Reports on Intel's Westmere 32nm plans contain one slightly disappointing piece of news. The six-core 12 MB cache Gulftown - the 32 nm successor to the current Core i7 high-end LGA1366 Nehalems - is now firmly a 1H 2010 entry, rather than the intro part for the 32 nm process. And that's from the mouth of our friend Stephen Smith, our favourite Veep at the Digital Enterprise Group. By extension, that would apply to its dual-socket equivalent that follows the soon-to-be-out Gainestown Nehalem EP. Another point, just confirmed today, is that the 45nm 8-core biggie, the 'Beckton' Nehalem EX - critical to Intel's recapture of the 4-8 socket enterprise and HPC performance lead - is seemingly also a 1Q 2010 part. We really hoped that this part, expected to roll over its competition (both X86 and otherwise) would be out this calendar year, but looks like it's not to be. Will the delay have any impact? If AMD doesn't suffer a premature death due to a possible Arab funding failure, it should have more than just a DDR3, HT3-enabled smoother-scaling flavour of Shanghai. Those six-corers like Istanbul, and even a 12-core MCM version - provided the socket is big enough - are possible, too.
If any of these come out at a time close to the first quarter of 2010, it will pose some competition to Beckton. How much? Depends on what AMD can deliver performance-wise. Whatever the monster multi processor update is from AMD then, it won't (or shouldn't) run at less than 2.8GHz clock-wise. Meaning that Intel may have to go above 2.66 GHz for the Beckton launch if it does happen in early 2010. As for the UP and DP, with an updated stepping past the most recent D0, Intel should be able to ratchet up the Gainestown on DP and Bloomfield on UP to 3.6 GHz by year end. It should keep it well ahead even of DDR3-enabled Shanghai, at least unless AMD decides to release the six-corer in UP and DP package before the Gulftown cometh. In summary, the 2010 (re)schedule of Gulftown and Beckton doesn't impact the Intel performance position in any meaningful way, but it does give AMD an opening to grab a head start, at least in the MP market, where its Shanghai MP Opterons do keep ahead of FSB'd Dunningtons in HPC stuff as of now. The DP and UP segments remain led by Intel, unless AMD pulls a naughty speed surprise out of the bag in the meantime. ( www.atomicmpc.com.au )
Post a Comment