The
PR (performance rating) started as an AMD Marketing tool.
It started as a way of comparing the performance of the Model 6 (Palomino) processors compared to AMD's Thunderbird (Model 4) CPUs. Refinements in the design of the Model 6 CPUs provided the same performance level as the Model 4 CPus, but at lower clock speeds (lower frequencies). AMD carried on the
PR as they migrated to the Model 8 (Thundertbird), Model 10 (Barton) and now the new family of 64-bit CPUs.
A better processor design is capable of doing more work per clock cycle, so a Barton Model 10 running at 2167mHz (PR3000+) is rated to do as much (or more... the "+") work as a Thunderbird Model 4 running at an actual 3000mHz.
(Note: I seriously doubt anybody ever got a T-Bird up to 3000mHz unless they used LN.)
Peformance differences are affected by a lot of things besides cache size. I'm sure some one here at TechIMO knows a lot more about CPU architecture and design, and can either explain what things got better/more effective at a given speed, or point to a resource.
A lot of people think that the
PR is supposed to show equivalence between an AMD and an Intel CPU. AMD Marketing never officially proposed or promoted this, but didn't actively deny it when people made those assumptions. Besides, in running a whole variety of different benchmarking programs, you will usually find that the
PR rating of an AMD CPU usually does match quite well with an equivalent clock-speed Intel CPU.