Staff Reports
NS&D Monitor
1/17/2014
The Department of Energy is trying to leap-frog supercomputing to ultimately reach exascale—computers a thousand times more powerful than today’s best—and the next big step is called CORAL. That stands for Collaboration of Oak Ridge, Argonne and Livermore, three of DOE’s national labs. Not only are the three labs working together, but the effort also requires the cooperation of two of DOE’s Office of Science and the National Nuclear Security Administration.
As part of the collaboration, each of the labs will get next-generation supercomputer that will cost about $125 million and have capabilities exceeding 100 petaflops (100 million billion calculations per second). The CORAL request for proposals was released in early January, and the due date for proposals is Feb. 18. Buddy Bland, who directs the Leadership Computing Facility at Oak Ridge National Laboratory, said a total of about 100 experts from the three labs will be evaluating the proposals in the coming months and helping make the technical choices.
Labs Need Replacement Supercomputer in 2017
According to Bland, the three-way collaboration came about in part because each of the three labs will be in need of a replacement supercomputer at about the same time, around 2017. Oak Ridge (Titan), Livermore (Sequoia) and Argonne (Mira) each has a successful supercomputer that currently is ranked among the 10 fastest computers in the world. Bland said the search for the next and best supercomputers will be a full and open competition, with the project trying to identify two different architectures with great potential. It’s expected that different vendors will come up with the different architectures desired for the CORAL program, but it’s not inconceivable that the same company could present two different designs of choice, Bland said. “We’re going to see what the marketplace will bring,” he said.
ORNL has a longstanding relationship with Cray, which built the Jaguar and Titan systems as well as others housed at the Oak Ridge supercomputer center, and Argonne and Livermore both have longstanding relationships with IBM. But Bland said that doesn’t mean those vendor loyalties will continue through the CORAL process. He said he wouldn’t be surprised at all to see changes. Although the nominal cost of each of the next-generation supercomputers is $125 million, the funding for the program is subject to congressional appropriations. Initially, there will be two NRE (non-recurring engineering) subcontracts awarded, valued at $25 million apiece, to showcase the research and development of new computer systems, “in order to accelerate technology, improve capabilities, improve application performance, and lower the total cost of ownership of the delivered systems.”
Subcontracts Expected to be Awarded This Year
Those subcontracts will be issued by Lawrence Livermore, but the three laboratories will jointly negotiate them and evaluate their merits. Bland said the three labs are all equals in the CORAL project. Each of the labs will award a subcontract to build its next supercomputer, with delivery anticipated in the 2017 timeframe. ORNL and Argonne will choose machines with different architectures, and then Livermore will choose one or the other architecture for its next-generation supercomputer.
Both the NRE and build subcontracts are expected to be awarded this calendar year, Bland said. The new Oak Ridge supercomputer will not require a new building, but it will mean upgrades in the electrical power and chilled water capabilities, he said. Bland said the new supercomputer will be located in the bottom floor of the annex that was constructed at the back of Building 5600 a few years ago. It has never been occupied, he said.
