Cray Moving to Intel-based Supercomputers 3 Comments

The most famous supercomputer maker of all time, Cray Inc. (Seattle, Wash.) has moved to Intel processors for the first time in its long history. Originally founded in 1972 by Seymour Cray who practically invented the supercomputer in the 1960s at Control Data Corporation (CDC), Cray Research was sold to SGI in 1996, and then resold to what became the current Cray Inc. in 2000. Since then Cray’s supercomputers have been best known for their high-end systems using AMD’s Opteron processors. Cray’s next-generation supercomputer code-named Cascade, however, will be its first high-end system based on Intel (although Cray has sold the CX1 and CX1000 work-group systems using Intel-based processors for several years).

“Cascade–which is the code name of our next generation supercomputer to be released in first half 2013–will be our first high-end system to be based on Intel’s processors,” said company spokesman, Nick Davis, who verified that Intel’s Xeon E5-2600 processors will form the heart of Cascade. “And the Xeon Phi coprocessors will be available in our Cascade supercomputers as well.”

Cascade’s development is funded by the Defense Advanced Research Projects Agency’s (DARPA) High Productivity Computing Systems (HPCS) program, whose aim is to redesign the interconnection topology for multi-processors in order to be more resistant to failure, and easier to program, than currently available supercomputers. And Cray’s next-generation Intel-based Cascade supercomputer will make use of this revolutionary new interconnect technology, code-named Aries, as the successor to its current Gemini interconnect.

Earlier this year, Intel purchased Cray’s interconnect hardware development program and related intellectual property for $140 million. Aries is reported to incorporate the PCI-Express (PCIe) interconnect topology over the previous point-to-point links used with Gemini, enabling Intel Phi coprocessor boards to have an equal status with the main Xeon E5-2600 processors that form the backbone of the Cascade supercomputer.

So far Cray has pre-sold four Cascade supercomputer systems, the latest being a $21 million installation for the Commonwealth Scientific and Industrial Research Organisation (CSIRO) at the Pawsey Centre (Perth, Australia). It will use the Xeon Phi to process the massive data streams from the world’s largest radio telescope–the Australian Square Kilometer Array Pathfinder (ASKAP). Earlier this month, Cray also announced another $12 million Cascade supercomputer sale to the Finnish IT Center for Science Ltd. (CSC, Espoo) for use in studying climate change, energy generation, materials science, gene interactions and medical research. Last month, Cray announced a third $40 million Cascade sale to the U.S. Department of Energy’s National Energy Research Scientific Computing Center (NERSC) for climate modeling, biology, environmental sciences, combustion, materials science, chemistry, geosciences, fusion energy, astrophysics, nuclear and high-energy physics, and scientific visualization. Cray’s first two Cascade sales were to the Academic Center for Computing and Media Studies (ACCMS) at Kyoto University (Japan) and to the University of Stuttgart’s High Performance Computing Center Stuttgart (HLRS, Germany). All five systems are expected to be delivered in the first half of 2013.

________________________________________________________________

Colin Johnson is a Geeknet contributing editor and veteran electronics journalist, writing for publications from McGraw-Hill’s Electronics to UBM’s EETimes. Colin has written thousands of technology articles covered by a diverse range of major media outlets, from the ultra-liberal National Public Radio (NPR) to the ultra-conservative Rush Limbaugh Show. A graduate of the University of Michigan’s Computer, Control and Information Engineering (CICE) program, his master’s project was to “solve” the parallel processing problem 20 years ago when engineers thought it would only take a few years. Since then, he has written extensively about the challenges of parallel processors, including emulating those in the human brain in his John Wiley & Sons book Cognizers – Neural Networks and Machines that Think.

 
Posted on by R. Colin Johnson, Geeknet Contributing Editor
3 comments
Richard Rankin
Richard Rankin

I remember the late Seymour Cray saying "If you want to plow a field, do you want the plow pulled by two bulls or 1000 chickens?" Maybe Cray should refer to Xeon Phi cores as "chickens" or call the machine "Henhouse". I guess names like "Jaguar" sound better to the marketing team.

Richard Rankin
Richard Rankin

I'm glad to hear about this because I can use the technology. Who can afford these at $20+ million? How many do they sell in a year? This stuff interests me greatly but this has got to be a drop in the market bucket. Oracle Database, Analysis and storage machines need some competion, over-priced at a million apiece (they are Intel-based). The real market for this technology will be in business intelligence, web market intelligence, algorithmic and high-speed trading, insurance calculations, etc. This stuff is their bread & butter. A guy just went to prison fro leaving a major brokerage and taking code with him. Small and mid-size firms will want small to mid-size analytic machines and clusters. The ability to run a dozen patter-matching algorithms on massive amounts of data. The ability to collect that data and manage with hierarchical storage. This is off-the-shelf hardware today! In 5 years I will equal the Cray with a $10K-$20K machine built with parts I order from Amazon. Software is the issue but the algorithms are there. Writing custom software will be big bucks. "If you build it, they will come." I need access to hardware to start designing machines and software. I'm using Nvidia compute boards now - 2500+core boards coming later this year. Come on Intel, let's get this show on the road!

R. Colin Johnson
R. Colin Johnson

By wooing Cray over to its microprocessors, Intel is aggressively expanding its dominance in massively parallel supercomputers--just-in-time for its Many Integrated Core (MIC) architecture starting with the 50+ core Xeon Phi coprocessor due out later this year.