How Many Desktop Duallies In 2005?

Add Your Comments

Digitimes has an article stating that Intel won’t ship more than 500,000 Pentium Ds in 2005.

This is quite, quite different from what Intel is saying, to quote:

“We’re shipping 100,000 this quarter, and we’re going to ship millions by the end of the year,” said Gerald Holzhammer, vice president of Intel’s Digital Home Group. “This is a big deal for us. It’s the first time dual core will make a real impact on the marketplace.”

Obviously, two very different crystal balls. Which one is crack, and which one is cracked?

Here’s a few items to consider:

If I were a Taiwanese mobo maker, I’d probably not want to put too much effort into Smithfields. Let’s face it, it’s a cobbled-together quick-and-dirty place holder until the real deal shows up at 65nm in about six months, maybe less. That’s a pretty short life cycle.

On the other hand, after the success of the Centrino platform, Intel wants to bring its “One ring to rule them all” road show to both the home and business desktop, which if successful will put the squeeze on the Taiwanese.

Note the title of the Intel guy, VP of the Digital Home Group. That’s the whole point of this guy’s division, to bring the Intel platform into the home entertainment field, and keep others out of it.

When you hear Taiwanese mobo makers are saying that they’re not going to have all too many chipsets to play with, well, you have to at least suspect that the Digital Home Group and other Intel groups are the main culprits.

Of course, the success of Centrino does not guarantee the success, much less instant success, of any desktop descendants.

So the divergent numbers reflect on the one hand a rather pessimistic view, and on the other, what may be very wishful thinking on the part of a rather biased participant.

The truth probably lies somewhere inbetween, though I would suspect the Taiwanese are closer to the truth.

What About AMD?

Not surprisingly, I got some emails “explaining” to me why AMD couldn’t possibly ship out a $300 dual-core processor any time soon. The term “fab capacity” came up a lot.

Let’s examine this closely:

When AMD’s Dresden plant was built, it was supposed to be capable of making roughly 10 million CPUs per quarter at 130nm running flat-out.

Currently, AMD is producing a bit over 8 million CPUs per quarter. About half of them are 130nm Athlon XPs, the rest Hammers, some 130nm, probably more 90nm. It is safe to say AMD is getting less than $100 per XP, probably a good deal less than that. It is just as safe to say that AMD is getting rather more than $100 per Hammer sold, probably a good deal more than that.

It’s taken AMD almost two years to get desktop Hammers to the crossover point, but from this point on, Athlon XPs will be quickly phased out. AMD will have to sell millions more Hammers than they have in the past.

Per die size, 90nm Hammers are roughly the same size as the 130nm Athlon XPs, so there will be no unit production bonus shifting current XP production to Hammer production. There will be a bonus shifting 130nm Hammer production to 90nm, though it’s more in the nature of removing an anchor from capability.

Athlon 64 sales were minimal (around 100K) a quarter until prices dropped below the $200 mark. Hammer sales are very heavily skewed to the low end (AMD said it, not me). There’s no evidence of AMD Hammer shortages; all indications are that supply and demand are pretty much balanced.

It’s very hard to argue that AMD can’t afford to sell a dual unit for $300 when they are using half their capacity to make single XPs for less than $100. It’s very hard to argue that you can’t make a decent profit by charging $300 for an item that might should cost Dresden $50 or less to make.

One might say that AMD is better off selling two single-cores than one dual, but given that AMD’s single-core sales are mostly at the low-end anyway, there’s probably little revenue or profit difference at the $300 level (and AMD would probably be justified in charging a bit more than Intel).

What’s more important strategically is that this is an opportunity for AMD to shine and make a good first impression on this new playing field. If you want buzz, if you want word-of-mouth, you’ll get a lot more effective buzz if a lot more people actually own the things. This is a field where AMD can match Intel close-to-one-on-one in a new field for at least a while. AMD doesn’t get many shots at doing that. They ought to take advantage of it under any circumstance, and especially so if Intel is having teething problems.

If what the Taiwanese are saying about Intel’s offerings is fairly close to the truth, it is highly unlikely that a $300 dually is going to overwhelm Dresden. Maybe a few hundred thousand extra units per quarter. There should be enough capacity at Dresden for that, and even if there isn’t, then what were all these foundry deals with third-parties about?

It should be noted that in a year, AMD will be happily selling dual-cores at the Sempron level. Yes, they’ll be at 65nm, which saves a bit, but you know they aren’t going to cost anywhere near $500 or even $300. They probably won’t even cost half of that.

If you sell a dually at $300 today, people will know that they won’t get better for over a year. Start selling them at $300 six months from now, and people will be less inclined to buy them knowing they’ll be obsoleted fairly shortly.

Yes, Intel may be having some problems, but that’s even a better reason for AMD to push ahead on duallies. You have superior performance, you don’t need a special mobo to run the chip, why not try to match Intel sale for sale among the early adopters? Make this new playing field an even one for 2005, and you’re in much better shape against Intel in 2006.

Instead, we see a strategy that will sell AMD maybe 100-200K units the rest of the year. Even if the Taiwanese have hit the nail on the head for Intel, that means Intel will dominate in this new field right from the start.

It makes no sense. Unless . . . .

The Catch

All that I’ve said relies on one assumption: AMD is not having any serious problems producing the goods.

Unfortunately, based on AMD’s past history, that’s an awfully big assumption to make.

Over the past few years, every time AMD has done something that hasn’t made much sense, like not talk about new products, or hyperpricing them, one kind of production problems or another has been the reason. Every single time.

When this happens, AMD does the same thing every single time. They deny (or actually, say things that sound like denials) any problem, until they’ve fixed them. Then, and only then, will they ‘fess up.

No, they don’t flat-out lie, but they provide slick, technically-accurate but misleading answers, much like President Clinton said, “I did not have sex with that woman,” using a legal rather general definition of oral ministrations.

To give just one example, in the past, they’ve said things like “Yields are fine.” That sound pretty good until you realize that “yields” all by itself mean little. A working Hammer that can only run at 800MHz or 1400MHz is hardly any good.

In other words, if Problem C is giving Dresden nightmares, if someone asks, “What’s wrong?” they’ll say “There’s no Problem A or B.” If someone is determined or lucky enough to ask, “Is Problem C the problem,” (and I can’t say I’ve heard anyone nail it at their conference calls yet) the answer will be the same, “There’s no Problem A or B,” sometimes with a “We’re not going to talk about that.”

Then six months later, you’ll hear them say, “Oh, we had Problem C, but we’ve fixed that.”

Given this track record, while we don’t exactly know what the problem is, it is pretty safe to say there is some problem, somewhere.

Maybe yields of working duallies are the problem. Maybe yields at the desired speeds are the problem. Maybe there’s no technical problem at all, but AMD hasn’t been able to afford converting capacity at Dresden from 130nm to 90nm as quickly as they might like. Maybe Hector Ruiz still thinks he’s at Motorola, and still thinks CPUs deserve to cost a lot. Who knows?

However, for us, the why doesn’t really matter, the what does, and the “what” is “AND can’t deliver and can’t take advantage of opportunities when it counts.”

And that’s sad.


Leave a Reply

Your email address will not be published. Required fields are marked *