Fight for supercomputer title
SGI and IBM are both claiming bragging rights for the fastest machine. Who's the real king of the hill? Depends on who's doing the testing.
- Shankland covered the tech industry for more than 25 years and was a science writer for five years before that. He has deep expertise in microprocessors, digital photography, computer hardware and software, internet standards, web technology, and more.
Compaq also jumped into the large scale computer race today, announcing a computer architecture with Sandia National Laboratories. The advanced "clustering" technology will target areas such as data warehousing, scientific visualization, and Web search engines. (See related story.)
But the real slugfest is between Silicon Graphics and IBM. Silicon Graphics and Los Alamos National Laboratory said their Blue Mountain machine is the fastest, less than two weeks after IBM and Lawrence Livermore National Laboratory made the same announcement with their Blue Pacific computer.
The issue of who's king of the hill depends on which hill you're talking about. Energy Department officials spent several minutes at a news conference today tiptoeing around the subject of how to measure which computer is fastest.
"Both [are], in a certain sense," depending on which performance benchmark you're using, said DOE Undersecretary Ernest Moniz.
But Rick Beluzzo, chairman and chief executive officer of SGI, felt no such restraints. "There is only one that can be in first place, and we're in first place. In my humble opinion, IBM announced early because they knew they were going to lose first place," Beluzzo said.
IBM again touted Blue Pacific as the world's fastest in a news release Today. IBM said it has increased its presence in the supercomputing arena while SGI has lost ground.
Asked to respond to Beluzzo's claims to the No. 1 spot, an IBM spokesman downplayed the performance differences between Blue Mountain and Blue Pacific.
"The intent of the [Energy Department's Accelerated Strategic Computing Initiative] is to push the boundaries of supercomputing," and any machine as powerful as IBM's or SGI's is impressive, the spokesman said. "In the end, we're working toward the same goal."
Using one performance measurement called SPPM, the SGI/LANL's Blue Mountain machine wins out by about 10 percent in a physics benchmark over IBM's Blue Pacific machine, said Steve Younger, associate director of LANL's nuclear weapons program.
But the list of the top 500 supercomputers uses a different measurement, a calculation called Linpack. Linpack hasn't been run on Blue Pacific, which is still being reassembled at the Livermore lab. Its first performance measurements were run when the computer was first constructed at an IBM facility in Poughkeepsie, New York.
Using the Linpack benchmark, the previous top computer is another DOE-funded machine at Sandia National Laboratories in Albuquerque, New Mexico. That machine, Red Mountain, has a Linpack score of 1.34 million. Blue Mountain, though, has a Linpack of 1.61, Younger said.
Supercomputing announcements have picked up recently, in part because of the SC 98 supercomputing conference this week in Orlando, Florida.
Both Blue Pacific and Blue Mountain were developed for an Energy Department program to replace actual nuclear tests as the way the United States makes sure its nuclear weapons work as advertised as they grow older.
DOE launched its Accelerated Strategic Computing Initiative (ASCI) as a way to push U.S. computer companies to advance supercomputer power using commodity processors.
IBM's Blue Pacific cost $94 million and is based on a 5,856-processor version of its RS/6000 series. SGI's Blue Mountain cost $120 million and uses a 6,144-processor system based on its Origin 2000 series.