Supercomputer Champ Remains (For Now)

Supercomputers are a lot more super than they were just six months ago.

According to the latest and 23rd edition of the TOP500 list, the number of supercomputers with
greater than 1 teraflop processing power jumped by 86 percent to 242, up from 130 six months ago.

The list of top computers gives IBM bragging rights on clusters, as well as Intel on its processors. China also cracked the list’s top 10 for the first time. But the top spot in the list remained the same for the third straight year. At 35.86 teraflops, Japan’s NEC Earth Simulator remains the most powerful supercomputer on the planet today.

The latest list, which is jointly compiled by authors at the
University of Mannheim, Germany, NERSC/Lawrence Berkeley National
Laboratory and the University of Tennessee, leaves the current supercomputing champ unchallenged.

It is, however, unlikely that the Japanese supercomputer will retain its top spot for much longer, with renewed efforts in the U.S. to regain the top spot with a supercomputer that is ultimately forecast to hit 250 teraflops. Not to mention continued gains by IBM’s supercomputing initiatives.

IBM hailed the list as a testament to its own
supercomputing prowess. Big Blue said it is now the number one manufacturer of supercomputers, with 44.80 percent of all systems on the top 500 list; that works out to about 50.12 percent of the list’s installed performance.

The only systems vendor that came even remotely close to IBM’s position is Hewlett-Packard , which grabbed a number two ranking with 28 percent of all systems, representing 18.5 percent of the installed performance. Beyond HP and IBM, no other vendor held more than 6 percent penetration on the list.

A pair of IBM’s BlueGene/L system prototypes, jointly developed by IBM with the Lawrence Livermore National Laboratory (LLNL), held the number four and number eight spots on the list. IBM called the Blue Gene/L prototype, “a radical new design for supercomputing.” The system is small and uses less power than comparable existing supercomputing platforms, according to IBM. At peak processing power of 16 teraflops, the number four ranked Blue Gene/L DD1 IBM prototype is powered by over 8,000 PowerPC processors.

The existing system is only at 1/16 of its total planned
final capacity and has jumped 69 spots up the list in the past six months (73 down to 4 since November).

“Whether we are talking about improving the accuracy of weather forecasts, designing better automobiles or improving disease research, we are seeing
the advent of a new supercomputing age.” said Dave Turek, vice president,
Deep Computing for IBM, in a statement.

The group said the final BlueGene/L system that will be installed at LLNL is expected to capture the top spot on the list by as early as the end of this year. IBM also holds the number six spot on the list with a p690 system powered by its own Power4+ processors. The Power4+ is also part of IBM’s p655 servers, which GM recently bought to use in auto development and virtual crash simulations.

LLNL is currently home to the world’s number two supercomputer, the Thunder Linux cluster, which boasts 19.9 teraflops that are produced by 4,096 of Intel’s Itanium 2 processors. Clusters of all breeds also made a strong impact on the list, representing 56 percent, or 281 out of 500 total. Out of 281 clusters, 150 Linux clusters were built by IBM.

Standard CPUs still power many of the clusters. Intel-based processors represent 57 percent (287 of 500) of all processors on the list, up from just under 24 percent (119 of 500) one year ago. IBM’s Power processor is the second most popular family at 15 percent, followed by PA Risc at 11 percent and AMD at just under 7 percent.

The list comes out just after supercomputer manufacturer Cray claimed in a press release that its processors were the most powerful processors on the list and had the highest efficiencies. Cray’s top ranked computer on the top 500 list came in at number 20 at the Oak Ridge National Laboratory (ORNL), the site where the company, together with the U.S. government, plans to build a 250 teraflop behemoth. Last week, Cray reported that its X1 supercomputer had the highest reported scores in a government-sponsored test using a new HPC Challenge benchmark test.

The top 500 list uses the Linpack tool to benchmark supercomputer performance. The HPC Challenge benchmark test is seen by some as a more accurate indicator than Linpack of overall HPC performance.

“Linpack is useful, but no single test can accurately reflect the overall performance of HPC systems,” Jack Dongarra University of Tenesse and one of the top 500 list compilers as well as contributor to the new HPC Challenge benchmark test said in a statement last week. “The HPC Challenge benchmark test suite stresses not only the processors, but the memory system and the interconnect. It is a better indicator of how an HPC system will perform across a spectrum of real-world applications.”

News Around the Web