Leveraging HPC expertise developed for stockpile stewardship to advance a broader range of science
Close-up image of wiring on one of LLNL's supercomputer systems

Grand Challenge Supercomputing Projects Push the Computing Envelope

Monday, October 6, 2014

“The Grand Challenge program encourages Lab scientists to leverage our high performance computing (HPC) expertise developed for stockpile stewardship to advance a broad range of science critical to the nation,” said Dona Crawford, associate director for Computation. “Allocating time on some of our most powerful supercomputers is an important investment we make in LLNL’s science and technology.” Teams with winning proposals were allocated time starting October 1 on Sierra, a 261-teraFLOP/s machine and Vulcan, a 5-petaFLOP/s machine. Sierra and Vulcan are systems dedicated to unclassified research through the Laboratory’s Multiprogrammatic and Institutional Computing (M&IC) program. These HPC machines consist of thousands of CPUs; the Sierra system has 1,944 nodes each with 12 cores, for a total of 23,328 cores, while Vulcan has 24,576 nodes each with 16 cores, for a total of 393,216 cores. CPU time is measured across the multiple CPUs in a computer. For example, two CPU hours can be one CPU used for two hours or two CPUs used for one hour.

“This year marked the first time that institutional demand for Vulcan cycles exceeded the available allocation,” said Brian Carnes, M&IC program director. “It’s encouraging to see so many teams willing and able to take on the challenge of advancing science by pushing the state-of-the-art in HPC.”

Project proposals were reviewed by both internal and external referees. Quality and potential impact of the proposed science or engineering effort were among the key criteria used in evaluating the projects.

Over the last 15 years, HPC resources dedicated to unclassified research have increased more than a 10,000-fold from 72 gigaFLOPS in 1999 to more than 5 petaFLOPS today. To put that in perspective, only seven countries in the world possess more computing resources in total than the Laboratory makes available for unclassified computing. Grand Challenge allocations have accelerated code development for stockpile stewardship and research in critical areas such as laser plasma interaction and first principles methods for equation of state.