Information Week reports that during a conference at their headquarters, Intel spoke to analysts and reporters about their future energy-efficient products, including Larrabee and wireless solutions. Justin Rattner, Intel’s CTO, also noted that Larrabee will be their “first tera-scale processor” and that it is aimed at a 2010 release, or possibly 2009 if things go especially smoothly.
Justin also reiterated that Larrabee will deliver “well in excess” of one teraflop of processing power. The company stressed the importance of their “path finding” process: previously, product development engineers were pretty much on their own guessing what design decisions made the most sense for the target timeframe.
In the company’s newest initiatives, such as Tera-Scale, researchers and product development people are working much closer together. The researchers are also responsible for finding “the best use for newly developed technology in products”.
The fact that Larrabee (or presumably, its first silicon implementation, as the codename often seems to refer to the architecture and all its potential derivatives) seems now more likely to be released in 2010 (or very late 2009) rather than early 2009 brings up a number of interesting questions, such as what process it will be manufactured on and what NVIDIA and AMD’s products will bring to the table in that timeframe.
In terms of process technology, the 32nm shrink of the Nehalem architecture (aka Westmere) is aimed at a 2H 2009 release, and most likely Q4. If Larrabee comes out after that, it would make sense to presume that it will also be manufactured on 32nm. Furthermore, if it wasn’t, Intel would actually be at a process disadvantage in that timeframe: NVIDIA and ATI are likely to be on the 40nm half-node at TSMC, while Intel would only be on 45nm. As such, this question is of keen importance, as it will determine much of the future competitivity of the initial implementation of the Larrabee architecture.
However, given that Larrabee has been compared with Gesher (aka Sandy Bridge, which is a 32nm chip) in a recent Intel presentation, it is far from impossible to imagine it will be produced on the same process. Finally, in terms of competing products, this most likely implies that Larrabee will directly compete with NVIDIA and ATI’s true next-gen architectures, rather than with loose derivatives of the G80 and R600. No matter what these new architectures will be called (G100 and R800?), they will certainly be very different from today’s, and that makes it much more difficult to predict how Larrabee will compare to the competition.
Intel, Processors, CPUs, Larrabee