Sun to Buy Supercomputer-Maker Cray?

by Timothy Prickett Morgan

Like a lot of businesses, the IT industry is full of rumors. Some of them pan out, and some do not. For instance, time and again over the past few years, I heard rumors that Fujitsu would either partner with Sun Microsystems on a unified Sparc Unix server platform or acquire Sun outright. Eventually, the partnership, which was announced earlier this month, happened. Now Sun is rumored to be thinking about acquiring supercomputer-maker Cray.

This rumor could turn out to be complete hogwash, of course. But it does present some interesting possibilities.

First of all, buying Cray would be, in some measure, a return to Sun's roots in the IT business. Sun was a Unix workstation vendor for academic and government research institutions long before it was a server maker, or the main provider of systems for telecommunications companies, and the dominant platform for early commercial Internet computing (in the mid-1990s). The workstation business created by Scott McNealy, Bill Joy, Andy Bechtolsheim, and Vinod Khosla planted the seed for a high-performance-computing market distinct from outside of the vector supercomputers created by Seymour Cray for his eponymous company and line of computers.

But Sun's involvement in the HPC market goes way beyond the workstation. Sun's CTO, Greg Papadopoulos, was one of the chief designers (the other being Danny Hillis) of the CM5 massively parallel supercomputer from Thinking Machines. (Sun bought the carcass of Thinking Machines, as it also bought the carcass of Kendall Square, another parallel supercomputer maker.) Mark Tolliver, Sun's former chief strategy officer, worked at rival Hewlett-Packard and held the top marketing job at Maspar, a defunct parallel computer maker that was eaten by Digital Equipment. Mark Canepa, who is in charge of Sun's storage unit, and cut his teeth at HP, is running the Apollo workstation unit that HP acquired in order to go after Sun's fledgling workstation business and eventually driving the HP acquisition of vector supercomputer maker Convex. (The Convex machines are at the heart of the HP Superdome servers.) When SGI decided to buy Cray, in 1996, there were two competing lines. So Sun acquired the Sparc-based Cray 6400 supers from the Cray Research division of Cray. With that acquisition, it also got Clark Masters, the lead designer of that machine and the guy who has headed up Sun's enterprise server unit for years, and Shahin Khan, Sun's chief competitive officer and, most recently, the general manager of its cross-divisional HPC business unit. Khan did product marketing at Cray Research and Floating Point Systems, a company Cray acquired that had Masters as its main system designer.

The acquisition of the Cray 6400 was pivotal for Sun, giving it an influx of techies and marketing people who understood both the Sparc chip and the big systems. Sun rebranded the Cray 6400 as the "Starfire" Enterprise 10000 and sold thousands of these very expensive yet very powerful machines in the dot-com boom.

Both Masters and Khan kept their jobs in the management shakeup in April that saw a lot of Sun executives either leave or be moved from prominent positions. Last week, Masters was tapped by Sun's head salesman, Robert Youngjohns, to take a point position selling Sun's hardware, software, and services into government organizations. Khan is reportedly leaving Sun. This could be to make room for Cray, or for some other reason.

Sun and Cray are a good fit in many ways. First, like Sun, Cray has caught on that it can differentiate in the market by adopting the Opteron processors from Advanced Micro Devices. In fact, Cray understood this long before Sun did, winning a $90 million contract to build the "Red Storm" parallel Linux supercomputer for Sandia National Labs. In March, Cray got even more serious about the Opteron when it paid $115 million to acquire an upstart supercomputer maker called OctigaBay, which has invented a very sophisticated Opteron supercomputer that runs Linux and can scale to over 12,000 Opteron processors to deliver 58 teraflops of number-crunching power, using the 2.4 GHz Opteron 246 processors. OctigaBay was a serious threat to Cray's fledgling Opteron supercomputer business, and buying it was a very shrewd move. The OctigaBay 12K systems are supposed to be in beta now, with volume shipments toward the end of this year. (You can find out more about the innovative architecture of the OctigaBay 12K in "OctigaBay Takes Opteron-Linux to New HPC Heights".)

Cray would also be an attractive target for a Sun acquisition for other reasons. For one, it would complete the loop, with Sun getting the other half of Cray that it didn't buy from SGI in 1996, and which SGI sold to a relatively obscure and money-losing supercomputer maker called Tera Computer in March 2000. Tera quickly adopted the Cray name for its company, then started lining up deals for the Cray gear. And now, four years later, Cray has a relatively steady vector supercomputer business, thanks to the launch of the Cray X1 supers and a boost in demand for traditional vector machines by governments around the world for various applications, including security in the wake of September 11. It also has respect, thanks to the X1, Red Storm, and OctigaBay machines, and has a sizable contract with the U.S. Defense Advanced Research Projects Agency (you know, the organization that really did create the Internet), to help the government work toward delivering petaflops (that's 1,000 teraflops) supercomputers by 2009 or 2010.

The DARPA work is significant for two reasons. One, it is cool. And, two, Sun and IBM were the only other two vendors who were given contracts for the project. Cray is designing a concept supercomputer called "Cascade" as part of a $43.1 million DARPA contract. The details are thin, but Cray has said that Cascade will include new processor designs that make better use of memory hierarchies because the L1-L2-L3 cache memory scheme we currently use is not efficient. Cray is working on processor-in-memory technologies that it hopes will deliver bigger sustainable memory bandwidths and lower latencies, which is the name of the game when building a giant cluster of computers. The Cascade machine will also try to include shared memory and distributed memory programming models while trying to allow existing applications to port to the new Cascade architecture. Sun's DARPA contract for $49.7 million was for a project called "Hero," which is more a software than a hardware project. Hero aims to take a machine with hundreds or thousands of processors and create development tools that will mask the underlying parallelism of the machine and make it look, at least to applications, like a single big processor. This is what SMP electronics and server operating systems do today for commercial servers. Sun is also developing the benchmarks that gauge the programmer productivity and performance of the DARPA prototypes.

Another reason why Sun can think about buying Cray is that it can afford to do it. At press time, this rumor had not leaked to the general IT trade press and Cray had a market capitalization of about $550 million. Sun just got $2 billion in its settlement with Microsoft. If Sun paid a slight premium for Cray, it might pay as much as $650 million for it, which is reasonable for a company that took in $237 million in sales and $21 million in profits in 2003. Sun would have to believe that its sales force, particular that lead by Masters in the government sector, could significantly boost Cray's sales to do the deal. If Sun bought Cray and could double its sales of supercomputers, most of that money could fall to the bottom line. Cray is experiencing some bumpy transitions in the first and second quarters, as it ramps up Red Storm, OctigaBay, and X1E supers, and profits have been hurt a bit. But, generally speaking, Cray is 45 times smaller than Sun and has thrown off more profits in the past 12 months. If Sun can afford it, and it believes in the HPC market, a Cray acquisition might make sense.

Cray was unavailable for comment at press time, and a spokesperson for Sun said that the company cannot comment on speculations or rumor.