Indeed we see here exactly what we expected from the previous pages. While difficult to see, Prescott outperforms Northwood in the 8-16kB area, where Prescott can remain working in its L1 area, while Northwood is forced outside to its L2 area. A similar effect can be seen more clearly in the 512-1024kB area, where Prescott's performance is much better, as it stays in its large L2 area, while Northwood is forced to move the data into main memory.

The question then becomes one of real world performance. We've seen that Prescott is stronger with certain data sizes, and weaker with others, but how will that relate to real world performance? Further, how will Prescotts other improvements factor in, as well as its lengthened pipeline? Let's take a look.