Below is a brief summary of the presentation by Hester, detailing K8L's specification, something called DICE and AMD's vision of modular "building blocks".

AMD Talk Modular
Before the details were given on K8L, Hester made it clear that AMD's new philosophy moving into 2007 will centralize around modular design. It is clear that AMD believe the modular approach will allow them to deliver architectural tweaks quickly and efficiently when the need for them arises in market segments.

If you look inside a micro-processor today you will find perhaps 10 to 15 major functional groups, and these are mostly applied within a monolithic design. It is possible to identify each functional group individually but the interfaces between them are not as clear or definable as they could be. The problem with this is that it takes a lot of engineering resource to modify how these individual yet undefined pieces work to create an updated processor model.

AMD’s modular concept is to better define each of the building blocks so to easily construct optimal designs for each of the market segments. This will allow AMD to do two things. Firstly, it will enable them to have one core architecture but develop very different processors for each market segment, and secondly, it will be far easier for AMD to change specific functional groups so to increase performance in the server market, or desktop market, or the low power market, basically whenever they need to react to the needs of a specific market they can quickly and cost effectively provide solutions. We may never again see a new architecture in the traditional sense as AMD can continually update any of the building blocks to effectively create a new processor family.

click to enlarge

Examples of modular design in action

If you look at the diagram above it indicates how modular architecture works. The “building blocks” as AMD like to call them, are shown within the yellow border and can be configured differently so to better construct optimized processor designs for each of the different markets. The blocks in red are two example configurations of how it can be used to better serve different markets, even unforeseen markets. The one on the left is an example of a dual-core design built to serve the desktop market, on the right an example of a quad-core design better suited for server demands, but both using the same core building blocks.

The gray box named “other” represents extra capacity for new functional blocks which can be added in the future. Possible uses here could be for co-processor functionality, and so on. The genius behind modular should be clear to all.