The news surprised even some AMD employees: The company was expecting triplets in early 2008. The chipmaker had kept the plan so secret -- and the thought of a three-core processor seemed so unlikely -- that last month's announcement attracted significant press attention and industry buzz, but not all of it was good.

Rumors swirled as to the reasons for the maneuver, with industry wags declaring that AMD must have ulterior motives while user forums erupted with conspiracy theories. The truth is actually a bit simpler: AMD is just trying to follow the advice of baseball great Willie Keeler and "Hit 'em where they ain't."

The 75 Percent Solution

From a design point of view, AMD's triple-core Phenom processors will not differ from their quad-core brethren except for having one fewer core. This translates into a 65-nanometer-process CPU with 512K of Level 2 cache per core and 2MB of shared Level 3 cache. The last is an important distinction, as the internal L2 cache is impacted -- falling from quad Phenoms' 2MB to 1.5MB -- but the separate L3 cache is not.

AMD does this by taking a quad-core Phenom and fusing off one of the cores, while leaving the separate L3 cache alone. This option is an advantage inherent to a native multicore design, which allows added flexibility compared to Intel's Core 2 Quad strategy of incorporating two dual-core processors on a single die. A three-quarter-Quad would be difficult for Intel to produce and suffer from a cache imbalance, since Intel's L2 cache is dynamically shared between the cores.

As noted in the diagram below, each of the Phenom's three processor cores has its own 512K of L2 -- and 64K+64K of L1 -- cache, while the other components like the 2MB of L3 cache, System Request Interface and Crossbar Switch remain separate, and have equivalent base functionality no matter the number of cores present. The integrated DDR2 memory controller and HyperTransport link also operate on this philosophy, and provide the same bandwidth for a triple-core Phenom as it would for a quad core model.

Reduced power consumption and heat generation should also be advantages, with the new Phenom likely to slide into a thermal envelope between AMD's dual- and quad-core models. This will certainly be important on the desktop side, but could be a real boon if AMD moves the triple-core concept into the mobile market. It's no secret that Intel is prepping a mobile quad-core Penryn CPU for 2008, and a lower-priced, more battery-friendly AMD design could be a strong competitor.

Platform support will also be a nonissue, as a motherboard that supports the four-core Phenom will work with its more core-challenged sibling as well. The system will communicate with the processor in the same way, with the latter's data interfaces, switches, and integrated memory controller handling any internal differences among dual-, triple-, and quad-core designs. AMD has long promoted its consistent platform base, and now Socket AM2 owners will have one more upgrade option.

CPUs, Lies, and Videotape

Among the rumors and speculations swirling around the AMD announcement are that the chipmaker is doing this because it can't produce its planned quad-cores, or that AMD doesn't want the Phenom compared to Intel's quad-core parts -- odd, since the former will be the first to market late this year, with the triple-core version following in early 2008. Performance comparisons will be critical, but the triple-core chip certainly isn't a way for AMD to hide from Intel's quad-core parts.

Many pundits have called the triple-core a redheaded stepchild of the AMD line, and that using an uneven number of cores practically bastardizes the whole multicore concept. AMD has replied by pointing out the Xbox 360, which has done quite well in the gaming market using a triple-core IBM processor -- and noting that, with many games being ported from consoles to PCs, Xbox 360 code might be well suited for the new Phenom.

One point AMD freely concedes is that triple-core will mean greater production efficiencies by allowing the company to salvage quad-core Phenoms that might otherwise be tossed. For example, if during AMD's testing, three cores hit their rated speed but the fourth does not, then AMD can just fuse that bad boy off. It's also conceivable that a quad-core that runs at 2.6GHz may be able to run at 3.0GHz as a triple-core, letting AMD sell it in either configuration depending on demand.

This also presents a potential negative as efficiencies rise; AMD could find itself pressured to fuse off some perfectly good quad-core Phenoms and cut their presumably higher prices if the triple-core concept really takes off in the marketplace. This also brings up questions about how long the triple-core product line will last. AMD's most recent roadmap lists a triple-core model for its 45-nanometer-process line in 2008-2009, but we all know how fast plans can change.

A Compromising Position

While manufacturing and production flexibility are certainly important factors, the introduction of a triple-core Phenom still feels like a pure marketing move, and a good one at that. It's always nice to tweak a competitor's nose by doing something they can't, or finding a market niche all your own, and AMD might have done both. Whether or not it's physically possible, we doubt that Intel will offer triple-core processors in this generation, so AMD theoretically has the field to itself.

In its Phenom presentation, AMD has already highlighted its marketing plan: to tout the triplets as better than dual-core without the higher cost of quad. This also gives AMD a very logical three-tiered -- entry-level, mainstream, and high end -- product hierarchy. Performance is the only question left to answer: The triple-core Phenom will still need to prove itself superior to the Core 2 Duo.