Replies to post #40797 on Advanced Micro Devices Inc (AMD)

[sgolds]

Tenchu, you are straining to support Intel's MHz Myth while criticising AMD. Consider that your argments may be colored by your investing preferences, a dangerous thing for certain!

What's the "error"? Clock speed is clock speed, useful for comparing two different CPUs in the same family of products. Obviously a Celeron running at 2.6 GHz isn't going to be the same as a P4 running at the same speed.

That is the fallacy in your argument. The average Celeron purchases does not know that the Celeron runs significantly slower than the equivalent P4. Intel has been cynically fooling these folks for years. Even now when they have the 300-series numbering (a good move) they are permitting advertising that highlights the MHz right next to the model number, nullifying the goodness of that model number.

Now AMD is joining in the game. It is a sad day for the CPU business.

[Not a Short]

What's Sempron supposed to be compared to, and who's stopping AMD from starting Sempron with a ModelHertz rating of 20 million?

Amd defined a line in the sand with the first Sempron all other semprons are compared to the benchmark suite results of the prototypical Sempron. See my other posts today for info on what benchmarks they use.

In short there can't be a 20,000,000+ sempron unless they can someday make the cache, fsb, memory so fast that the bechmark suite is no longer valid (I doubt it considering disk accesses are part of the system too).

[j3pflynn]

Tenchu - You know very well that it's in comparison to Celeron, just as Athlon XP(contrary to PR legalspeak)was in comparison to P4.
Paul

[wbmw]

Tenchu, Re: Clock speed is clock speed, useful for comparing two different CPUs in the same family of products. Obviously a Celeron running at 2.6 GHz isn't going to be the same as a P4 running at the same speed.

I used to argue the same thing until I realized that you can't combine engineering with marketing. Megahertz is a metric - just like cache size or even operating voltage - and it's a decent way to differentiate product skus. However, you can't educate people as to what megahertz actually means, and no matter how hard you try, it just comes across as another number. Arbitrary ratings are actually better, as long as they convey a simple and consistent message to the end user.

I think Intel's BMW-ish model numbers and AMD's Opteron model numbers do just that. A 335 or 248 has absolutely no engineering background at all, but it does provide the consumer with a way to make a more educated purchasing decision. They know that a Celeron D 335 is better than a Celeron D 330, no matter how many clock cycles there are per second or how many bits of static memory there are, or even how many bytes per second of data transfer there are on the interfaces. Given the number of more subtle micro-architectural differences these days between products, it simply makes more sense to use model numbers.

Now, the problem with Sempron model numbers is different, because they are neither simple nor consistent. If AMD wants a 4-digit model number, they should put it in a range where it does not look like megahertz, because that is a ploy to intentionally confuse the end user: not simple. Second, they should not pick a number that overlaps the numbers of a higher performing part, because that implies that the lower performing part offers equal or greater value: not consistent. Opteron and the Athlon 64-FX do a better job with model numbers because they are both simple and consistent.

Re: who's stopping AMD from starting Sempron with a ModelHertz rating of 20 million?

So you fear metrics that are prone to mass inflation.

Personally, I don't think there is a fear of that happening. AMD would be shooting themselves in the foot with a model number of 20,000,000 because it is no longer simple to the end user. Moreover, no one is going to be fooled into believing that AMD's part is several orders of magnitude higher in performance to Intel's part. Case and point: Intel's model numbers use 3 digits instead of AMD's 4, yet there is no performance confusion between the two. Moreover, AMD gives Athlon 64-FX a 2-digit model number and charges a premium. Again, there is no confusion because people assume that products are different when model numbers span a large range or have different numbers of digits.

Keep in mind that the microprocessor business is not the first to use model numbers. The auto industry has been big on model numbers for a long time, for example. The mass market has essentially already been trained on how to identify with model numbers, and the trick here is to ease people in on the idea that the number is now arbitrary, rather than an actual engineering metric. You don't sell cars based on horsepower, even though it might show up on the spec sheet, and the same can apply to CPUs without the fear of mass inflation in the model number system.