Dueling Multicores: Intel and AMD Fight For the Future. Part 1 - UPDATE!

Intel Still Chasing Moore's Law? Continued

Why this focus? The reason is a simple one: it's no longer possible for new single-core processors, even with continually shrinking lithographic processes, to be clocked into the 3.5 - 4.0 GHz range without toasting themselves. So the only way possible for AMD and Intel to achieve continued performance gains at the rate we've currently enjoyed, is through parallelism. In short, new CPUs must divide their workloads among two or more processor engines. If a single new industry standard architecture is to be defined, its guiding principles and its market success may depend almost entirely on how these future workloads are divided, and how parallelism is achieved.

Nathan Brookwood from Insight 64

As Tom's Hardware Guide reported from the Spring IDF , Intel is introducing four new dual-processing engine design platforms almost simultaneously this year. Though ostensibly geared for separate markets, Intel may be relying on its customer base to determine which platform gathers the most momentum. We enlisted Insight 64 analyst Nathan Brookwood to help us sort through the distinctions between Intel's new dual-core implementations:

For the new Pentium D series ("D" for "dual-core," in place of the "4"), Intel is introducing the long-awaited Smithfield, which will be its first dual-core platform to support both hyperthreading and EM64T. Its first incarnation in 2005 will be Lyndon, which will be produced using 90 nm lithography, and will introduce an alphabet soup of new Intel technologies, including its self-correcting Active Management Technology (IAMT). 2006 promises an upgraded version called Averill, which according to Intel's Pat Gelsinger is "where we bring dual-core into the mainstream of corporate computing." Smithfield also marks the beginning of a customer incentive called the Stable Image Platform Program (SIPP), which Gelsinger says will provide "a consistent platform for corporate I.T. so that they don't need to change their software offerings on that platform over time." Smithfield's first notebook derivative Yonah will wait for power envelope improvements before its release, scheduled for the first half of 2006. The jewel of the Smithfield family, somewhat down the road, will be Presler, a desktop version manufactured using the next step in miniaturization: 65 nm lithography. For the Itanium 2 series, the Montecito platform includes two processing cores (engines) etched into a single piece of silicon using the new 90 nm lithographic process, with all the wiring between the two cores also etched into the chip. Here, Core 0 and Core 1 do not actually share any resources on-chip; they communicate with one another through the conventional front-side bus. The first derivative architectures from Montecito available during 2005 should be the two-socket, low-voltage Millington, and the two- and four-socket standard voltage Montvale series. For Xeon server platform processors, Intel is planning the Dempsey series, which will initially feature two processor units on separate chips, packaged together. Though not technically dual-core, Dempsey's CPU 0 and CPU 1 will communicate with one another in the same way as they do for Montecito. Dempsey will help Intel premiere its I/O Acceleration Technology (I/OAT), which promises to boost network performance through hardware-driven TCP/IP acceleration. Dual-core derivatives of Dempsey, called Bensley for the server market, and Glidewell for workstations, should be available in the first half of 2006. These two will likely retain Dempsey's server-centric technologies, including I/OAT and IAMT. For Xeon MT 4 socket servers, Intel is introducing the Paxville series in 2005, which is the most technically ambitious of the three dual-core platforms. Here, not only do both cores share the same die, but they share a memory buffer through which they can move data, bypassing the front-side bus entirely. The buffer provides the core's link to the FSB, rather than both cores linked to the FSB individually, as with Montecito and Dempsey. A four-socket (MT) version, called Truland, may be the first Paxville derivative to support EM64T, while Tulsa will be Truland's successor when Intel makes the shift to 65 nm perhaps as soon as next year.