IBM Takes Top Supercomputer Honors Again

A system created by IBM and the Department of Energy's Lawrence Livermore Lab hits 478.2 trillion calculations per second.

IBM and the Department of Energys Lawrence Livermore National Laboratory in Livermore, Calif., took top supercomputer honors … again.

The IBM Blue Gene/L system installed at the DOE research facility, runs at 478.2 teraflops or 478.2 trillion calculations per second, earning it the most powerful system on the Top 500 Super Computer list, which is published twice annually by the University of Mannheim, Germany, the University of Tennessee and the National Energy Research Scientific Computing Centers Lawrence-Berkeley National Laboratory. It was released at the International Supercomputer Conference in Reno, Nev., on Nov. 12.

To maintain the top spot, IBM and DOE researchers boosted peak performance increased its peak performance from 280.6 teraflops, with 40 additional server racks added to the Blue Gene/L systems 64 racks, each with 1,024 compute nodes that hold 2,048 IBM Power processor cores, said Herb Schultz, marketing manager of IBMs Deep Computing division.

The Blue Gene/L system at the DOEs Lawrence Livermore lab is far ahead of the second most powerful system in the world, which happens to be an IBM Blue Gene/P system installed at Germanys Forschungszentrum Juelich research institute. This supercomputer has a peak Linpack benchmark performance of 167.3 teraflops. (The Linpack benchmark is the standard used to measure the performance of these supercomputers.)

/zimages/3/28571.gifClick here to read more about the NECs new supercomputer initiative.

The list, released Monday, showed that supercomputers are getting much more powerful in just a short amount of time. When the Top 500 list was released in June, three supercomputers broke the 100 teraflop mark in peak performance. The November list shows seven of the top 10 breaking 100 teraflops of performance.

The new list also shows that companies such as IBM, Sun Microsystems and others should be able to hit the next major milestone of supercomputing by next year, namely building machines that can hit a petaflop of performance, or one quadrillion calculations per second.

By attacking these types of problems with such large-scale systems, Schultz believes that IBM and other supercomputer vendors are gaining experience in tackling more mainstream IT issues for customers in financial services, oil and gas exploration and public utilities that need to run applications for massive workloads.

"This sort of looms over what they are doing," Schultz said. "You have financial institutions running these huge financial modeling applications and you have oil and gas companies looking for resources in obscure places."

IBM built four of the top 10 supercomputers on the new list, while Cray built three and Hewlett-Packard built two machines. SGI also built one of the supercomputers listed in the top 10. Overall, IBM built 232 supercomputers on the Top 500 list, while HP built a total of 166 supercomputers.

For the first time ever, an India research facilities has a supercomputer in the top 10. Earlier this year, the Computational Research Laboratories in Pune, India, installed an HP Cluster Platform 3000 BL 460c systems that had a peak performance of 117.9 teraflops. This makes it the fourth most powerful system in the world, according to the new list.

A total of 354 supercomputer systems—about 71 percent—on the Top 500 list use Intel processors, while 78 systems use processors from Advanced Micro Devices. Another 61 systems use IBM Power processors.

Editors Note: This story was updated to correct the percentage of supercomputer systems on the Top 500 list using Intel processors.

/zimages/3/28571.gifCheck out eWEEK.coms for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.