Sun's Rock doomed from start, analysts say

Sun's development Rock processor was a troubled project that may never have stood a chance, analysts said following reports that the chip had been axed.

Earlier this week, the New York Times reported that Sun had cut development of the 16-core Rock chips, which were designed to go into high-end servers. The multithreaded processor would have doubled the core count from Sun's fastest server processor today, the eight-core UltraSparc T2, and was targeted at enterprise servers to process data-intensive applications like databases.

Rock's development had been a high-priority project within Sun, and the company poured plenty of money into its development. Sun had high expectations for the chip, which blended high multithreading capabilities with quick instruction processing. Rock was originally due for release in 2008, but Sun said last year that it had delayed the release until the second half of 2009.

Ultimately, the continuous delays made Rock's development a costly affair for Sun. Rock also suffered as incrementally better chips from rivals such as IBM, Intel and Advanced Micro Devices hit the market.

Rock may have gotten buried under the weight of its costs and high expectations, said Gordon Haff, principal IT adviser at Illuminata. Sun has refused to comment on Rock's future, but Haff said that the report of its demise "certainly is a very believable rumor."

Rock's development took a hit as budgets got lower and the chip saw many glitches during the development process, said a financial analyst who spoke on condition of anonymity because he had not been authorized to speak with press. The budget for the development shrunk as the company lost server market share to its competitors like IBM and Intel, he said.

Key Sun employees involved in its development also left the company over the past few years, which affected Rock's development, the financial analyst said. The departures included former Executive Vice President of the Microelectronics David Yen, who left for Juniper Networks in 2008, and respected chip designer Marc Tremblay, previously a chief technology officer in Sun's microelectronics business, who bolted for Microsoft this year to work as a distinguished engineer.

Even if Rock had made it to market, it would have been an uninteresting processor as companies like Intel and Advanced Micro Devices are offering high-performance chips at more reasonable prices, said Dean McCarron, principal analyst at Mercury Research.

Intel and AMD make x86 server chips that are used in industry-standard servers running the Windows or Linux OS. As more companies adopt those chips, Sun's return on investment on the Rock chips -- written for the Solaris OS -- would have declined, McCarron said. Putting more money into Rock's research would only make Sun's pockets lighter, he said.

Sun is reportedly dropping the chip as it prepares to be acquired by database giant Oracle. Though there's little evidence of Oracle's involvement in Rock's cancellation, it may have played an informal role in the background, said Dan Olds, principal analyst at Gabriel Consulting Group.

Oracle in April announced its intent to buy Sun for US$7.4 billion, and Oracle CEO Larry Ellison said at the time it was most interested in Sun's Solaris OS and Java software. Ellison initially provided few details about what Oracle would do with the Sparc chips, but later clarified that it intended to stay in the hardware business and would increase its investment in Sparc.

Sun's competitors, including Intel and IBM, jumped on the uncertainty surrounding Sparc's future. On the day Oracle announced its intent to acquire Sun, IBM highlighted the success of its Power chip, stating customers had migrated to IBM servers from Sun systems. A few days later, Intel CEO Paul Otellini said that Sun's departure would create a larger market for its Itanium processor. Analysts speculated that chip development would be the first to fall as Oracle looks to clean up the company after its acquisition.

Oracle could in fact keep Sun's hardware business at some level, Olds said. Oracle may end up retaining some parts of the Sparc business, perhaps continuing development of its low-end Sparc chips, code-named Niagara, and high-end Sparc64 chips. Rock was originally designed to merge Niagara and Sparc64 designs in a single chip, but the development of both chips could continue separately.

Sun is still selling a good number of Niagara-based servers, Olds said. Sun is in a contract with Fujitsu to design and manufacture Sparc64 chips.