This article should provide you with a useful foundation for understanding bandwidth, latency and performance in a variety of systems and settings. Done in the usual comprehensive Ars Technica style it pretty much covers everything from beginning to end.

When people talk about bus bandwidth this way what they're really describing is only one type of bandwidth: the bus's theoretical peak bandwidth. The peak bandwidth of a bus is the most easily calculated, the largest (read: the most marketing friendly), and the least relevant bandwidth number that you can use to quantify the amount of data that two components (i.e. the CPU and RAM) can exchange over a given period of time. In most product literature this theoretical number, which is rarely (if ever) approached in actual practice, will be cited whenever the literature wants to talk about how much bandwidth is available to the system. Let's take a closer look at how this number is calculated and what it represents. Read article now