• E-mail this page
• |  Print this page
• | 
• | 
• | 

At Cray, Good News Is Hard To Savor

On April 10, Cray said the Nasdaq warned it could be delisted from the exchange, pending a hearing, after Cray failed to file its 2005 annual report on time. The report is being delayed while the company sorts out whether it needs to restate its 2004 net loss and increase that number by as much as $3.3 million after Cray booked revenue from product development work it did for a customer, even though it failed to bill the customer for the order. The error also happened to occur in a year in which Cray lost a whopping $203.9 million.

One potential fix would be to account for the charge in 2005, but the company can't file last year's annual report until its auditors understand the problem, Cray CEO Peter Ungaro said in an interview Thursday. The company hopes to set a hearing date with Nasdaq "any day now" and make its case for an extension within the next few weeks, he said.

The threat of being delisted is only the latest indignity for Cray, whose founder, Seymour Cray, revolutionized high-performance computing in the 1970s and '80s by building the fastest machines around before his death following a car accident 10 years ago. On a preliminary basis, the company lost $9 million on sales of $65 million during the fourth quarter last year--the latest in a string of losses. Its stock price is in the tank--shares of Cray (Nasdaq: CRAY) closed April 13 at $1.56, up 7 cents. Last November, former chief scientist Burton Smith left the company to join Microsoft, departing shortly after the supercomputing industry's most important conference of the year in Seattle. Last summer, Cray promoted Ungaro, its president and a former IBM exec, to CEO, replacing James Rottsolk.

Can the company turn it around? Ungaro says Cray has made needed changes in its technology, staffing level, and other expenses to return to profitability later this year. As expenses have fallen, it's also seen an increase in outside project funding, he says.

On the plus side of the ledger, Cray's machines remain valued by customers including the Homeland Security Department, NASA, the military, and the country's national labs and universities, even after multiple corporate incarnations over the years, including ownership by Silicon Graphics and a buyout by Tera Computer six years ago. Cray's Red Storm supercomputer at Sandia National Laboratory is the world's sixth fastest, running at 36.2 teraflops, and it's on track for 50 teraflops of performance this year. In the emerging market of India, two computing centers signed deals for Cray systems last month. And Cray struck a deal late last year with chipmaker Advanced Micro Devices to supply it with processors through 2010, ensuring that Cray has access to the computer industry's hottest chip platform.

Moreover, Cray is one of three computer companies (along with Sun Microsystems and IBM) competing for Defense Advanced Research Projects Agency funds said to total $250 million a year over five years that will be awarded in July. The award is part of a DARPA high-performance computing research program to build a system capable of a petaflop of performance by 2010. That's 1,000 trillion operations per second, or three-and-a-half times more power than in today's fastest supercomputer, an IBM system rated at 280.6 teraflops. DARPA is expected to select one or two companies to continue developing technology under the next phase of the program.

Now for the minus side. AMD's Opteron technology is just one of four computing architectures Cray has to support, as I wrote about last month. It also sells supercomputers based on vector processors, field programmable gate arrays, and a specialized architecture tuned for mining unstructured data that's used by Homeland Security, according to the company. Supporting those four technologies keeps research and development costs high: Even after slashing R&D spending by more than half in the third quarter of last year--never a good sign--Cray still spends nearly 20% of its revenue on R&D, well above industry norms. Ungaro warns against reading too much into quarterly fluctuations in R&D spending, since they can include large expenditures for plants that can vary by quarter.

The upshot of Cray's new "adaptive supercomputing" road map is that by 2010, it plans to sell one system capable of housing boards that contain vector processors, scalar Opteron chips, and FPGAs. Compilers and other software would be able to route pieces of applications to the processor type that can run them fastest, says Cray Senior VP Jan Silverman. Next year, Cray plans the intermediary step of releasing a common programming environment for all its platforms.

If Cray is one of the companies DARPA selects for phase three of its petaflop program in July, it would be a strong vote of confidence in Cray's approach, embodied in a machine code-named "Cascade." Government and commercial buyers like platforms that look like they have a future.

At the SC05 conference in Seattle last November, Cray CTO Steve Scott told me AMD's road map for processors, boards, and packaging could get Cascade to a petaflop by 2008. Ungaro says the Cascade system and adaptive supercomputing plan will be key to Cray's proposal for the July award. Submissions are due next month.

If Cray can ace IBM or Sun in the DARPA competition, manage its way out of its Nasdaq embarrassment, and return to profitability, its management team might take their place in the annals of turnaround artists. Cray's plans look solid on paper, but the company keeps getting egg on its face. Stanching the flow of bad news could be a good first step toward picking up market share by convincing customers that Cray computers belong in places other than their rarified niche.

« Software Security Groupies Kiss And Tell | Main | Daily News Podcast For Friday, April 14 »