If Cray were the federal government, everyone would be singing its praises this year. In 2009, the iconic supercomputer maker, for all intents and purposes, balanced its budget and ended up with more cash reserves than it has ever had. But Cray is a business, and the thin line between profit and loss still represents the boundary between success and failure.
First to the numbers. For 2009, Cray reported a record revenue of $284 million, against a minuscule $600 thousand loss. That’s a significant improvement over 2008, where the company saw just slightly less revenue, $282.9 million, but suffered a net loss of a $40.7 million. Better yet, Cray finished the year debt-free, having retired all its convertible notes. In addition, the company now has $113 million in cash and short term investments, roughly twice what it had available at the end of 2008.
Cray CEO Peter Ungaro pointed to the growing sales of custom engineering, the CX1 deskside system, and the XT mini (XT5m) systems as three areas where they were able to make up for some lost ground in their bread-and-butter big systems sales. Service revenue, driven by rapid growth of custom engineering work, grew by 33 percent year-over year, contributing $84.9 million to the bottom line.
While Ungaro was disappointed about not being able to deliver an outright profit, he seemed especially pleased that 2009 revenue was less dependent on sales of a few large machines, and more evenly spread across smaller systems (the CX1 and XT minis) as well as service revenue. Because of Cray’s reliance on big supercomputer contracts, which tend to occur at somewhat long and unpredictable intervals, the knock on Cray from investors has always been that its sales were “lumpy.” With a more diversified supercomputing systems and services portfolio now in place, some of those lumps are going to be smoothed out.
The blame for the $600K profit shortfall was attributed to the DARPA contract fumble. As reported in January, a contract modification for the Phase III HPCS work, related to a missed milestone, was delayed and cost Cray around $15 million in R&D compensation. Had the contract mod been completed during the fourth quarter and the milestone achieved, the company would have easily been profitable for the year.
The HPCS contract mod is now in place. Under the revised agreement, $92.5 million worth of R&D reimbursement will be shuttled to Cray over the next three years, contingent, of course, upon the completion of specified milestones. The end game for Cray’s HPCS effort is deliver the Cascade supercomputer prototype near the end of 2012. According the Ungaro, they expect to deliver Cascade systems to customers shortly thereafter.
With 2009 now in the rear view mirror, Cray is forecasting 2010 will be a year of (surprise) record revenue and profitability. To help fulfill that prediction, the company is launching two new supercomputers this year: the XT6 and the next-generation “Baker” supercomputer. The XT6 supers and the associated XT6m minis will incorporate AMD’s upcoming 8- and 12-core Magny-Cours Opteron processors. Those systems were previewed at SC09 and are currently on track to be released into the wild in the second quarter of this year.
Right on the heels of the XT6 is the company’s next-generation XT, codenamed “Baker,” which is scheduled for release in Q3. The new system encompasses both new interconnect hardware and a new software stack. “This will be, without a doubt, the most significant upgrade to our XT line since its inception in 2004,” Ungaro told investors. “And we believe Baker will be a very competitive offering at the high end of the supercomputing market.”
Hardware-wise, Baker is based on an XT6 compute blade, but with Cray’s new “Gemini” interconnect included as a replacement for the current SeaStar2 technology. Gemini, which promises better performance, reliability and features, is still in development, but is supposedly on schedule for its Q3 debut. Apparently Cray has built the XT6 blades such that the SeaStar2 hardware can be upgraded to Gemini if need be. An XT6-Gemini upgrade could presumably be used as a backup plan by would-be Baker customers if the launch of the new system runs into glitches.
Cray, of course, is hoping this will not be necessary. Revenue and profitability in 2010 will hinge upon the company hitting its Baker launch in the third quarter. Theoretically that should allow enough time to install the new systems for its early customers (e.g., NERSC and KMA) and realize the revenue after those systems are accepted. Because of the timing of the Baker launch, the company is already telling investors to expect a “significant majority” of 2010 revenue to be recognized at the end of the year.
Later this year, Cray also intends to bring a new line of HPC systems to market, positioned between the CX1 and the XT6m. No details were forthcoming, but the idea is to basically fill the gap between the CX1 microcluster and the XT minisuper, further diversifying its product portfolio.
Cray is also looking to expand its service sales this year, forecasting its total revenue at around $110 million. Again, they’re looking at the customer engineering group to drive most of this growth.
Interestingly, one of the first custom engineering contracts in 2010 involves a non-HPC application. Although Cray execs were not at liberty to name the customer or give out many details about the contract, apparently a large commercial company tasked Cray to research and prototype a future infrastructure for a cloud computing-type datacenter. Significantly, the contract represents Cray’s first foray outside of the supercomputing arena.
Some extra income may be captured from the second year of government stimulus funding, both in the US and elsewhere. Last year, the Jaguar upgrade at ORNL was funded by stimulus money, propelling the system to the top of the TOP500. Ungaro hinted there is at least one more stimulus-related US contract in the pipeline, as well as some other possibilities globally. But on the whole, he sees the HPC stimulus effect ramping down in 2010, even as the overall strength in the government sector looks good for supercomputing.
Meanwhile, the competition at the high-end of HPC is heating up. IBM is launching its POWER7-based HPC servers — the IBM Power 755 — this month in advance of the XT6 and follow-on Baker system. Ungaro said he thought the new IBM gear will be a good product line, but he feels Cray is strongly positioned with Baker and the XT6 line to compete head-on with its arch-rival. Cray also has to consider that competitors like SGI, and its new Altix Ultraviolet line, and Bull, with its bullx supers in Europe, also may cut into Cray system sales. Making a living off supercomputers has never been easy, and 2010 will be no exception.