Herb, Re: "You must be talking about what the Aberdeen report refers to as the “Pinocchio factor” that QuantiSpeed™ flawed methodology “will grow over time as the psuedo-equivalency gradually will become patently in inaccurate” and so AMD is already now planning to come up with a new measure of performance rating called the “True Performance Initiative” (TPI)."
The real problem is that.
1) There is no progress with the TPI, and
1a) There may not be progress for some time, due to the difficulty in coming up with a standard for processor performance.
Meanwhile,
2) AMD's model numbers are not sufficient for AMD to use until they do come up with a TPI.
2a) As you mention, they grow more optimistic all the time,
2b) They require updates to every change in micro-architecture that AMD or Intel does, and
2c) Despite denial at AMD HQ, there is every bit of a market perception that they are being compared to the Pentium 4, and not the previous generation Athlon.
This presents a little bit of a catch-22 for the AMDroid folk. For one, they believe that model numbers, while not perfect, is better than not having model numbers. If there was something better, then AMD would use it, but for now, they are "forced" to use a sub-standard measurement that will grow more faulty over time.
I read the Aberdeen report, and many of the points that it made were good ones. If AMD expects to be consistent in their rating system, they would increase the stability of their name brand, but they will lose credibility in what that brand name represents. On the other hand, if they update model numbers here and there to coincide with improvements in micro-architecture, they will maintain a performance advantage, but they will forsake the effectiveness of the brand name, to the frustration of both their customers, and the end user.
If they get rid of model numbers, then they are faced with the megahertz wars once again, and chances are they will not keep up. If they rush a TPI early, then they will of course lose everything, as a rushed system will surely have faults that can be equally exploited.
AMD made three major mistakes with the model number system, and now they will have to live with them.
- The first was using a different baseline for performance in mobile and desktop markets. They call a 1.33GHz Athlon XP a 1500+, but they also call a 1.30GHz Athlon 4 a 1500+. Clearly, the latter part will offer less performance, but AMD rounds it off, and hopes that no one will notice. Indeed, few people benchmark mobile systems, but such a move was still bold enough that it couldn't pass by entirely undetected. Now, AMD is threatened with having to define a baseline that clearly doesn't exist in the first place.
- The second bad move was to not take into account the performance improvements that Intel could achieve by doubling the cache, and adding 33% to the front side bus. With Northwood and 533MHz FSB in their chipsets, Intel has made AMD's modest model numbers look far less conservative. Now, AMD has to live with a competitive part that will likely scale just as well as theirs, except that it will be measured in real frequency, while theirs is ((frequency-333MHz) * 3/2).
- The third bad move is the plan to introduce new sets of model numbers midstream through their product lines. AMD did this both with the Athlon MP and with the Athlon 4. I soon expect them to do it again with Duron, once Intel transitions Celeron to a Willamette base. What kind of credibility does AMD have when a 1.2GHz part is called 1.2GHz, and a 1.3GHz part is called 1500+? Is there that big of a difference between those two parts? Can AMD still claim that they compare model numbers against their older core, when a 1.2GHz Athlon MP is the same core as an Athlon MP 1500+? Of course not. It sounds silly.
AMD made a few mistakes because they were threatened by more competitive parts from Intel. Now they are stuck with these problems, and they have to come up with more creative ways of explaining them away, and hoping that people latch on to the TPI concept.
I think that this is also dawning on the AMDroids, since they can't be entirely blind to AMD's mistakes. They just don't want the alternative, which would be no model numbers, because then Intel would really lay the hurt on them. As such, they are willing to lash out against the Aberdeen Group, and propagate the Inquirer rumor that Intel funded that research.
I don't know one way or the other, but I don't see what the difference would be. Either way, the research brought up some good points, but the AMDroids will ignore them to pursue the hope of lawsuits, which would protect them from further scrutiny.
All I can say is that Intel probably won't quit raising awareness about Quantispeed. It's not because Intel has anything better, because they don't. But fortunately for them, the battle that they wish to wage doesn't involve defending their own ground. Megahertz will still continue to be a unit of CPU clock frequency, and no one can argue against it. Arguing performance will always come back to the fact that megahertz doesn't measure performance - it's just that people see it that way.
AMD, on the other hand, needs to put up or get out, and I think it will make things increasingly difficult for them in the future.
wbmw |
