As we mentioned yesterday, this article slipped in what is probably the biggest news story in some time: socket A will “likely” last another generation.
Now this isn’t any rumor from unknown sources. This is coming from AMD itself.
Let me quote the most important parts of the story:
“. . . AMD said it will likely move its Athlon XP, which has been on the market since 2001, to the 90-nanometer manufacturing process and continue to make the chip through 2004.
“The new Athlon XPs, meanwhile, are coming to market to compensate for delays to the upcoming Athlon64 lineup. The first Athlon64 chips are set to hit the street in September, but won’t start coming out in massive volume until the debut of 90-nanometer manufacturing, Marty Seyer, vice president and general manager of AMD’s Microprocessor Business Unit, said in an interview earlier this month.
“Until the 90-nanometer manufacturing begins, Seyer said, the Athlon64 will be identical in size to the Opteron chip at 193 square millimeters. The Athlon XP, by contrast, measures 101 square millimeters, which makes it far cheaper to make.
“These factors mean there will be a limited number of relatively pricey Athlon64 chips on the market at first. Thus, extending the Athlon XP line will let AMD continue to make chips for the midtier and budget PC market.”
What About The Hammers That Are Supposed To Be Called XPs?
Many of you may say, “Wait a minute. AMD isn’t talking about real XPs; they’re talking about the Hammers that they plan to call XPs.”
It’s an understandable reaction, but evidence both inside and outside the article indicates that these are real XPs AMD is talking about.
For one thing, the AMD spokesperson referred to the AthlonXP being 101 sq. mm.. That’s the exact size of Barton at 130nm. It’s pretty unlikely a stripped-down Hammer would come in at exactly the same size as a Barton.
The comments made by an Abit spokesperson earlier this month now make a great deal more sense:
“The KV7 has a solid future ahead of it. Athlon 64 will definitely have a huge impact on the market, particularly the high end. However, the vast middle of performance users and gamers who haven’t won the lottery will not be much affected by this. . . . Keep in mind that this will only affect the very high-end of the market: mainstream sales will likely be dominated by K7 based solutions for another year.”
The article keeps making reference to “moving” the XP line to 90nm and “extending” the product line. There’s not a hint of Hammerness in any of the comment.
While it’s not inconceivable that AMD is jumping through linguistic hoops and actually referring to the alleged cut-down Hammer, the weight of the evidence points another way.
Nor is this necessarily an either/or. I suppose it’s possible AMD could do both.
Or perhaps the reports of an x86-64-less Hammer were really reports of a 90nm XP that someone just assumed had to be a Hammer.
What Is AMD Up To?
The spin AMD tried to give in the article was “Oh, Athlon64 is sooo big, we need a die shrink to make it prime time.”
Excuse me, but this isn’t Joe Sixpack providing the answer. This is a company whose job it is to make CPUs. This isn’t exactly like your mom plopping down by hand globs of dough to make batches of cookies. AMD has certainly known for a long time how big a 130nm Hammer with 1Mb cache is, if nothing else, they’ve been making Opteron for at least six months.
Sure, these take up a lot of die space, but if that’s recent news to the people making them, they all ought to be fired for blatant incompetence.
Nor does it explain what the problem would be with making socket 754 chips with less cache. After all, that’s been on their roadmaps for quite some time.
Making Athlon XP 90nm doesn’t do anything for AMD short-term either. It’s not like they have 90nm production ready and waiting to be used. These Athlon XPs won’t be out until AMD has converted some capacity over to 90nm, and if that’s the case, why would you not want to use that for second-generation Hammers instead?
There’s only two rational reasons for wanting to do this:
AMD didn’t feel they needed a safety net before. They feel the need for now. This tends to indicate that all is not well someplace with 90nm Hammer production.
Perhaps it will prove an unnecessary precaution, perhaps not.
The Six Month Strategy
There are lots of possible reasons why AMD is doing this, everything from problems in making the things (which could be anything from product yields to pulling manufacturing capacity offline to convert to 90nm) to the product directly (as in not being fast enough at 130nm) or indirectly (Windows x86-64 support) not really being ready for the mainstream.
It’s probably futile to diagnose AMD’s problems at this point. All signs point to a confused organization under the gun scrambling around and trying to make lemonade out of lemons.
AMD needs extra money, if for no other reason than they’re sitting on an aging product line with an eroding ASP. For whatever reason, things aren’t working out as planned. So what do you do?
AMD looks like they’re going to try to sell a limited number of CPUs at very high prices. Just to toss some numbers around, if they could get $400 each for 300,000 Hammers, that’s $120 million. While it certainly won’t bring AMD to profit, that would be enough to reduce their losses and give the financial world the impression that they actually are getting their act together. It’s actually rather the opposite, but it buys them some time to actually get their act together.
AMD probably can sell a limited number of first-generation Hammers at high prices without severe long-term reprecussions. There are those who do really need 64-bit and are willing to pay a lot for it; there just aren’t that many.
The great risk AMD looked like it was going to take was making more than the market was willing to take, leaving it with a lot of unsold processors and taking a big risk of Hammer being labelled a failure. By keeping the market small, AMD will keep that from happening.
Announcing that it is “likely” that AthlonXPs will go to 90nm gives AMD the flexibility to go with them if the second-generation of Hammers have problems (which should at the very least ease the minds of some financial types), or just changing their minds and dropping the idea if things go well.
Of course, having 90nm Athlon XPs around produces a negative; AMD will continue to compete against itself. But you can’t always get what you want.
This obviously could be great news to socket Aers; if these 90nm XP processors see the light of day; they get the chance for another CPU boost while keeping their current platform.
But AMD giveth, and AMD taketh away. This is the second go-round for 90nm XPs in the AMD roadmap, and four or five months from now, it may go away again.
The best advice is to sit tight, and don’t do anything. It’s hard to plan when a company keeps changing its mind; you can’t count on these guys.
Six to nine months from now, you’ll be looking at either a faster XP, a second-generation Hammer, or both.
Set your alarm clock, and hibernate until then. I wish I could join you. 🙂