Blue Gene/L, already ranked as the fastest supercomputer on the planet, has been doubled in size, according to researchers at Lawrence Livermore National Laboratory in Livermore, Calif.
Livermore has been running a 32,000 processor system since December, but three weeks ago trucks began delivering the components that allowed Livermore to add another 32,000 processors worth of power to the supercomputer, effectively doubling its processing power.
Though there are still some adjustments being made, the system is now operational, said Robin Goldstone, group leader with the Production Linux Group at Lawrence Livermore. “It’s mostly functional. They’ve actually run calculations on the 32,000 nodes,” she said Wednesday. “They’re shaking out the last few bad nodes.”
Blue Gene/L is made up of approximately 32,000 two-processor nodes, giving it about 64,000 processors in total, Goldstone said.
A 33,000 processor prototype of Blue Gene/L, assembled by IBM Corp. last November was ranked the fastest computer on the planet on the Top500 list of the world’s fastest supercomputers. IBM’s prototype was benchmarked at 70.72 trillion calculations per second, or teraflops, using the Linpack benchmark, which puts the system through a series of mathematical calculations.
Lawrence Livermore’s new system is expected to be capable of approximately twice that performance, making it nearly three times as powerful as the next system on the list, NASA’s (the U.S. National Aeronautics and Space Administration’s) 10,240-processor “Columbia” supercomputer. Columbia has been benchmarked at 51.87 teraflops. Goldstone declined to comment on the Livermore system’s benchmark performance.
The 32,000 node Blue Gene/L represents the second stage of a three-part build-out of the US$100 million supercomputer that is expected to be completed by June. When fully assembled at Lawrence Livermore, Blue Gene/L will be a 130,000-processor system with a theoretical peak performance of 360 teraflops, according to IBM.
Perhaps the most remarkable characteristic of Blue Gene/L is how compact it is. When the complete system is assembled into a total of 64 server racks this June, it will be a about half the size of a tennis court, much smaller than most of today’s supercomputers.
Blue Gene/L will consume less power too. The final system is expected to draw approximately 1.6 megawatts of power. To put this in perspective, another supercomputer that Lawrence Livermore will be bringing online this June, the 100 teraflop ASCI Purple system, is expected to require 4.8 megawatts.
The difference is that ASCI Purple will be made out of general purpose servers, similar to IBM’s eServer p655, whereas BlueGene/L’s compute nodes contain little more than memory and processors.
“We’ve kind of reached the limit with these commodity clusters,” said Goldstone. “They just generate too much heat and too much power.”
IBM is now in the process of commercializing Blue Gene and is selling a 5.7 teraflop single-rack version of the system, called the eServer Blue Gene Solution, to high performance computing customers. The company has also agreed to deliver Blue Gene systems to a number of research institutions including the San Diego Supercomputer Center and the University of Edinburgh.
This month, the Armonk, New York, computer maker plans to operate a 100 teraflop Blue Gene system at its Thomas J. Watson Research Center in Yorktown Heights, New York. This system, which IBM claims will be the world’s largest privately-owned supercomputer, will be used, in part, for life sciences research.