Sun Shelves UltraSPARC V for 'Niagara' and 'Rock'

The sun is setting on UltraSPARC V. Sun Microsystems is updating its processor roadmap in favor of "high throughput" products, and is terminating UltraSPARC V -- code-named Millennium -- and the low-end Gemini processor.

Consequentially, Sun is streamlining its resources and efforts towards the fabrication of "Niagara" and "Rock" – two next generation chips.

Ahmad Zandi, Technical Marketing Group Manager for Sun's Processor and Network Products group, told BetaNews that the change comes as a consequence of a companywide initiative. Sun predicts that higher throughput applications will become dominant within the next several years. As a result, processors will require the capacity to handle "ten's of threads," massive amounts of memory and be complimented by a sufficient reduction in I/O latency.

According to Zandi, Sun's Niagara is designed to handle a baseline of 32 threads in parallel. Niagara will be marketed as a "network facing" processor geared towards Internet service providers who run network-intensive applications.

Solaris 10 will be optimized for Niagara, having the ability to recognize different types of packets. Solaris will then direct the packets to the appropriate thread to be processed.

Sun's Rock, due out after Niagara, is "data facing" and will tackle high performance computing tasks. Rock will jut out from Niagara's baseline performance by handling even more threads and will offer lower heat dissipation.

"Rock is intended to handle the whole gamut of workloads rather than just the network-facing ones that Niagara is designed for," Illuminata's Gordon Haff told BetaNews. Rock is expected to hit the marketplace during the 2005-2006 timeframe.

Sun is preparing a memory subsystem designed to handle "radical" numbers of threads, balancing out systems to ensure that the CPU is not starved for data. But the company has yet to describe how this will be accomplished.

Commenting on Sun's strategy, Gartner's Martin Reynolds told BetaNews, "There sounds like some magic is coming," noting, "There must be a missing component."

Reynolds speculated that this magic might stem from Sun's acquisition of Afara, a licensee that developed multithreaded SPARC chips. Reynolds believes that Afara may have fixed Sun's "windowing problem" – the elegantly complex register windowing system. "The presence of the windowing system may have been a good reason to drop UltraSPARC V," remarked Reynolds.

Sun's Zandi said that the decision to kill UltraSPARC V came because the processor was traditional, "In the class of IBM."

In the meantime, an enhanced version of the UltraSPARC IV processor dubbed UltraSPARC IV-Plus (Panther) will bridge the generational gap until Sun can deliver on its performance promises. UltraSPARC IV "must carry them though," Reynolds told BetaNews.

"If Sun were to double performance of current products they'd have what would looks like a mainstream product," said Reynolds. "UltraSPARC III just not fast enough, and delivers one-half to two-thirds as much performance as the competition, according to recent benchmarks. The only reason to buy UltraSPARC III is for binary compatibility or for niche applications. Customers are moving away from Sun's poor price performance."

Reynolds ranked Sun's current generation of products third behind Intel's Itanium and IBM's Power PC processors, with Intel coming into the winner's circle first.