Supercomputers made by Cray still account for 14 of the top 25 on the newest list, which will be released to the public Thursday in advance of the Supercomputing 99 conference. But the list, which a group of supercomputing researchers update every six months, shows some bad news overall for Cray and its beleaguered parent company, SGI.
SGI and Cray held 188 of the 500 spots in the June list, but that share slipped to 133 on the current list. Big Blue, meanwhile, took 23 new spots on the list to reach a total 141, and Sun increased 18 places to 113. However, SGI and Cray machines still account for a larger fraction of the total computing horsepower on the list.
But the future holds many questions for Cray. For one thing, it's been three months since SGI announced a reorganization in which SGI said Cray would be set on its own. Despite news that the company had a financial partner to help with the job, the deal still is in negotiation with the as-yet-undisclosed company, said Bill Minto, director of product management for Cray.
"We're in active negotiations on the sale, and we expect that to go through," Minto said.
"Its current position is tenuous," said International Data Corporation analyst Debra Goldfarb. "A lot depends on who buys Cray. There are a number of potential suitors, and one in particular that has a formal, qualified bid going through due diligence right now."
One of the factors in the negotiations is splitting up intellectual property such as patents and ironing out the technology-sharing agreements, Minto said.
Cray has been "nicely profitable" except for the most recent quarter, which was hampered by delayed purchases, Minto said.
"SGI took the steps needed to turn Cray around," said communications director Steve Conway, noting that the employee count was cut from 4,500 in 1996 to 850 today. SGI also moved Cray into using more mainstream technology, where possible, in order to cut costs.
The top three computers on the list--made by Intel, IBM, and SGI, in that order--are all products of the Energy Department's nuclear weapons program. However, computer experts warn that the ranking is based on a single performance measurement, a mathematical calculation called Linpack, which doesn't necessarily reflect how well the computers will do at a wide variety of real-world jobs.
Another uncertainty is how well Cray's technology will fare. The company has two product lines: one the "highly parallel" T3E design with lots of processors, and the other the "vector" SV1 design with fewer but faster processors. Those product lines will be merged into a combination machine due in the second half of 2002, the SV2, Minto said.
The SV2 is based on a hybrid of traditional Cray designs with SGI technology that allows thousands of processors to communicate quickly, Minto said. The goal for the SV2 is for users of either of the current designs to be able to upgrade painlessly, he said. Those upgrades are an issue because software must be fine-tuned in order to work well on the machines.
Cray's hardware record is mixed, Goldfarb said. The T3E, when introduced a few years ago, was a very successful machine. But Cray's current SV1 is hindered because it's not fast enough at transferring data internally. And in the vector market, "They don't have a high-end product to sell today," she said.
The SV2 has an innovative design, but--largely as a result of all the new hardware--Cray has delayed it several times, Goldfarb said.
One supercomputing expert questioned whether the new design would arrive in time. "That's a very protracted schedule. Nobody's going to wait until then," he said.
Another cloud on Cray's horizon is competition with IBM, Japanese companies, and even potentially SGI.
IBM's ascendance doesn't worry Minto. The market Cray can address is worth about $850 million today, and "we expect that market to grow," he said. While that's enough for a tidy profit for Cray, "it's too small a target for the large companies."
Goldfarb agrees that Cray is shielded because few companies are interested in its relatively narrow market. "The business model is hard to justify if you're a mainstream server vendor," she said.
SGI won't be a competitor, because the different types of machines are good for different high-performance tasks, Minto said. SGI is better where a programming job can be split among lots of processors that communicate relatively infrequently, whereas Cray machines are designed for super-fast communication between relatively few processors.
"We're focused on providing support for that very high degree of communication intensity," he said.
The Japanese supercomputer makers have faster designs in the current generation of supercomputers, and "there is definitely a lot of pressure out there to lift the tariffs," Goldfarb said. The Japanese companies have been selling aggressively in Europe and Asia, she added.
IDC analyst Chris Willard didn't rule out Cray as a possibility for seizing the high-end market.
"We see that there's a real market for high-end computing. It averages around a billion dollars a year," Willard said. "There's an opportunity for a company to assume leadership in that market."