But the rating system AMD is using is flawed and becomes more so with each release.
PR = 1500 + ((True_MHz - 1333) * 1.5)
That formula shows how at 1733MHz (the fastest Athlon XP available today), the PR rating evaluates to: 1500 + ((1733 - 1333) * 1.5) ==> 1500 + (400 * 1.5) ==> 1500 + 600 ==> 2100+. The difference between the true MHz and 2100+ rating represents a 21.2% overstatement from clock speed.
However, note also that as you move higher, a 2.0GHz Athlon XP's (2000MHz) value, for example, becomes: 1500 + ((2000 - 1333) * 1.5) ==> 2500+. The true MHz and 2500+ rating now represents a 25% (!!) overstatement from clock speed, an increase of 3.8% over that at 2100+! And the overstatement only *GETS WORSE* as frequencies increase.
AMD's upcoming Hammer, which is reported to be rated at 3400+, will equate to a 2600MHz (using the above formula), an overstatement of 30.8%. I'm forced to say it, but that is unacceptable.
The following series demonstrates this increase in discrepancy as frequencies rise (note that Ri = Relative Increase from previous model number, Ti = Total Increase from 1500+ baseline):
1333MHz = 1500+ = 12.5% overstatement
1400MHz = 1600+ = 14.3% overstatement (RI=1.8%, Ti=1.8%)
1466MHz = 1700+ = 16.0% overstatement (RI=1.7%, Ti=3.5%)
1533MHz = 1800+ = 17.4% overstatement (RI=1.4%, Ti=4.9%)
1600MHz = 1900+ = 18.8% overstatement (RI=1.4%, Ti=6.3%)
1667MHz = 2000+ = 20.0% overstatement (RI=1.2%, Ti=7.5%)
1733MHz = 2100+ = 21.2% overstatement (RI=1.2%, Ti=8.7%)
A difference of 400MHz represents a difference of 8.7%. While a total delta of 8.7% between low-end and high-end models is almost acceptable, anything beyond this point will produce an overstatement that is not acceptable. And the law of diminishing returns will play a greater and greater role (until Athlon XP undergoes an architectural change). In short, the model number formula breaks down for the three reasons stated previously.
In my opinion, AMD will be *FORCED* to take a different approach to measuring performance on its processors (hopefully AMD will do it in concert with an industry-wide movement). If the company does not change its approach and there is no industry-wide movement, AMD will surely face significant criticism from enthusiasts, consumers, and competition alike for its increasingly skewed values.
AMD is now faced with Intel's might and power! Even more damning to AMD's image, Intel will directly benefit from continued use of the model number system as processors ramp up in speed. Why? Because early AMD model numbers often represented a conservative comparison, e.g., a 1600+ Athlon XP was faster than a Pentium 4 1600MHz CPU. But as model numbers increase the overstatement also increases, and higher model numbers will definitely not show the overstatement the lower ones did. AMD's chips will begin to seem slow and outdated because the increase in clock speed necessary to carry out a linear progression of work performed is not commensurate with Intel's. Intel gains a noticeable amount with each release ... AMD's releases aren't moving quite as fast.
As a result, even without the introduction of Northwood, it would've soon been such that Intel's processors at a given numeric value significantly overshadow the performance of Athlon XPs at a model number of the same numeric value. This would even have been true on the standard benchmarks used by AMD to demonstrate its TPI! As the law of diminishing returns and the increased overstatement will only get worse, Intel can smile, because it's all working against AMD.
The Pentium 4 architecture, with its much faster front-side bus and less work per clock, has more "breathing room" before those same laws affecting AMD begin to slap it in the face. And, in the very near future, Intel can claim the following: "Even with their 'bad science' model numbers, our processors still outperform on the very tests they were using a few months ago to show how much faster they were!" And the laughter will be heard over the cigar smoke.
Bottom line: People will buy Intel's story hand over fist because it *WILL BE TRUE* (in the very near future). In short, AMD needs to address the performance separation between its model numbers and the Pentium 4 performance, for without a significant architectural increase (such as increasing Athlon XP's FSB to 333MHz, or incorporating the memory controller on the Athlon die), Intel's Pentium 4 will, undoubtedly, remain the speed king throughout the rest of Athlon's life.
Facebook
Twitter