A year ago, when AMD introduced their Athlon XP family of processors, they also unveiled their True Performance Initiative that was intended to show customers the real performance that processors from AMD can offer. The first core to power the Athlon XP processors was called “Palomino” and had numerous architectural advantages compared to its predecessor “Thunderbird” core. These architectural innovations conditioned the fact that different AMD processors` of the same core frequency performance was not the same. That time the company started to mark its CPUs with model numbers instead of core-clock which should help customers to compare between various AMD processors. Circumstantially, the model numbers also allowed to compare the Athlon XP CPUs to Intel`s Pentium 4 microprocessors, where the model number corresponded to the core-clock of the rival. Basically, it was very useful for consumers, as they had almost clear picture of CPU speed behind their eyes.

With the introduction of the latest AMD Athlon XP 2400+ and 2600+ it became clear that performance rating AMD used for its processors does not longer represent the true speed some CPUs can or cannot provide in certain benchmarks or applications when compared to appropriate Pentium 4 models. Firstly because K7 architecture is not as efficient at frequencies about 2GHz as it used to be at 550 – 1500 MHz, secondly due to software optimisations for Pentium 4 processors and SSE2, thirdly because the rival`s microprocessors eventually got 533MHz system bus and 512KB of L2 cache, receiving additional performance boost without increasing the core-clock. In future, the situation will be even worse for AMD and its performance ratings: Intel implements HyperThreading technology in the fourth quarter, giving their CPUs another architectural advantage and increases the size of L2 cache on their Pentium 4 “Prescott” processors to 1MB by the third quarter next year. With both architectural supplements, chips from Intel get another speed increase.

In order to be able to proclaim their new model numbers reasonably, AMD has added several new benchmarks in their performance determination method and even asked PricewaterhouseCoopers LLP to examine and verify their assertions. AMD has also changed their formula used to calculate the rating for the CPUs (see this news-story). Nevertheless, it seems that the current formula will not live long: according to this site, AMD hopes to release the new speed measurement tool by early next year. Taking into account that x86-64 CPUs will appear sometime in the first quarter 2003, we believe that the new method to determine the performance of a CPU will be proposed specially for code-named Hammer processors. We think that changing the way of describing processor`s performance is a quite logical move as AMD`s future CPUs will have totally different architecture against today`s processors. Hence, we will not be able to compare the rating numbers of Athlon and Athlon x86-64 chips.

Although performance ratings in general will make it easier for a consumer to distinguish between the CPUs, we still have no information if such initiative may be supported by Intel, who still has more than 70% of microprocessors market share. Considering comparatively low core-clock of Banias CPU, Intel will have to explain the end-users that these processors are even faster than the Pentium 4-M and I have no ideas except a kind of performance rating or index number to do so. In fact, Intel already had their iCOMP index long ago, but they dropped it in the mid-nineties.

Well, it seems that the industry really needs a tool to measure and describe CPU performance. If the archrivals combine their efforts, we will have a precise and adequate solution. If not, we will be confused as we never have.