The Fall of AMD

(“Fall” as in season, not cessation. :))

The Inquirer has an article about AMD’s fall collection. Much of the estimated dating comes from Lynn Computer Products, which has a history of telling its customers when it can expect to see future products.

(It also has a history of getting products a little later than others, so odds are the products will first show up a bit sooner than they say.)

The reseller forecasts a 2400+ in October, a 2600+ a few weeks after that, and finally a 2800+ in December.

There’s a number of curious statements coming from the system integrators also interviewed for the article.

First, a 333MHz FSB is now being called not a big deal (which it isn’t, we overclockers do that all the time), which wasn’t what was being said in June.

Second, one of the sources stated that AMD would push the 2400+ into September, if that was possible.

Understand the competitive environment. We are looking at $190 2.4GHz PIVs in a month. That’s going to tumble high-end Athlon prices even more than they already are. It is hard to see how AMD is going to be able to get much more than $100 for an XP2100+. Neglected Nag 2200+s have only recently become fairly common; I’d bet that price will tumble now.

AMD needs to get these processors to at least the big OEMs as soon as possible in September/October for the Christmas season. If that doesn’t happen, it’s a sure sign that it’s not a matter of AMD timing, but sheer inability to do so.

Intel will have at least 2.8 and 2.6GHz available for Christmas (whether they can get the 3.06 in on time is questionable). The prices on them will be skyhigh, but that means even assuming a big discount, AMD could get really good money on 2600+s and 2800+s.

Why would they give that chance up unless they simply couldn’t deliver?

Names, Names

Then we have the part of the article which says that AMD has a big problem distinguishing these Whipped-Up Nags from Clawhammers and Bartons.

Huh?

What is so hard about telling the difference between a 2800+ XP and a 3400+ XP Clawhammer processor? Unless, of course, that Clawhammer isn’t a 3400+ processor.

Clawhammers will be called Athlons, too? Well, who were the dodos who came up with that? Wasn’t like the government mandated it.

You need new names? Took me a whole minute to come up with a solution. I won’t even charge for it. Call Bartons “Athlon XP+.” Call Clawhammers “Athlon XXP.”

Actually, AMD does have a very big, very real problem having Thoroughbreds and Bartons floating around at the same time, and this is what it is:

The calculation of PR will be different for Barton than earlier Athlons. Essentially, the PR rating will go up 600. As we pointed out at the time, this could cause AMD some real identification problems.

Introducing a 2800+ Thoroughbred will really mean AMD stepped into it. We will have the spectacle of a Thoroughbred processor running 400MHz faster than the Barton, but both will have the same PR rating.

Even more amusing, the Thoroughbred 2600+ will run 267MHz faster than the Barton, but the Barton’s rating will be 200 PR points higher.

Explain this to an inquiring Joe Sixpack and make him believe it. This I want to see.

Best Guess

AMD has already hinted that they’ve had some problems with the Neglected Nag. My best guess is that those problems have been a lot bigger than AMD has been willing to admit. Big enough that AMD had to redirect resources away from Hammer and back to the Neglected Nag.

My suspicion is that the Athlon design combined with a too zealous optical draw to minimize die size is largely to blame.

Now this could be a minor, or a major problem.

It’s a minor problem if the problems with the Neglected Nag are only just a matter of neglect, and fixable with proper care and feeding.

The major problem would be if AMD just can’t get much past 2GHz with the current design at .13 micron (at least not with acceptable yields). This in-and-of-itself wouldn’t be too big a deal, unless it also affects Hammer, and I think it does.

Think about it. A 2800+ Throughbred would be a 2.2GHz processor. A 3400+ Barton would be the same. A 3400+ Clawhammer is only supposed to be a 2GHz processor, and given AMD added 600 points to Barton for adding 256Kb cache, isn’t it at least possible adding 256Kb to Clawhammer gets you the 4000+ promised in 2003?

If you thought explaining to Joe Sixpack that fewer MHz really meant a faster CPU was a challenge, imagine selling “2=4.”

Think some more. How is Hammer different than an XP? It has a built-in memory controller. It has x86-64. It has a couple extra pipeline stages. And it’s tiny like Thoroughbred.

Now read the following from the article:

And finally, the trickiest bit of all, it has to do all kinds of clever things at its Fab 30 in Dresden to make sure that the Clawhammers and Opterons are ready to roll off the line.

I can believe that. Hmmmm, if I wasn’t too sure I was able to meet deadlines, I’d say something like that. too.

Speculation? Sure. I could be wrong. For AMD’s sake, let’s hope I’m wrong.

But this fits the relatively few facts we do have a lot better than “We won’t put out the best we can for prime selling season because we just can’t figure out what to call them.”

For those who don’t believe anything unless it officially comes from the company, what happens when the company refuses to speak?

Does a problem vanish if no one talks about it? Or do you call that something else?

Ed

Be the first to comment

Leave a Reply