Just two years ago, CEO Jonathan Schwartz blogged about getting the first Rock silicon and how it would support 256 terabytes of RAM. It was supposed to have four cores, each with four processing engines and was originally to debut in 2008 and replace UltraSparc V. And now, it may be gone:
Ultimately, I suspect that this anonymous poster at RealWorldTech has it right when he suggests that Rock's run-ahead, speculative execution approach was never able to deliver the kinds of speedups that its designers were hoping for across enough workload types. Or maybe there was a major bottleneck somewhere else, like socket bandwidth, that prevented the chip from getting anywhere near its theoretical peak performance. Or, alternatively, anonymous and I could both be wrong, and Sun's Rock engineers could be throwing away some revolutionary hardware simply because their company ran out of time and money and got bought by a software giant.Given that the chip had been sampled at least since 2007 and was supposed to release some time this year, later thant he original 2008, I have to wonder if the issue really was unfulfilled expectations. You'd think that might have been clearly before now. What may be happening is something that Eric Lai at Computerworld covered: the issue of a projects being killed after a corporate acquisition.
When one company acquired another, it often takes on some hefty debt, or at least wants to recoup the investment as quickly as possible to prove to investors that the move was wise. The acquirers become the "experts" who know what is best for the business of the acquired. There's been speculation that Oracle was really only interested in Sun's software and would sell the hardware business off to Fujitsu to avoid competing with long-standing business partners. We'll never know, but it sounds as though the Rock was just another victim, albeit a once high-profile one, to the economics of M&A.
Rock image via stock.xchng user willsbma, site standard license.