Sun Nixes Unified Sun-Fujitsu-Siemens Sparc Hardware Platform

by Timothy Prickett Morgan

There's been plenty of talk about Sun Microsystems and the possible courses of action that it might take to chart itself a new path in the aftermath of the dot-com downdraft. There has been talk of Fujitsu, a long-time partner of Sun's, or the Fujitsu-Siemens partnership acquiring Sun. For the most part, the reason why Sun has been talking so openly about its roadmap for the future, which it calls Throughput Computing, is to quell any concerns that its customers might have over Sun's prospects in the ever-aggressive server market. The top brass at both Fujitsu and Sun have recently quashed the idea of Fujitsu or Fujitsu-Siemens buying Sun, but there is another possible option that these two companies have: They could create a single Sparc server platform from their respective technologies and then try to get OEMs to sell the boxes designed and manufactured by the two of them.

Neither Sun nor Fujitsu Siemens seems inclined to go in this direction, as is amply demonstrated by the fact that they both just rolled out roadmaps that take them into 2006 and beyond on widely divergent Sparc-compatible and Solaris-compatible server platforms. Such a move would require both vendors to make some tough choices. But it might end up making their respective server businesses more profitable, too. So it is something to think about.

Fujitsu's Sparc64 chip designers are working on rolling out a whole line of improved processors and servers to keep the performance of its Solaris servers going up and up. The Sparc64 V processors are just coming out now at 1.35 GHz using a 130 nanometer copper and SOI process, and by late this year or early next year will hit 1.62 GHz, pushing up eventually to 2.4 GHz. The Sparc64 VI will be a dual-core implementation of the Sparc64 with two on-chip 4 MB L2 caches. These chips could debut in mid-2004, and are slated for a speed bump to at least 3 GHz by 2005. It could probably ramp the clock on the Sparc64 VI dual-core chips to 4 GHz if the Sparc64 VII chips were running late. These chips will have four cores per chip and will run at between 5 GHz and 6 GHz. The resulting chips plugged into a modified 128-way "Kaiser" PrimePower server should give Fujitsu and Siemens a RISC/Unix box that has about 30 times the processing power of the PrimePower machines the two debuted in 2001. Fujitsu is only planning on putting one thread per core in all of its future Sparc64 chips, and this is a big difference that it has with Sun's future plans.

Sun is designing future Sparc chips and the servers that surround them based on the idea that instruction-level parallelism--something that Intel Corp has focused on with its 64-bit Itanium processors and also that IBM, Hewlett-Packard, Fujitsu, and others have chased with their RISC chip designs--is not as important for a lot of the workloads out there in the commercial world as is thread-level parallelism. Sun's UltraSparc-IV is due in early 2004, alongside IBM's second-generation dual core Power5 chip and HP's first dual-core chip, the PA-8800. The UltraSparc-V will come out in early 2006, with the processor appearing in systems in the second half of 2006. The UltraSparc-V processor is based on a 90 nanometer process from Texas Instruments, and will be able to switch from one mode optimized for single thread performance to another mode optimized for multiple thread performance, much as some chips can switch from 32-bit to 64-bit modes today. The UltraSparc-V will have two threads, and the Sun Fire servers using them will have approximately five times the performance of the current Sun Fire servers using the 1.2 GHz UltraSparc-IIIs. For lower end machines, Sun has the "Gemini" H-Series processors for its blade machines, which have SMT support, due in 2004 using a 130 nanometer process, and the 'Niagara" H-Series line of processors, which have eight cores and 32 threads as well as integrated memory, network, and security controllers on a single chip, due in late 2005 or early 2006. Clearly Sun is counting on an inflection point in the market in late 2005. Until then, expect Sun to talk a lot about threads.

Sun is, according to Andy Ingram, vice president of marketing for processors and networking products at the company, pretty happy with the existing relationship it has with Fujitsu and Siemens. He says that Fujitsu Siemens pays a royalty on each license of Solaris that it sells, and that Sun also gets money from helping Fujitsu engineer and test its Sparc clones to run each release of Solaris. Exactly how much is unclear. To his way of thinking, Sun is already spreading the cost of Solaris development over both Sun and Fujitsu-Siemens, and there is little to be gained in this area. "And on the Sparc half of the equation, there are some issues," concedes Ingram. "Differentiation is a problem." While Fujitsu Siemens licenses the guts of the Sparc chip from Sparc International, an independent consortium Sun established many years ago to try to foster clones of the Sparc and also licenses some memory interconnect technology from Sun, according to Ingram, the fact is that both Sun and Fujitsu Siemens see their respective hardware platforms as key differentiators. (Incidentally, Fujitsu is still a big reseller of Sun's servers in Japan, as it has been for nearly a decade.)

When asked directly about whether Sun and Fujitsu-Siemens might nonetheless give the idea some consideration, Ingram was not terribly hopeful that there might be a single Sparc platform some time in the future. "I don't know the answer to that," he said. "There are a lot of possibilities, but it is hard to reckon the probabilities of what these two companies might agree to."

Perhaps most significantly, Fujitsu-Siemens seems content with serving its key customers in Germany and Japan and getting some sales in North America and the rest of Europe. Fujitsu-Siemens needs Sun to make Solaris-compatible machines, so it doesn't want to rock the boat in the Solaris market any more than Fujitsu and its subsidiary Amdahl rocked the boat in the market for mainframes that were clones of IBM Corp machines. "It's really a question of whether or not Fujitsu-Siemens is expanding the Solaris market, or is it just playing the same game it did in the plug-compatible mainframe market," says Ingram. "They make a nice box, and we obviously like the operating environment. But they seem to be doing the same thing in Sparc as they did in mainframes."

If the economy tightens and Sun and Fujitsu-Siemens need to tighten their belts, they could unite their respective Sparc platforms and create a single line. But it will take a much worse economy than the world currently has to force them to such a decision. There are limits to pride, engineering, and profit, and inertia has its own way of preventing such possibilities from being considered.