Jaguar, Roadrunner Race to Be Fastest
Just five months after IBM's hybrid Roadrunner became the first supercomputer to break the lofty petaflop barrier, a second, more traditional machine has made the same leap.
And at least one industry watcher said the move of Cray's XT Jaguar supercomputer -- with the help of a $100 million upgrade -- into the petaflop realm is swinging the doors wide open for other systems that are on the verge of following it to a new level of power and speed.
"We'll start to see more and more machines passing this point," said Jack Dongarra, a co-creator of the Top 500 list and a distinguished professor at the University of Tennessee. "I would not be surprised to see one of the [IBM] BlueGene/P machines passing this point soon. We broke though this mark. That's a big thing. Now everybody follows that first machine to enter. We'll have a whole wave of machines cross that barrier."
The U.S. Department of Energy announced earlier this week that the latest implementation of the XT Jaguar supercomputer, housed at the DOE's Oak Ridge National Laboratory in Oak Ridge, Tenn., hit a peak performance of 1.64 petaflops, or more than a quadrillion mathematical calculations per second. Last June, IBM's Roadrunner was the first to break the barrier, hitting a sustained speed of 1.026 petaflops. A few weeks later, Roadrunner was officially crowned the fastest computer in the world when it made the top spot on the semi-annual Top500 List of supercomputers.
The latest Top500 List of supercomputers is expected to be released next Tuesday at the Supercomputing Conference in Austin, Texas. Dongarra told Computerworld that he won't divulge which machine will be at the top of their list but he called Jaguar a "very powerful machine."
Steve Scott, chief technology officer at Cray, Inc., said in an earlier interview that while the Cray machine surpassed Roadrunner's June numbers, the IBM machine, which is operating at DOE's Los Alamos National Laboratory, could have been upgraded since then.
Dongarra said IBM's hybrid supercomputer has been updated since June, but would not disclose its latest performance numbers. IBM today would not say whether Roadrunner has been upgraded since June.
"Roadrunner is a rather specialized machine. It's a hybrid [system] with AMD Opteron processors, and the Cell chip with its PowerPC core and eight added processors," said Dongarra. "To have three different processors, means you need three different programs, one for each type of processor. Jaguar, is a more traditional set up. It looks much more like a conventional parallel system. That translates into an ease of programming. You have to write one program in a traditional parallel fashion. Jaguar will be easier to use and easier to program by a wider group of people. It's also simpler to port existing software to that platform."
A source inside IBM, who asked not to be named, noted that Jaguar uses 7 megawatts of power, while Roadrunner uses 3.9 megawatts. "A system that does the same amount of work for twice the energy is a dinosaur the same day it's introduced," he added.
Jaguar isn't considered anything of a dinosaur at the Oak Ridge lab, noted Buddy Bland, project director for the leadership computing facility at Oak Ridge. He said that it's exciting to think that the lab's supercomputer may lead the Top500 List, but it's even more important that it allows them to get more work.
"This is not about having a big toy. This is about delivering the most capable scientific instrument in the world,' said Bland. "It's much more about what this is going to do for science."
He noted that this year, the supercomputer has been used for 30 different research projects that range from new battery designs to fusion energy research, and climate studies.
Cray engineers have been upgrading the scalable Jaguar machine at the Oak Ridge lab since 2004. The scalable supercomputer now has 362 terabytes of memory and a 10-petabyte file system.
The $100 million upgrade involved adding 200 cabinets, taking the system from 84 cabinets to 284, according to Scott. Each cabinet can hold up to 192 separate quad-core Opteron chips from Advanced Micro Devices Inc. The entire machine runs 45,000 chips, which adds up to 180,000 processors, Scott noted.
Bland noted that the system takes up about 5,700 square feet and is slightly larger than the size of an NBA basketball court. It takes about 70 people to operate the supercomputer and work on applications, as well as run the network and storage. Jaguar also has 6,000 miles of interconnect cables linking the processors together and 148,000 memory modules.