Top 500: Good Show for Nehalem, Germany, Arabia

The TOP500 list of supercomputers is always a big deal in the never-ending quest for bragging rights among hardware vendors. For this latest list, the 33rd, Intel scored a big win.

Just three months on the market, and Nehalem already has 33 positions among the 500 fastest supercomputers in the world; 22 for the Xeon X5500 and 11 for the E5500. All told, Intel (NASDAQ: INTC) holds 80 percent of the list, with 393 x86-based supercomputers and six Itanium-based computers.

“Obviously we’re delighted to have 33 systems just three months after launch,” an Intel spokesman told InternetNews.com. “It shows how powerful the Nehalem is and how easy it is to build supercomputers with off-the-shelf components rather than custom solutions.”

But all isn’t lost for AMD (NYSE: AMD). Two supercomputers sporting its brand new six-core Opteron, also known as “Istanbul,” made the list, and Istanbul is even newer than Nehalem. All told, 43 computers on the list are Opteron-based, including the two fastest computers in the world.

Holding first place again is the Roadrunner system at the Department of Energy’s Los Alamos National Laboratory. Roadrunner, built by IBM in June 2008, features a combination of Opteron and Cell Broadband Engine processors and is the first system ever to break the petaflops per second Linpack barrier.

In second place is the Cray XT5 Jaguar system, also Opteron-powered and run by the DOE. This one is at the Oak Ridge National Laboratory. It was measured at 1.059 petaflops per second when it was first installed. Due to its heavy workload, no further measurements were possible.

New supercomputers in the top 10

There are four new entries in the top 10, two in Germany. At number three is JUGENE, an 825.5 teraflop system based on IBM’s Blue Gene/P system. At number 10 is JUROPA, a 275 teraflop system built from Bull Novascale and Sun Microsystems SunBlade x6048 servers.

The list has been released as part of the International Supercomputing Conference taking place this week in Hamburg, Germany. All measurements are made with Linpack, a benchmark that dates back to the 1970s, featuring Fortran-based mathematical tests to measure a computer’s floating point computation speed.

While the numbers are impressive, these massive supercomputers usually don’t bring their full power to a single task. In most cases, these computers are partitioned into many smaller systems and used to run dozens of heavy-duty scientific projects. Even a slice of these computers is still faster than anything most people will have access to.

By way of comparison, an Intel Core 2 Duo E6700, a 2.67GHz dual core processor that’s typical of your average home computer, runs at about 12 gigaflops. There are 1,000 gigaflops in a teraflop, and 1,000 teraflops in a petaflop. So these machines are extremely fast.

They are also highly-specialized, notes Martin Reynolds, vice president and research fellow with Gartner. “The continued rise of specialist machines represents the most interesting shift,” he told InternetNews.com. “It shows these architectures are finding a place. They are difficult to use, but the speed you can get to a solution outweighs the time it takes to program them.”

While server sales are in a near free-fall, there is no shortage of horsepower in the supercomputing space, where servers cost millions of dollars. Reynolds isn’t surprised.

“If you need it, the value of a supercomputer is huge. If you are using it to find new oil and gas, you need it quickly. The value is getting the results to market quickly. If these supercomputers can bring your concept to monetization more quickly, then you pay what it takes for them,” he said.

Next page: Other noteworthy stats

Page 2 of 2

Other noteworthy stats

The other two newcomers are a Cray XT5 system called Kraken installed at the National Institute for Computational Sciences at the University of Tennessee and an IBM BlueGene/P system called Dawn installed at DOE’s Lawrence Livermore National Laboratory.

When the list was last updated in November, China made history with its debut in the top 10. The Dawning 5000A was notable not only for being the first Chinese system to make it that high, but also for using Windows HPC, since more than 400 of the 500 on the list run Linux.

This year’s nation of record is Saudi Arabia, which landed at number 14 with an IBM BlueGene/P system at the King Abdullah University of Science and Technology (KAUST) in Thuwal, Saudi Arabia. The school isn’t even open yet. It’s expected to open in September.

The U.S. remains the leader, with 291 systems, the same as last November. The European share dipped from 151 last November to 145 now. Asia gained two systems for a total of 49. The remaining 15 are scattered around different parts of the world, such as South America, the Middle East, Russia and Eastern Europe.

Hewlett-Packard (NYSE: HPQ) kept its numbers lead over IBM (NYSE: IBM) – 212 to 188, but IBM maintained the performance lead. Cray is tied for third place with 20 machines, 10 of them in the top 50. SGI also has 20 machines, including a top 10 machine. Dell (NASDAQ: DELL) has 14 machines, Sun (NASDAQ: JAVA) has five.

Quad-core processors rule the roost, with 383 of the 500 systems running quad-core or better processors. Only four of the 500 use single core processors. There are four systems, including the top one, using IBM’s Cell processor, the same processor used in the PlayStation 3 console.

Just to show how fast things move, the entry level to the list is now 17.1 teraflops per second. Just six months ago, it was 12.64 teraflops. The system at number 500 on this list would have been at number 274 on the November 2008 list. All told, the 500 computers on this list have a total of 22.6 petaflops per second. Six months ago, it was a total of 16.95 petaflops, and a year ago it was 11.7 petaflops.

While a comprehensive list, it’s not truly complete. Many large corporations may be sitting on computer systems that could easily crack the list but won’t admit it. “Financial institutions certainly won’t talk about it. So there are places not on the list as well,” said Reynolds.

Others, like universities, love to brag about their accomplishment because it acts as a lure to students and researchers. “Because it captures the most visible systems, it gives you a picture of the most active parts of the market, so that’s good in a way,” said Reynolds.

The TOP500 list is compiled by Hans Meuer of the University of Mannheim, Germany, Erich Strohmaier and Horst Simon of NERSC/Lawrence Berkeley National Laboratory and Jack Dongarra of the University of Tennessee, Knoxville. Dongarra is the developer of the LINPACK benchmark.

News Around the Web