Sandia National Laboratory Develops Tougher Rating System to Evaluate Supercomputers
| Sandia National Laboratory, www.sandia.gov |
Computer experts at Sandia National Laboratory have developed the Graph500 Challenge, a test that evaluates a computer’s ability to solve problems involving seemingly random graphs. They say it is a more realistic way to compare computers than current tests like Linpack that measure how fast a computer does basic numerical problems.
Graph500 presents computers with large complex graphs with vertices representing data points, such as people in a medical study, and edges representing relations, such as common symptoms, between data points. Such problems stress a computer’s ability to store and communicate large amounts of data in irregular, fast-changing patterns. The Graph500 benchmark also presents problems of different sizes — huge, large, medium, small, mini, and toy.
In a recent competition where Graph500 ranked several supercomputers, no machine could handle problems in the huge or large categories.
“I consider that a success,” says Richard Murphy, a Sandia computer scientist who helped create the test. “We posed a really hard challenge and engineers are going to have come up with innovative new hardware and software to handle ‘large’ or ‘huge’ problems in the available time. And some supercomputers that placed very highly on simpler tests like Linpack were also tested on Graph500, but their scientists didn’t submit the results because their machines would shine much less brightly.”