Pentium 4 analysis part2

While real-world benchmarking provides some insight as to how many different design decisions ultimately pan-out in terms of performance, it's not always possible to determine the impact of any given single component to a microarchitecture from such results. What does a Quake 3 timedemo tell us about the effective performance of the floating-point unit or the accuracy of the branch predictor? Is it even possible to determine the impact of these factors without interference from the graphics accelerator, memory interface, storage device, operating system, and even, in some cases, the monitor? In short, not with Quake 3.

To isolate and evaluate these distinct components of any given microarchitecture, we use benchmarks based on key algorithms which exploit the region of the processor we are interested in while minimizing interaction with other components as much as possible. In essence, we can use these "synthetic" benchmarks magnify both the weak and strong points a given CPU for closer evaluation. By very design, these benchmarks are not intended to be unilateral performance indicators, but instead to provide insight into exactly what that real-world performance is made of.