NERSC Picks IBM System for Next-Generation Supercomputer with 3 Trillion Calculations per Second Capability

BERKELEY, Calif. -- The U.S. Department of Energy's National Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory today (April 28) announced that it has selected an IBM RS/6000 SP system as the center's next-generation supercomputer.

The IBM system was chosen based on its ability to handle actual scientific codes and tests designed to ensure the computer's capability as a full-production computing system in NERSC. These tests indicated that the system, when fully installed, will provide four to five times the total current computational power of NERSC, already one of the most powerful supercomputing sites in the world. This agreement, a fixed-price, five-year contract for $33 million, is the largest single procurement in the 68-year history of Berkeley Lab.

"When one of our scientific research facilities takes a step like this, it's critical that the decision meet both today's scientific needs and the potential for even greater future demands," said Energy Secretary Bill Richardson. "Not only will this partnership with IBM achieve these goals, but our Berkeley Lab computing center's expertise will help IBM improve their computers to make them even more effective. This partnership is another example of the Department of Energy's leadership in the field of computational science."

The new system, which will incorporate IBM's newest processor and interconnect technology, will be installed in two phases. When completed, the system will increase NERSC's computing capabilities by more than 400 percent.

Phase I installation, scheduled to begin in June 1999, will consist of an RS/6000 SP with 304 of the two-CPU POWER3 SMP nodes that were recently announced by IBM. This system will be the first implementation of the POWER3 microprocessor, with two processors per node. The 64-bit POWER3 can perform up to two billion operations per second and is more than twice as powerful as its predecessor. In all, Phase I will have 512 processors for computing, 256 gigabytes of memory and 10 terabytes of disk storage for scientific computing. The system will have a peak performance of 410 gigaflops, or 410 billion calculations per second.

Phase II, slated for installation no later than December 2000, will consist of 152 16-CPU POWER3+ SMP nodes, utilizing an enhanced POWER3 microprocessor. The entire system will have 2,048 processors dedicated to large-scale scientific computing. The system will have a peak performance capability of more than 3 teraflops, or 3 trillion calculations per second.

"Although some computing centers describe their system's performance in terms of theoretical peak computing, we look at our systems in terms of how much they can enhance our clients' ability to solve large-scale scientific problems," said Bill Kramer, head of NERSC's High Performance Computing Department and leader of the procurement effort. "That's the real measure of performance in our view."

As part of the purchase contract, NERSC will work with IBM to develop computer-utilization benchmarks and methods to assess and improve the effectiveness of the SP system in a production environment. While the theoretical peak performance of supercomputers can be amazingly fast, that capability does not always represent real-world computing. To ensure that the new NERSC system is well-suited to the workaday world, NERSC and IBM have agreed to develop and test a "SUPER" (System Utilization Performance Effectiveness Rating) benchmark for the new computer. This set of tests will measure how well the SP delivers scientific work under a realistic workload.

"Theoretical computer speed is comparable to the top end of a car's speedometer, and while your car might be able to do 150 mph on the open road, you're really more interested in how it will carry out your day-to-day driving chores," Kramer said. "While we anticipate that most of our users will appreciate the new machine's high speed capability, our main concern is that they have the computing resources they need, when they need it. This contract ensures the system will live up to NERSC's standards for performance and reliability."

The SP is a highly scalable system made up of building blocks called SMP nodes. This architecture will allow NERSC users to increase the size of their computations and make the results more meaningful. For example, one area of research utilizing NERSC's computers is creating accurate models of materials, such as magnets. More powerful computers allow scientists to create larger models, of 1,000 atoms or more, and gain a better understanding of how magnetic fields are affected by temperature. Such research has applications in fields ranging from computer disk drives to power generation.

"The continuing partnership between IBM and the Department of Energy is further testimony to what can be accomplished when two leaders in the field of computational science push the boundaries of conventional thinking," said Rodney Adkins, general manager, IBM RS/6000. "I can't think of anything more noble than being a part of making lives better, whether it's through helping design cleaner engines or increasing life-saving knowledge about our environment through climate modeling. As always, the knowledge we gain from this initiative will also benefit our commercial customers around the world through powerful business solutions."

The SP's architecture will also allow NERSC to run a variety of different-sized computations simultaneously, thereby providing faster turnaround of results for users across the country.

NERSC provides high-performance scientific computing and data storage resources to about 2,500 researchers at national laboratories, universities and industry across the nation who are working on Department of Energy-funded programs such as combustion, climate modeling, fusion energy, materials science and computational biology. NERSC's seven supercomputers, the largest of which is currently a 640-processor Cray T3E-900, are utilized 24 hours a day, seven days a week, and are up and running more than 95 percent of the time, so the computers must be both highly reliable and high-speed.

Established in 1974 as the Controlled Thermonuclear Research Computer Center to provide computing cycles to fusion energy researchers, NERSC's first computer was a borrowed Control Data Corp. 6600. Considered a computing powerhouse at the time, the 6600's overall performance was two-and-a-half million calculations per second.

NERSC's new machine will be 1.2 million times more powerful. The new system will also be 13,000 times more cost-effective.