65nm Talk

There’s some talk about Intel pushing the arrival of 65nm chips some months ahead of schedule.

First, presuming Intel does it, all they’ve done is end up where they initially planned to be. 65nm chips were initially supposed to be out 4Q 2005, then that got pushed back to 1Q 2006, now they’re back to suggesting 4Q again (end of 3Q means the same thing).

Second, it shouldn’t be surprising at all that Intel would have 65nm chips available. After all, they have a development fab to do just that in Hillsboro, Oregon. That’s what they’re supposed to do. Seeing chips out there now basically just tells us that Intel is not having the kind of problems they had with 90nm (and if you’ve lowered your goals as to speed, it wouldn’t be surprising for them to have an easier time of it).

Finally, getting 65nm chips out of development is one thing; getting a few fabs converted to 65nm and cranking out tens of millions is quite another. It’s not like Intel can say today, “OK, good enough,” and starting making millions tomorrow. Under normal circumstances, it takes a few quarters from product introduction for Intel to get enough new capacity make half the mainstream chips with the new process.

So while 65nm chips may show up this year, they’ll probably won’t be commonplace until early, or not terribly early, 2006.

However, Intel will have more incentive than usual to implement as quickly as possible. If the future is dual-core, Intel needs to make those chips at 65nm as soon as possible simply because 90nm dual-cores are huge, and Intel certainly isn’t going to make its usual profit margins selling essentially two Prescotts for little more than the price of one. It takes 65nm to get the die size of a dually to be roughly the same size as a Prescott.

Where Does This Leave AMD?

Behind.

AMD has had most of the technical high cards in its hands for the last six months, and has done squat with them.

Consider this: Hammer is the best CPU out there. Up to now, by historical measurements, desktop Hammers have been a sales failure. It usually takes six-nine months for a next-generation CPU to become the major unit seller for the company. Desktop Hammers have been around eighteen months, and won’t manage that for another three months.

The only brilliance AMD has shown with Hammer is tossing up enough smoke and mirrors for people to not realize AMD hasn’t been able to produce them en masse. They haven’t been able to deliver the product, apparently because at AMD, SOI really stands for Slave Of IBM. From the little we know, it seems like they’ve been largely waiting around for IBM to (reluctantly) give them the formulas for good SOI sauce, then struggling to make it work.

Yes, revision E will be out shortly, this looks to be the answer. A quarter or two thereafter, we’ll see another humongous 90nm dual-core come from them, which they won’t be able to charge a lot for except at the high-end because Intel won’t. Just about when it shows up, Intel shows up with 65nm chips, and starts ding-donging the world about why dual-cores are now essential, and AMD is stuck selling big chips at relatively low prices until they get their new fab going sometime in 2006.

Another squandered opportunity.

Yes, AMD is saying that they’ll have 65nm chips by the end of the year, and they had better be able to crank out a few by then, just to validate the new 65nm fab, but that’s not the same as going into serious production. After all, the history of Hammer ought to tell us the difference between “making” chips and making chips.

The article linked above suggests that even the computer press is getting wise to this, and that sentences are no substitute for silicon.

How Relevant Is All This To Me, Mr. Overclocker?

The answer to that is probably the biggest piece of news of all: very little, and that ought to tell you we live in a new era.

In the past, Intel getting a big lead over AMD in a process shrink would be big news, simply because a shrink meant a big speed (as in MHz) increase. Not any more. They may end up somewhat faster, but certainly no 40-50% of more like we have seen in the past.

Any dramatic increase in speed will be held hostage by the presence or absence of multithreading programs. Code, not circuits, will be the key the next product cycle.

No, this fight will be a battle of the beancounters, who can make this duallies cheaply, quickly, and the results will be shown in the profit-and-loss statements, not the benchmarks.

It is quite likely that over the next year or two, AMD will win all the performance comparisons, but get beaten up in the financial wars. Not beaten to death, just left incapable of doing any beating with a better product. Maybe “stunted” is the best term to use.

I’m not much for conspiracy theory, but I’d really like to be a fly on the wall at a few IBM meetings. I suspect a few people have some interesting thoughts about what they might like to do with AMD in a couple years.

Ed