100-Petaflop Supercomputers Predicted By 2017

Next year's Top 500 list of the world's most powerful supercomputer will list systems capable of performing calculations as much as eight times faster than the fastest computers available today.

IBM already has deals for a 10-petaflop system in the Argonne National Laboratory near Chicago and another in the University of Illinois at Urbana-Champaign. In addition, the company is on target to also have operational a 20-petaflop system at the Lawrence Livermore National Laboratory in Livermore, Calif., in 2012.

IBM won't be alone in delivering such high levels of horsepower. Cray told InformationWeek it is ready to deliver a 10-petaflop system and has been in talks with the Oak Ridge National Laboratory, which already uses a Cray XE6 supercomputer. No deal has been finalized.

Jack Dongarra, a computing expert at the University of Tennessee who helped design the technology used in testing systems on the Top 500 list, said he expects more 10-petaflop-plus systems next year. "Everything is moving along according to Moore's law, so things are doubling every 18 months, roughly," Dongarra told InformationWeek.

Dongarra predicted computer makers in 2017 will have 100-petaflop systems, with "exascale" systems between 2018 and 2020. An exascale, or extreme-scale, computer is expected to be a thousand times faster than a petascale system. "All of these things are dependent on funding," Dongarra said in an interview. "This will happen if the funding is in place for those machines."

Dave Turek, VP of deep computing at IBM, predicted that exascale computing will be available in 2019. He also said the contracts are in place for the three latest petaflop systems. "We don't expect the government to back out," he said in an interview. Financial terms were not disclosed.

The world's fastest computer today is China's Tianhe-1A, a 2.57-petaflop system that topped the November edition of the quarterly Top 500 list of the world's most powerful supercomputers. The much faster systems planned by IBM include next-generation Blue Gene supercomputers, called Blue Gene/Q, in Argonne and Livermore, and a Blue Waters system at the University of Illinois. The Blue Gene systems run on specially made systems-on-a-chip, while Blue Waters runs IBM's Power7 processors.

The Argonne system, called Mira, will be made available to researchers from industry, academia, and governments worldwide. The Livermore system, called Sequoia, includes 1.6 Pbytes of memory and 1.6 million cores. The supercomputer will be used for advanced "uncertainty quantification" simulations and weapons science.

IBM has discussed Sequoia and the Blue Waters systems before. Mira, which IBM announced Tuesday, is part of a $180 million federal award for Argonne to shift its focus from nuclear research involving the use of highly radioactive materials to the use of supercomputer-driven research. Argonne is a direct descendant of the World War II Manhattan Project, which led to the development of the first atomic bomb.

The computing power of Mira and Sequoia is expected to boost many forms of research. For example, Sequoia's horsepower makes it possible for meteorologists to predict the path a tornado would take through a town, making targeted evacuations possible, according to IBM. Mira's ability to perform 10 quadrillion calculations a second is the equivalent of each man, woman and child in the United States performing one calculation each second for almost a year.

Argonne and the Department of Energy have selected 16 initial projects for Mira. The research ranges from reducing energy inefficiencies in transportation and developing advanced engine designs to driving advances in energy technologies.

IBM introduced Blue Gene in 2004, as part of a $100 million development effort that included IBM, Argonne, and Lawrence Livermore.