2” project led by the Georgia Tech Research Institute (GTRI) for the development of applications, benchmarks, and metrics in support of DARPA evaluations of four TA1 team system designs.
The chess benchmark serves as an exemplar of graph-based decision problems, designed to stress candidate UHPC machines in atypical and important ways. In particular, an efficient chess program is a dynamic graph-based algorithm that must be able to distribute or redistribute large amounts of parallel work, synchronize and communicate state information across the machine, and selectively abort entire groups of running calculations.
“These capabilities are of use to a number of mission-critical Department of Defense applications related to search or tactical analysis,” said Thomas Sterling, LSU CCT and Computer Science professor.
Work on this project is being carried out by the LSU Center for Computation Technology’s ParalleX Group led by Professor Thomas Sterling and includes Dr. Steven Brandt, Chirag Dekate, and Phillip LeBlanc. For more information on the LSU CCT ParalleX group projects, visit http://px.cct.lsu.edu.
Source: LSU Center for Computation Technology