Sun Breaks Sparc Unit Free Again
Published: March 29, 2007
by Timothy Prickett Morgan
Server maker Sun Microsystems this week announced that it was splitting its Sparc processor development group from its Systems group, and putting David Yen, a key executive who once ran Sun's chip business, back in charge after he had a 10-month stint running the merged Sun and StorageTek storage businesses. The move coincided with the naming of Jon Benson, formerly the vice president of tape engineering, as senior vice president of Sun's storage business.
Yen is not stranger to moves at Sun, and has been used in recent years as a point man to cope with problems in Sun's business units; he has had his own share of woes as the person who headed up the development of various Sparc processor lines as well. The UltraSparc-III processors, for instance, were several years late and when they finally did make their debut in September 2001, they were underpowered relative to alternatives and the systems using them were costly. It was Yen who killed off the "Millennium" UltraSparc-V and "Gemini" UltraSparcIIIi processors in April 2004, when he was also named to head up the Scalable Systems unit, which merged Sparc chip development with the Sparc-based Sun Fire server line to get some tighter coupling between chip and server development. Up until then, Sun's Sparc chip business operated as a separate unit, developing processors that other Sun units planned to use, but sometimes did not use very effectively. (Remember the MAJC Java processor? Whatever happened to that?)
April 2004 was, of course, when Jonathan Schwartz, Sun's current chief executive officer, was named as chief operating officer, setting the stage for his rise. Getting the Sparc business back on track was key back then, since the bulk of Sun's server revenues and an even larger portion of its profits came from these machines. After years of customers unplugging Sparc boxes for other RISC/Unix or Linux on X86 systems, Sun knew that it needed to get better UltraSparc processors into the field. The dual-core "Cheetah" UltraSparc-IV chips, which Yen steered to the market, were a step in the right direction, and with the "Panther" UltraSparc-IV+ kickers announced in September 2005, Sun finally had a chip in the field that it did not have to be embarrassed by when it came to performance or bang for the buck. And, consequently, customers have begun investing in Sparc boxes again.
The delay in the jointly developed "Jupiter" servers that Sun is hoping to bring to market soon with partner Fujitsu is a bit of an embarrassment, but it is Fujitsu's dual-core Sparc64 VI "Jupiter" chip that is delayed. Sun and Fujitsu announced their partnership to develop a kicker to the Panther systems in June 2004, and said they would use Fujitsu's Sparc clones. The so-called Advanced Product Line machines were expected around mid-2006 originally, and now are expected sometime in the first half of 2007. That means within the next three months.
Yen also had a hand in the move away from traditional Sparc processor designs and the embracing of multithreaded designs, such as the current "Niagara" Sparc T1 chips and the future "Rock" chips. Sun has now sold some $500 million or so worth of Sparc T1 servers, and perhaps more importantly, it is seen as being an innovator in the energy-conscious data center of today rather than the performance-crazed data center of the past. Niagara is still a drop in the worldwide server bucket--less than one percent of annual server sales--but it gives Sun a positive message to sell and a product that is presumably profitable and growing faster than the market at large. It sure beats the alternative, which is the kicks in the teeth that Sun took from late 2000 through about the middle of last year, when it was clear that the Panther and Niagara chips were pretty good products.
Yen was unavailable for comment about his appointment as vice president and general manager of the reconstituted Microelectronics group at Sun. And Schwartz did all of the talking in the statement that Sun released about the executive changes, proclaiming that Sun would be expanding its focus on chip designs. (Which begs the question, can you really expand a focus?). The Microelectronics group will be a hodge-podge of projects and products that fit Yen's expertise, including Sparc processors, network co-processors like Sun's "Neptune" cryptographic chip, as well as high-performance computing.
"The broad acceptance of the open source Solaris operating system running on Dell, HP and IBM hardware shows that Sun's innovations have value and appeal beyond our own servers and storage products," Schwartz explained. "With numerous successes including the success of our UltraSparc T1 processors fueling the growth of our chip multi-threaded servers, the tapeout of our Rock processors defining new terrain in high-productivity computing, and innovations like Project Neptune opening entirely new markets for our technology. Now is the time to fuel that same success with our Microelectronics products. As with our software, decoupling our silicon from a strict reliance on Sun's systems raises our profile and opportunity globally."
As you might expect, that was obviously Sun putting the best spin on Yen's move and the naming of a new head of the storage unit. But Yen really went into the storage business to rationalize its product line and kill off a bunch of products, much as he did with the Sparc business years ago. And if he is being put in charge of chips again, it is probably not just because Sun sees this wonderful opportunity, but because something is wrong.
As we previously reported, there is a rumor going around that Texas Instruments, who just so happens to be Sun's chip fabrication partner for its Sparc processors and has been since day one (with Fujistu having a short gig to make UltraSparc-II chips back in the dot-com days when Sun could not keep up with demand because demand was huge, not because it underestimated demand), has fired some 500 chip experts who develop advanced chip fabrication technologies. (This information comes from Carl Johnson, the main analyst at semiconductor watcher InfraStructure.)
This TI rumor hit about the same time that Sun and Intel got buddy-buddy, with Sun promising to use Intel Xeon chips in its "Galaxy" X64 server line and Intel promising to support Solaris on those chips. Maybe it is a coincidence. Maybe the TI layoffs are not what they seem. But something else might be up, especially if Sun is shopping around for a new fab partner.
Intel clearly has the manufacturing volume and technologies to implement Sun's Niagara and Rock chips, but the design of a chip and the very specific manufacturing processes are tightly coupled. Sun can't just send some chip masks to Intel and say "give me 2 million of these next week." Sun and Fujitsu could partner on chip design and manufacture, but the delays in the APL product line--whatever they are, since neither vendor talks about it--probably are not making Fujitsu look as appealing as a chip supplier as it might otherwise. Sun's Niagara-2 and Rock-1 chips use TI's already-designed 65 nanometer processes, but future Niagara-3 and Rock-2 chips will use 45 nanometer processes. I happen to think that by the 45 nanometer era, there will be tremendous economic pressure for chip makers to partner to share costs, and that we could end up with only be two fabricators of processors: Intel on one side and everyone else on the other. It is hard to say where Sun will end up. I have a sneaking suspicion that what Yen is really being asked to do is sort that out--and to do so in such a way that future Rock and Niagara chips are not delayed, and perhaps even accelerated to market.
Will 45 Nanometer Chips Make Two Warring Camps?
Sun Merges Server Units, Taps Key Exec for Storage
Sun Lays Off Workers in Sparc Server Group
UltraSparc-IV+ Chips Give Sun's Midrange Servers Twice the Oomph
John Fowler Appointed to Head Sun's Network Systems Unit
Sun Reports $760 Million Loss, Shakes Up Executive Leadership
Sun Pulls the Plug on Future Millennium, Gemini Chips
Post this story to del.icio.us
Post this story to Digg
Post this story to Slashdot