Cray revenues slammed in Q1
Come in Baker/Gemini
Nothing illustrates so clearly why supercomputer maker Cray has been expanding into entry and midrange supercomputer markets as the financial results that the company posted in the first quarter ended in March.
With only a high-end product line, Cray had nothing to sell as big HPC shops are waiting for the future Baker systems and their Gemini high-speed interconnect. And thus revenues plummeted 61.9 per cent to $23.4m, and even with DARPA kicking in $12.5m to cover some research and development costs, Cray had a net loss of $11.6m.
Supercomputer product sales were hit particularly hard, falling 84.8 per cent to $9.1m as customers who were not waiting for Baker/Gemini systems nonetheless were waiting for the XT6 and XT6m massively parallel supers - based on AMD's new twelve-core Opteron 6100s that began shipping ahead of schedule in March.
The company's services business, thanks in large part to custom engineering work where Cray does bespoke hardware, software, or data center development, increased by 28.7 per cent, to $19.3m, in the first quarter. While this is very good growth, it just cannot fill in the massive hardware gap.
"Clearly, we had a fairly slow quarter," said Cray's president and chief executive officer Peter Ungaro in a conference call with Wall Street analysts to discuss the quarter. That could be the understatement of the decade.
Ungaro was not being crass or cavalier, though, and quickly reminded everyone that Cray said this was going to happen because of the Opteron 6100 transition and the move to the future Baker machines and their Gemini interconnect, which are not due until the third quarter.
While Cray's revenues were hammered in the quarter, Ungaro said that it was a strong quarter for orders, both in the domestic market here in the US as well as internationally, and that the momentum is continuing in the second quarter. In fact, Cray has two contracts still in negotiation where the company is the last vendor standing; these contracts include product and services spanning from 2010 through 2013 and have an aggregate value of $90m, including services.
The custom engineering business had more than $20m in bookings in the first quarter, too, including a deal where Cray is helping Microsoft design HPC centers. The $20m in orders for custom engineering included two other companies, but Cray didn't say who or dice and slice who spent what.
Ungaro said that Cray has just got the next iteration of the Gemini interconnect back from its foundry partner, Taiwan Semiconductor Manufacturing Corp, and is putting it through the testing phases. Given its name, the Gemini ASIC that creates the interconnect for the Baker systems presumably doubles up the bandwidth and capacity of XT systems compared to the current SeaStar2+ interconnect currently used in XT5 and XT6 machines.
The Baker systems will use the same two-socket blades that currently go into the XT6 supers and their midrange brethren, the XT6m. The Baker systems also include a specific software stack, designed to scale Linux and its applications across what could be millions of processor cores. Ungaro said there was work still to be done on this software stack, but was at this point fairly confident it would perform as planned.
The Gemini interconnect ASIC is the lynchpin for the whole Baker plan and, therefore, Cray's financial 2010. While the entry CX1 and new midrange CX1000 CPU-GPU hybrid machines are interesting, they are not selling in sufficient volumes to make any difference to Cray in 2010.
The initial tests for Gemini must be going alright because Cray tightened its delivery for Baker systems from "the third quarter" when it spoke with Wall Street thirteen weeks ago to "early third quarter" when speaking Tuesday.
Still, it is going to be a thin couple of quarters, with sales expected to be around $30m in the second quarter and $50m in the third quarter. The company reiterated that given the delivery and acceptance of Baker/Gemini systems in the third and fourth quarters as it currently sees them unfolding, it can still post sales of between $305m to $325m, and be profitable in 2010 as well. That includes $110m in services revenues.
To help cover costs, the contract Cray has with DARPA, which is for the development of future supers and which is booked not as revenue but to reduce R&D expenses, includes two more milestones this year. If Cray passes them in the third and fourth quarter as expected, that will cut costs by around $12m in each quarter.
If you do the math, that means Cray has to have around $200m in revenues in the fourth quarter, and revenues only get booked at Cray once a machine is shipped, installed, tested, and then accepted after testing by the customer to hit the performance they expected. You don't just drop these monster HPC machines in and run like you do a two-socket x64 box.
"We've mapped it out," said Ungaro to a skeptical Wall Street. "We believe it is achievable." But Ungaro also qualified that many times, saying that if there is a bug found in the Gemini ASIC, that could force a tweak and perhaps add a quarter or so to the delivery time for Baker/Gemini systems.
If that happens, 2010 will get hosed for Cray. But, then again, think about how easy the compares will all be in 2011. And by then, the Intel Xeon-based CX1 and CX1000 lines may actually build up some steam and capture some revenues, helping to even out Cray's quarterly revenue and profits.
In the meantime, Cray has $103.3m in cash and will burn some of it to build up parts supplies and get all the testing done on the Baker/Gemini systems. ®
Sponsored: Benefits from the lessons learned in HPC