Latest Graph500 Ranking of Fastest Supercomputers Released by Leading Universities at SC15

The eleventh Graph500 list was released today at the Supercomputing 2015 conference (SC’15), with Japan’s K-Computer maintaining its top spot for the second consecutive time.

Fujitsu, IBM and China’s National University of Defense Technology dominated the top 10, with the BlueGene/Q architecture holding eight of the top 10 positions. Other notable entries are:

  • Largest Problem: DOE/NNSA/LANL Sequoia at Scale 41
  • Best single node performance: Institute of Statistical Mathematics ismuv2k (SGI UV 2000) ranking #40 on the list with 175 GE/s
  • Largest single node problem: UV 2000 (#79), 19.6 GE/s
  • Highest Performance with High Memory Utilization: TitanXforsite (#46), 132 GE/s

The Graph500 ranks the performance of more than 200 of the world’s most powerful supercomputers, which are used to analyze “big data” for cybersecurity, medical informatics, social networks and other scientific fields. The list is released twice each year to coincide with top high-performance computer (HPC) conferences, and it is the dominant international benchmark for analytics platforms. The Graph500 is compiled by the Georgia Institute of Technology, Indiana University, the University of Notre Dame and Boise State University.

“It’s about the data movement,” said Richard Murphy, director of advanced computing solutions pathfinding at Micron Technology Inc. and affiliated faculty at Boise State University.

“Supercomputers are built according to the jobs they will execute, and the bottleneck for analytics codes is often memory bandwidth rather than peak floating point capability,” said David Bader, professor and chair of Georgia Tech’s School of Computational Science and Engineering in the College of Computing.

The Graph500 executive committee also announced on Tuesday a major new initiative in analytics benchmarking—a new streaming analytics benchmark and an infrastructure to enable the incubation of new analytics benchmarks from the community. The effort will formalize its comparative methodology to begin the process of creating predictive analytics that map real application performance to benchmark measurements. The group believes this community-driven approach is essential to advancing the state of the art in data-intensive platforms.

“A community-driven approach is essential to any benchmarking process,” said Jack Dongarra, professor at the University of Tennessee and creator of the Linpack Benchmark. “We need a set of metrics for the evaluation of any HPC system.”

Of the upcoming advances, Peter Kogge of the University of Notre Dame says, “Just as we have learned a lot about dense regular computations from the Top500, the Graph500 has generated a knowledge on irregular structures. The upcoming additions to the Graph500 benchmark set should continue fostering such advances.”

The next Graph500 list will be released in June 2016.

David A. Bader
David A. Bader
Distinguished Professor and Director of the Institute for Data Science

David A. Bader is a Distinguished Professor in the Department of Computer Science at New Jersey Institute of Technology.