It is similar to AMD's HyperTransport in that it is a switched fabric architecture (not a bus, although expect to see it called that by the great unwashed, i.e., press).
PCI-Express and PCI-X do not refer to the same thing. Sorry, Intel came up with the confusing nomenclature, not me!
PCI-X is a socket and bus definition for a high speed peripheral slot. You can connect PCI-X slots to a PCI-Express fabric, but I expect that AMD will connect the PCI-X slots to a HyperTransport fabric. It is my understanding that new video controllers will be adopting this slot definition soon. (I am not sure if the PCI-X slot is introduced as part of the PCI-Express definition, it may be.)
You can find a very good document presenting the basics and rationale for PCI-Express here:
It is interesting to note that PCI-Express and HyperTransport differ in an important way: PCI-Express is meant as a peripheral or communications fabric while HyperTransport can be used as a CPU interconnect or communications fabric. Theoretically, you can use HyperTransport for your multiprocessor and PCI-Express as the peripheral connection on the same motherboard. Quoting from the above pdf:
Non-Goals:
Coherent interconnect for processors, memory interconnect, cable interconnect for cluster solutions.
Figure 3 on p. 4 shows the generic concept behind the switch.
Also, take a look at the Figure 6 on p. 6 - this clearly shows the relationship between the PCI-X peripheral slots and the PCI-Express fabric (represented by the purple thing labled 'Switch'). It is interesting that PCI-X does not even route through PCI-Express in this configuration!
I am just now learning about this architecture, so if I got some of the details wrong you will have to forgive me (or not).
Market Data 
Markets 
AI
