The shaky-cam shot, below, shows the Nehalem die in its quad-core guise. What we can tell is that the monolithic core, sporting some 731M transistors and eight-thread compute capability, isn't small.
Due to be launched in late 2008, Intel is initially positioning single-socket Nehalem setups as a base for a high-end desktop, and two-socket Nehalem boxes as server/workstation material.
During a roundtable discussion on Nehalem at Spring IDF 2008, Intel's representatives were being rather coy about just how other SKUs would be fashioned. For example, would the proposed eight-core model simply be two quad-core chips glued together, as is the case with four-core Core 2? Just how will Intel architect the two-core model, dubbed Gainestown, when the monolithic die supports four cores from the get-go? Would it switch off certain cores, or, more likely, re-architect Nehalem?
How would the SKU with the integrated graphics shape up, from a silicon viewpoint? Would it require a separate chip for the graphics' plumbing, or would it be more like AMD's APU? What's Intel's plan for cache sizes on varying cores?
We'll continue to badger Intel to provide the answers to these questions, but if you, the reader, takes away just one thing from this piece, it's that Nehalem, on the desktop, won't be a cheap product in late '08 - a look at the architecture tells us that. Don't expect to see £150 SKUs, and once you factor in the cost of a new supporting motherboard, for the 1,366-pin chip, upgrading to the quad-core Nehalem variant won't see much change from £500.
Yes, it will be faster than Penryn on a clock-for-clock basis, but the volume space will clearly belong to the present line of Penryn CPUs until way into 2009. Indeed, given such a statement, AMD's quad-core desktop update to its Phenom processor, Deneb, may not compete against Nehalem for a while, taking pricing into account.
