June 4, 2003 10:06 AM PDT
Sun's processor plans slip a notch
David Yen, executive vice president of Sun's microprocessor group, said at a meeting with reporters Tuesday that its UltraSparc IV processor will emerge in the first quarter of 2004, rather than late 2003, as Sun had said earlier. Furthermore, its planned successor, UltraSparc V, could appear in early 2006, instead of late 2005, he added.
A one-quarter slip isn't a major problem, especially in the server market where product cycles don't churn as quickly as with PCs, said Kevin Krewell, analyst and senior editor of the Microprocessor Report newsletter.
"Beyond a quarter, you start having an impact," said Krewell, who had expected UltraSparc IV to debut in the fourth quarter of 2003. Sun CEO Scott McNealy said in October 2002 that UltraSparc IV would appear in products in "about a year." Regarding UltraSparc V, Krewell said, "They definitely seem to be hedging a bit. It could be that the reality has set in."
UltraSparc IV, built on a process with 130-nanometer features, will fit into the same sockets as the current UltraSparc III but will combine two UltraSparc III processors on a single slice of silicon. The UltraSparc V, built on a more advanced 90-nanometer process, will be a major new design that can execute two batches of instructions at a time and change personality according to the type of work assigned to it.
Both chips will be sold in parallel with new products that employ "chip multithreading" technology. CMT uses numerous small, simple processors for executing many operations simultaneously instead of large chips a single operation very fast. CMT designs, acquired through Sun's purchase of start-up Afara Websystems, will initially be used in lower-end servers that handle Java and Web services transactions. After UltraSparc V, Sun will build the CMT technology into the high-end family as well.
Sun's current UltraSparc III was a late arrival, and the company must regain lost ground, Krewell said. "They know they screwed up on UltraSparc III, and they've got to play catch-up," he said.
The Menlo Park, Calif. company, whose primary business is selling servers that handle network storage and processing tasks, has tried to chart its own course rather than rely on processors from rival Intel. Though it has begun selling some Intel-based products, the company is steering customers toward use of its own version of the Unix operating system, Solaris.
In comparison, Intel processors are critical to server plans from Sun's competitors Dell Computer and Hewlett-Packard. In addition, rival IBM employs a dual-pronged strategy that emphasizes its own Power processors while making a place for Intel's Xeon and top-end Itanium chips.
Sun's first CMT design, code-named Niagara, will debut in 2005. It will have eight simple processor cores, each able to quickly switch among four different instruction chains, or "threads." The design is geared to keep the chip busy by switching to a new thread whenever the old one "stalls" when awaiting a response from the computer's much slower main memory.
CMT, which Sun also calls "throughput computing," is a radical and interesting approach, Krewell said. The design appears to be well suited to midrange tasks such as running Java programs and the Web services work that will underlie many next-generation Internet operations. However, he said he disagrees with Sun that CMT designs will be good "as a general-purpose high-performance processor."
Another Sun chip in the works, code-named Gemini, is for low-end servers and is due to debut in 2004. This model puts two UltraSparc II processor cores onto a single piece of silicon, Sun's Yen said, and the company has prototypes "already working in the lab." It's a likely candidate for use in Sun's blade servers, which the company plans to update with dual-core UltraSparc processors.
In the nearer term, Sun's UltraSparc IIIi, code-named Jalapeno and running at 1.06GHz, will speed up soon. A 1.28GHz model "is ready to go out," Yen said.
Sun designs its own processors but relies on Texas Instruments to build them. It's a relationship that Krewell says is imperfect because TI's main chip fabrication business is for a very different type of product than Sun's server processors, and because Sun can suffer if TI has problems getting a new manufacturing process up and running.