AMD: The Next Challenges . . .

The Inquirer reports that AM2 systems just aren’t much quicker, and aren’t going to get much quicker before release.

This would appear to give Intel a significant through hardly overwhelming lead when Conroe comes out.

What is this likely to mean?

One important point about Conroe and Company is that there’s not going to be a whole lot of them made in 2006.

Take a look here (btw, the article also has initial Conroe prices), and you’ll see that only 10% of their production in Q3 and 20% in Q4 will be Conroe.

Nothing unusual about that, those percentages are pretty much standard Intel rampup, but Conroe won’t become the mainstream CPU until early 2007, and its impact on overall AMD sales will be muted until then.

So AMD has some time before Conroe gets the chance to do real damage to them.

Let’s look at the situation at Dresden.

The AMD folks have been talking about a new feature here, some extra floating point units there, but again, you have to look for what they’re not talking about when you’d expect them to talk about it. What are they not talking about? They’re not talking about 65nm.

You have a company that until recently had one fab, Fab 30, started at 180nm, went through 130nm, now is making 90nm CPUs. Now they’re starting up another one, Fab 36. That was supposed to be 65nm, but that changed, and it’s going to be 90nm for at least a while.

Eventually, Fab 36 will be converted over to 65nm, but when will they be able to make a serious number of 65nm chips, serious as in half their CPU production?

And once AMD gets on the 65nm bandwagon, just what are they going to do with good old Fab 30? They’ve never really said. Maybe you’ve assumed Fab 30 would get converted to 65nm and continue cranking away, but you would assume wrongly.

As you can see here and here, the talk is more about making chipsets for others than CPUs for themselves.

It’s likely retrofitting Fab 30 to 65nm would be an expensive proposition, after all, Fab 30 was initially equipped as a 180nm plant, and while a lot of fab equipment can be used or tweaked to handle several process shrinks, this can only go so far.

At this point, though, important as it is, money becomes a secondary factor. You don’t decide today, “Let’s convert Fab 30 over!” and have it done by next week.

Complicating all this is moving to dual-cores. Roughly, going to dual cores means forfeiting any capacity increase from a process shrinkage.

So AMD is running flat-out with one fab, while trying to get another going. Both will be making a larger and larger proportion of outsized CPUs until they can move to 65nm, which they’ll do at some unknown date provided some unknown problems (equipment? SOI problems?) get fixed. You need 65nm to get the chips down to a decent level, but all your capacity is at 90nm, and you can only convert/create new capacity from one of the two places, and the more 65nm chips you make, the less anyone wants the 90nm chips from the old place.

And if this weren’t bad enough, your competitor just blind-sided you with better-than-expected results, so you’ll have to revamp or at least rush any improvements you had in the pipeline to stay competitive.

Handling all this over the course of the next year is not an enviable task, and you have to execute perfectly to be ready when Intel starts really pumping out Conroes in early 2007.

Yes, they have Chartered. Wonderful. Just when you’re juggling a stack of plates in Dresden, you have to tell people half-way around the world how to do it, too.

I’m not saying this is an impossible task, but things could easily get really ugly really quickly if problems crop up, and the key issue is 65nm.

Today, we’ve looked at the AMD/Intel competition on a macro level. Tomorrow, we’ll look at how it’s likely to affect you.


Be the first to comment

Leave a Reply