January 3, 2005
St. Paul, Minn. — Blue Gene/L project manager Steve Lewis steps into a room thundering with a massive cooling system at IBM's Rochester plant.
"Don't touch anything," he says. "It's an electrostatic sensitive area, so none of us [should] touch anything."
Lewis heads for several rows of silvery, but unremarkable looking computer racks, the area where Blue Gene/L is built and tested.
"The record setting computer was the first four racks in each of these four rows," says Lewis.
And that's only a quarter of the machine headed for its first customer. Blue Gene/L set two records this fall, the latter coming in at nearly 71 trillion calculations per second. Lewis says the record prompted calls from proud IBM alums in the area.
"They're very excited about what's come out of Rochester. The community's very excited. It's a big deal for the community, that this is where the world's speed record sits," says Lewis.
Development of Blue Gene/L started in New York, but it's being manufactured in Rochester. Lewis says all the pieces necessary to make the world's fastest supercomputer happen to be in IBM's Rochester plant, which manufactures servers.
Supercomputers like Blue Gene/L are used to assist a variety of tasks, including automobile and aircraft design, oil exploration, intelligence analysis and weather forecasting. They can cost tens of millions of dollars.
Blue Gene/L's launch customer is the Lawrence Livermore National Laboratory in California. Touting the speed record last November, Energy Secretary Spencer Abraham said the project is a deliberate investment in high performance computing to help promote U.S. competitiveness.
Dave Parry, who heads the supercomputer operations of California-based Silicon Graphics, says there's been a chorus of concern that the U.S. has fallen behind in supercomputing.
"There have been numerous studies over the last two to three years from almost every agency in the U.S. government that has anything to do with high-end technical and scientific computing, all of which have come to virtually the same conclusion," says Dave Parry, of Silicon Graphics.
The conclusion, says Parry, is that supercomputers are a key tool for scientific discovery and the U.S. is not spending enough on them.
The concerns extend beyond government. The Council on Competitiveness, a non-profit organization of corporate, academic and labor leaders says high performance computing is key to American industry's ability to compete.
Steve Conway says government interest in the industry dwindled in the 1990s. Conway works for Seattle-based Cray, Inc., indirectly a successor to the old Cray Research, the Twin Cities-based firm that dominated the industry for years.
"It looked like the U.S. was just substantially in the lead. Then all of a sudden in early 2002 [there was] a little bit of a Sputnik-like shock, when we found out that Japan had produced a supercomputer that in practice, on real problems, was anywhere from 10 to 100 times faster than anything the U.S. had," says Conway.
Conway says U.S. computers still lag Japanese performance on real-world problems, despite the new speed records based on testing software.
The Japanese computer, called Earth Simulator, held the speed lead for more than two years. Stung, the U.S. is now trying to catch up.
Congress recently authorized $165 million in new funding over three years for Energy Department research and development to advance high end computing. On an annual basis, that represents only about one percent of total supercomputer sales.
Still, the effort is clearly making a mark on the industry and in Minnesota. IBM is not the only supercomputer maker with a Minnesota presence to get a hefty government contract. Silicon Graphics, which bought Cray Research in 1996, got one too.
At Silicon Graphics' Eagan facility Karl Feind heads into the tunnel between two racks of computer cabinets as tall as he is. Inside, cables curl and hang along the sides like so many white snakes.
"And we can walk right through the middle of it here, and you can see it's a canopy over our head, where the cables connect the two hemispheres," says Feind.
Software engineers here worked on a new supercomputer built for NASA. Named Columbia in honor of the space shuttle astronauts who died, the machine set a new speed record of nearly 43 trillion calculations per second in October. Feind says there were lots of smiles that day.
"There were high fives everywhere in the hallway. There was a sense of awe that the software and hardware system we had created could run a single problem for eight hours without failing," says Feind. And he says there was a sense of relief.
SGI's Dave Parry says these are bullish times for the industry. He says the most recent quarter saw a doubling of sales of SGI's latest supercomputer model.
"And actually a quadrupling of product bookings in that product line. We're seeing that in combination with substantial wins in large deals, really around the world," says Parry.
But that strength is not enough to make SGI profitable. As a whole, the company has been losing money.
Seattle-based Cray, Inc. has also gotten some impressive government contracts and officials expect a big jump in sales next year following a decline this year. Like SGI, Cray has remnants of the old Cray Research both in the Twin Cities and Chippewa Falls, Wisconsin.
Steve Lewis of IBM is a little more circumspect about Blue Gene/L's sales, even though IBM is already the biggest supplier of supercomputers.
"It's a new machine, and so people have to kick the tires a little bit before they're ready to engage in a big fashion," says Lewis.
Analysts also have mixed views.
Long time supercomputer industry observer Gary Smaby is skeptical the industry is on the cusp of boom times, despite renewed government interest.
"I think it will help sustain the jobs that are already here. There will be continued demand for machines like Blue Gene and a machine like Columbia, not to mention some of the Cray machines, but that business is relatively flat," says Smaby.
IDC analyst Addison Snell is somewhat more optimistic. He predicts the industry will show sales growth of 15 to 20 percent this year. He expects continued growth, but he says demand for the largest and most powerful machines -- which would help SGI and Cray most -- will grow slower.
"I wouldn't rule out entirely the possibility that Cray or SGI could really get going again, and one of those two could get on to a really rapid growth spurt," says Snell.
Steve Conway of Cray says the industry is likely to grow here in Minnesota, maybe even faster than the industry nationwide, but it's still much smaller than it used to be.
"There aren't nearly as many people employed in this industry in Minnesota as there were 15 years ago, and that mainly is because Cray Research was a much bigger company at that time, and employed somewhere [around] 1,000, [or] 1,500 people in Minnesota alone," says Conway.
Supercomputing is a niche industry and may never be a major source of job growth for Minnesota. But the state again had a hand in setting the latest speed records, and that legacy may well continue.