|
Cray on the comeback
 |
| Happy days are here again for Dave Kiefer, vice president of engineering at Cray Inc. and the one many observers think is the person who deserves the most credit for the rebirth of the supercomputer maker. Kiefer, though, credits an outstanding Cray team.
|
|
|
The once-mighty supercomputer pioneer teetered on the edge of extinction, until some dedicated professionals brought it back.
By Mark Gunderman
They call it the "smoke test." It's the moment that a newly designed and constructed computer component is powered up to see if it works. Smoke signals coming from the circuit board are not a good sign.
That was especially true for engineers at Cray Inc. in November 2001, when it was time to power up the main module for the Cray X1. A disastrous test could well mean a legendary company was going up in smoke.
The team crowded into the room and fired it up.
"No one was overly concerned that we were going to have 'smoke.' We do way too much testing for that," said Dave Kiefer, vice president of engineering for Cray Inc.
Cray Systems Test Technician Chad Philson runs a diagnostic on one of the Cray computers behind him.
But how well the new system worked was going to have a great deal to do with the new Cray's chances of success.
"When you fire up that machine for the first time, the health of it determines when you are going to ship it," said Kiefer.
The brains of the XI proved to be pretty healthy. And now, so too is Cray Inc.
Cray Systems Test Technician Jo Ann Burrows troubleshoots a node module on a Cray computer.
Four years ago this month, the legacy of Seymour Cray broke away from its temporary owner, SGI. Team members arrived at their Chippewa Falls headquarters with little more than pencils in their pockets.
In a relatively short period of time, they succeeded in engineering the rebirth of a legend.
Days of discouragement
The glory days of Cray Research are well known to the high tech leaders of the Chippewa Valley, most of whom were with the company in those early days. Seymour Cray virtually invented the supercomputer and the company was the dominant force in the industry from the early 1970s through the early 1990s, owning roughly 70 percent of the worldwide market.
But new technologies such as Massive Parallel Processing (MPP), started to challenge the Cray dominance, spurring products that were nearly as fast but could be had for a much lower price. Leadership at Cray had dreams of building a multi-billion dollar company. With the end of the Cold War and cutbacks in government spending, it needed to broaden its market to do it.
Cray was doing some soul searching and struggling to survive. Mountain View, Calif.-based Silicon Graphics Inc. (SGI) bought Cray Research in 1996, beginning what turned out to be an unhappy marriage.
Competing goals
The second generation of Cray MPP machines — the T3E — had just shipped, and so had SGI's similar Origin series.
"We were both working on follow-up machines. They killed our follow-up within a month," recalls Gary Shorrel, program manager for the Black Widow, Cray's latest project.
The T90 was canceled about a month and a half later.
Cray's focus had always been the high-end machines based on vector processing — a market much smaller than SGI was targeting with its range of MPP-oriented machines and work stations. Some of the Cray vector people were moved to other projects and the high-end work suffered from staff cuts as SGI tried to stem a tide of red ink.
"It was frustrating because you had two different companies with competing goals," said Steve Scott, Cray's corporate scientist. "There was some of that back in the end of the Cray Research days."
The salvation was to come in the form of the X1 project, begun in the spring of 1998. The idea was to combine traditional vector processing with the concept behind MPP — something no one had done before. The potential was great, but so was the risk.
The former Cray people were trying to get this going while SGI had its corporate mind elsewhere.
"The old Cray days were never going to be re-created," said David Frasch, Cray’s chief technology counsel.
"The worst time for me was probably in the spring and summer of 1998," said Kiefer. "At that time, SGI was losing money quarter by quarter and they were having to make some business decisions on how much development they could support."
Kiefer spent a lot of time advocating for the project. There were fears the X1 would be killed and the remnants of Cray with it, but the project had some friends in high places.
"A number of government customers came to SGI and said, 'We need a follow-on vector machine,' " said Bruce Steger, Cray systems technology manager.
"We were a separate business unit as part of SGI at that time," Kiefer said. "They decided to sell Cray in the summer of 1999."
New life through Tera
Redmond, Wash.-based Tera Computer bought the business unit and the Cray name in April 2000.
"Tera was focused on the high-end capability computing. This was a better fit," said Kiefer.
