Sun to Debut CMP Technology in UltraSPARC IV

New technology puts two chip cores on a single processor, enhances threading capabilities.

• By Stephen Swoyer
• 10/17/2002

In September, Sun introduced a version of the UltraSPARC III microprocessor, based on a 130 nanometer process, which runs at 1.2 GHz. In the past, the company indicated that the 1.2 GHz chips would be the last iteration of UltraSPARC III before the launch of UltraSPARC IV.

Andy Ingram, vice president of marketing for Sun's PNP group, declined to specify when his company will actually deliver UltraSPARC IV, but pointed out that at its Sun World show several months ago, the Unix kingpin demonstrated a "real-live UltraSPARC IV chip."

As part of its CMP strategy, Ingram confirms, UltraSPARC IV will integrate two chip cores per processor, similar to what IBM Corp. has done with its Power chips. Hewlett-Packard Co. (HP) also plans to introduce similar support in its PA-8800 RISC processors. But Ingram says that UltraSPARC IV also will integrate advanced multi-threading support, similar to a technology—called "hyper-threading"—that Intel Corp. has introduced in its Pentium 4 chips. Ingram claims that UltraSPARC IV's new threading technology will permit a "processor [to] work on multiple instruction streams, multiple processes, at a single point in time."

"You can combine the concept of more [processor] cores, which you see with IBM, and multiple threads—which is what you're hearing, not seeing, from Intel—and that's what we're delivering with CMP," Ingram asserts. "You can have multiple cores per processor, and you can have multiple threads per core."

Industry watcher Nathan Brookwood, a principal with microprocessor consultancy Insight64, says that UltraSPARC IV is a "fairly straightforward" development effort and anticipates that Sun will ship UltraSPARC IV in late 2003. "[UltraSPARC IV] is essentially a repackaging of the current SPARC. It uses compatible busing, except that Sun will fit two of those cores on a single chip, ala what IBM has done with [Power]," he comments.

Ingram claims that UltraSPARC IV's dual-processor core technology will permit Sun's Sunfire 15000 servers, which currently scale to 72 processors with full I/O support, to double in capacity to 144 processors. The SunFire 15000 currently scales to 106 processors with limited I/O support.

Sun Rejects Itanium
While competitors have embraced Intel's next-generation Itanium processor, Sun has thus far refused to endorse Itanium or the IA-64 architecture upon which it's based. HP, for its part, has been an important Itanium booster since day one, and has already announced plans to transition its line of Unix servers away from PA-RISC and on to IA-64, beginning in 2006 or 2007.

For his part, Ingram acknowledges that HP has "good reasons" for moving away from PA-RISC to IA-64, but claims that Sun's UltraSPARC development project doesn't entail the same costs—namely, the support of a chip fab, and the construction of new fabs—that HP incurs by virtue of its in-house development and manufacturing of PA-RISC. For 12 years now, Ingram explains, Sun has outsourced SPARC's manufacturing to Texas Instruments

"We get a lot of advantages that HP and others have of owning their own chip, but we don't have the financial liability of owning the fab," he asserts. "That's a big difference in our strategy relative to HP, IBM, DEC or any one else who wanted to manufacture their own processors"

Notwithstanding the substantial costs associated with the construction of new chip fabs, which can run in the neighborhood of $3 billion or $4 billion each, analysts say that a desire to divest itself of PA-RISC probably wasn't a prime driver behind HP's decision to support Itanium.

"I would figure that the PA-RISC guys would have thrown some body blocks if that was the case," indicates Rob Enderle, a fellow with consultancy Giga Information Group. "Instead, HP went to Itanium because they would be first and foremost on the next generation Intel platform. They saw this relationship with Intel as a way to move them to a position of first among equals."

Even if Sun does outsource its chip-making responsibilities, Insight64's Brookwood points out, it must still support its Solaris development efforts, which probably amount to $300 or $400 million annually.

"Its competitors are spending similar amounts developing PA or Power, but HP is on an arc to get away from this development, and when it happens, they will be saving a lot of money," he argues.

For the record, Ingram reiterates that Sun currently has no plans to support Itanium. He cites IA-64's Explicitly Parallel Instruction Computing (EPIC) instruction set, in particular, as one of the chip's most significant drawbacks.

"I think that [Intel] went a bridge too far with [IA-64's] EPIC architecture … [because] it turns out to be inordinately complex," he contends, noting that the performance of applications on Itanium will hinge upon the effectiveness of the EPIC compilers that Intel and HP have separately introduced.

"The performance of the compiler is fundamental to the specific performance of your application, so you have to do a lot of tuning to make it run well on Itanium, so I think it's going to be a long time coming before software moves there."

HP's, Sun's Futures Hinge on Itanium Success
Rival HP is in a precarious position in view of its PA-RISC transition plans and IA-64. As it now stands, HP has two additional iterations of PA-RISC—PA-RISC 8800 and PA-RISC 8900—that will carry it through 2006. After that, it will ship Itanium-only systems. If demand for Itanium hasn't yet materialized, Giga's Enderle speculates, HP could suffer a disadvantage at the hands of Sun and IBM, whose UltraSPARC and Power development projects will continue apace.

On the other hand, analysts say that any fallout from Sun's decision not to endorse Itanium will probably be years in the making. "This is a long-term phenomenon. Sun's current set of customers … are going to always see another Sun system as being an easier path to upgrade than a migration to a non-Sun environment," Brookwood comments.

Kevin Krewell, a senior analyst for chip industry newsletter Microprocessor Report, believes that Sun has the ability to execute an "about face" if IA-64 "turns out to be such a barnburner that they can't ignore it."

Nevertheless, Krewell argues, Sun has based its reputation on the synergy between UltraSPARC and Solaris. In this respect, he says, "it's going to be very hard for them to change from that, and their customers are going to continue to expect value from it. From that perspective, they can wait a while before they have to make a decision [about IA-64]."