This article is more than 1 year old
AMD's Athlon XP numbering scheme misses the (bench)mark
Right idea, wrong way of doing it
Analysis Just how is AMD going to persuade potential PC buyers that its new processor naming scheme, introduced yesterday with the Athlon XP chip, is a meaningful measure of relative performance?
AMD has a point when it claims that megahertz alone are no longer a sufficient measure of the raw speed of a processor - something PowerPC fans have been rightly pointing out for quite a while now - and the company is right to seek an alternative way of stressing the fact. Our concern is simple: we're not sure that slapping '1800+' on a CPU enough to convince punters that it's faster than a 1.8GHz Pentium 4.
Make no mistake, that's exactly what AMD is trying to say. For all its claims that the new model numbers signify the Athlon XP's performance relative to the old Thunderbird Athlons - purely for legal reasons; it doesn't want the Federal Trade Commission and Intel on its back - AMD's sample OEM advert compares the 1800+ XP to "competitive 1.8GHz PC processors". Since AMD doesn't actually make a 1.8GHz processor, T'Bird or not, competitive or otherwise, the chip company clearly has its arch-rival in mind.
Said mock advert doesn't hide the fact that the 1800+ runs at 1.53GHz, and AMD is to be applauded for making that clear. That said, PC makers are already slipping up and claiming the part runs at 1.8GHz. AMD will no doubt put them right, and as the naming scheme becomes established, errors like this should be few and far between.
It's all relative
But if the average buyer sees '2GHz Pentium 4' and 'Athlon XP 1800+' on two separate PC specifications, they're still going to ask what the clock speed is - '1800+' seems as no more relevant a number than the Pentium's '4' does. And even if the buyer makes the mental leap and associates '1800+' with 'more than 1.8GHz', they're still likely to consider that inferior to the Pentium's 2GHz.
One response to that assumption is to state that the benchmarks show that the 1800+ outperforms the 2GHz Intel part. But how many consumers are au fait with the latest benchmarks? And if they are, then they certainly don't need a new model numbering scheme to know how a 'low' clock-speed Athlon XP compares with a 'high' clock-speed Pentium 4.
Equally, many of the folk currently willing to look beyond raw clock frequencies will be buying AMD processors anyway, so again, the new naming scheme is fundamentally unnecessary.
So if all this QuantiSpeed stuff is irrelevant to those in the know and confusing to buyers who aren't, why bother? Well AMD certainly has to make it clearer to the wider buying public that clock speeds aren't the be-all and end-all of processor performance. It could even partner with Apple, Motorola and others to promote some new potential standard for chip performance by which all chips could be compared.
Intel's Pentium 2500+
We can't see Intel coming on board, certainly not while the chip giant continues to use clock speeds to differentiate parts on performance. The P4 architecture may offer little or no improvement over the PIII at sub-2GHz clock speeds, but there are indications that the performance ramp will grow considerable as Intel pushes the P4's frequency to 2.5GHz and beyond.
That will leave AMD - if it's not careful - struggling to match it on clock speed and model number. '1800+' plus is all very well if your competitor is running at 1.8GHz - if he's running at 2.5GHz, if you'll need to get a '2500+' part out pretty smartish.
And if Intel does adopt a similar approach, expect it to number its chips higher than the competition's. That's the trouble with numbering schemes based on relative performance - you're free to choose whatever you like to your new chips to be relative to. A case in point: AMD says its ratings relate to the Thunderbird Athlon, but then goes on to line its new XPs up against P4s. That's because, as we said earlier, AMD sells against Intel.
To its credit, AMD is at least allowing the benchmarks that led to its naming scheme to be independently verified by Andersen. We're sure Andersen will confirm AMD's findings, but since all they prove is that the Palomino is faster than the Thunderbird, clock-speed for clock-speed, we're not much sure what help it will be.
Only if Andersen - or another independent organisation - establishes a universal baseline by which all chips can be compared, regardless of instruction set and based on real-world application performance - will any attempt at moving beyond clock frequencies ever be meaningful.
AMD's pseudo-independent approach doesn't come close - as we'll see just as soon as a rival offers a '2000+' part. However, it is a step in the right direction, and if it's True Performance Initiative garners broad industry support, we could one day have a good measure of chip performance that doesn't rely on clock-speeds or arbitrary comparisons to older chips. ®