When the deal closed, the new Cray Inc. had the SV1, a T3E upgrade and a few miscellaneous products to sell.
"The revenue was primarily being generated from service," said Steger.
The whole future of the business was being bet on the X1.
"Everyone knew it was very high risk," said Frasch. "If we had a massive failure, I don't think we could have survived."
The company faced two big challenges: Bringing the X1 product to market before industry advances further limited its market, and keeping the business afloat until the X1 brought in the revenue.
"What you had at that time was a 100-person company purchasing an 800-person business unit," Kiefer said.
Tera didn't have a lot of resources or infrastructure to handle even cutting that many payroll checks. And they needed cash to make payroll.
"We didn't have a lot of products and we didn't have any new products," Kiefer said. "But we did have people in the government very interested in the type of things we do."
"I was worried more from a business aspect whether we were going to make it or not," said Scott, who was confident in the technological concept of the X1.
"During 2001 we were at the bottom of the trough financially and we couldn't pay our suppliers," Scott added.
Shorrel heard stories that there were times when payroll was nearly not met.
On the technical side, the company was in a race with what is known as Moore's Law. To the layman, it roughly translates into the old expression "bang for the buck."
That is, the industry is producing ever-more powerful computers at lower costs. The ratio of price per performance improves rapidly in the business — about 60 percent a year — and every company needs its next generation of products to keep up.
On one end, some companies focus on keeping the costs of the products down. On the other end, Cray tries to increase the power of its machines exponentially, while still selling expensive machines. There are countless companies in between.
While the Cray team was confident the X1 would work, they recognized the need to get it to market as soon as possible. For every six months of delay, the potential market would shrink due to Moore's Law.
"Your margin slips away as time goes on. That's why the schedule is so important. The same product one quarter later you will get less money for," Scott said.
And so the team took on the technical challenges with a lot of hard work. Some concerns included a low yield on manufactured parts — that is, the percentage of circuits that actually work. Some chips obtained from a supplier did not work properly. It took months to work out such problems.
"There were a lot of long, long days put in by people who worked on that problem," Shorrel said.
Meanwhile, Kiefer was dealing with a restless market. Every potential customer raised the question of whether Cray would be around to deliver and service the goods.
Because the federal government was interested in the project, there were constant reviews of progress, and that helped reassure the market.
"There was pretty good knowledge of where we were technically," Kiefer said.
Shipping with Leinie’s
Eventually the team came to the point of firing it up and hoping it didn't smoke. That point is faced with every project, but the moment was more critical for the X1. Most new products are direct descendants of other products whose capabilities are well known. But the X1 was all new.
And it worked. The first one shipped Dec. 31, 2002, with a case of Leinenkugel's.
"An old tradition was revived. That's a relationship that goes back a long, long way, and customers like it," said Frasch.
Cray was not entirely focused on the X1 the whole time. Development of the follow-up product, dubbed Black Widow, is well underway. By early 2002 the company had outlined a product development strategy through 2010.
"For me it was real comfortable having a roadmap we could all work toward," Scott said.
The company had turned the corner by mid-2002.
"We had enough business lined up. We were profitable for a couple of quarters and we knew we were right-sized," Kiefer said.
Cray's stock shot up, orders came in and the dark days of Cray were over.
"You're making money, you know the future, you know what the products are going to be. People were having fun again," said Scott.
The future looks brighter, too, because it doesn't depend on one product. The Black Widow may be next in line to the X1, but Cray also has a government project called Red Storm. It's a computer model meant to assist in nuclear stockpile stewardship. Cray has a contract for the project with the Sandia National Laboratory.
Unlike other Cray products, it uses a common microprocessor.
"That puts us in play in a marketplace where we haven't been a player in a long time," Shorrel said.
But the company is very mindful of not getting overconfident because things have turned around. Everyone knows that Cray will never again have 70 percent of the supercomputer market, and if you can fall from that perch, you can fall from any perch.
"I don't feel you should ever feel comfortable. You can't sit back and relax," Kiefer said.
Mark Gunderman is editor of the Chippewa Valley Business Report. Contact us at (715) 723-5515, or through www.chippewavalleybusinessreport.com.
Email this story
Print this story
|
