They held this event yesterday, you can see the slides for it starting here.
Although AMD has been known to tell investors and analysts one thing one day, and change their plans quite a bit the week thereafter, this does provide at least a vague roadmap for the next few years.
Make Them Small and Cheap, and Move Quickly
That’s AMD’s goal over the next few years in a nutshell. They talk incessantly about how their CPU cores will be smaller than Intel’s, and thus they can make more CPUs from a wafer than Intel.
This is not as convincing an argument as one might think. The current Willamettes are ridiculously big, but the difference between AMD and Intel narrows considerably at .13 micron. The Northwood will still be about 40% bigger than the Athlon, but that’s a lot better than the current 100% or so.
We’re also not talking about a large dollar difference. Right now, it probably costs Intel $40-$50 more to make a Willy than it costs AMD to make an Athlon. Maybe that difference becomes $20 at .13 micron.
So long as Intel gets substantially more money for its processors than AMD, though, the competitive advantage of lower costs becomes irrelevant. It’s like Ford saying a Taurus has lower production costs than a Lexus. True, but so what?
In the short term, the primary benefit of lower costs is that AMD can still keep its head above water unless Intel pricing changes from predatory to near-suicidal.
AMD plans to use just Dresden the next few years to make CPUs. .13 migration is starting or will start shortly. That we knew.
What we didn’t know is that AMD is currently planning .09 micron migration starting in mid-2003, and will apparently partner with someone else to build a .065 plant opening in 2005.
When AMD migrates, it expects to take a year to migrate to a new process. In contrast, it takes Intel more like two. Intel may get to a new process first, but AMD expects to normally catch up and surpass them in the proportion of processors using the new process fairly quickly.
Throughbred Looks A Bit Stunted
If you only look at one slide, look at this one. This basically gives you AMD’s rampup
through early 2004.
It’s an estimate of how powerful AMD processors will be (apprarently using PR).
Throughbred actually isn’t going to be around all that long: Barton, using SOI technology along with .13 micron, will come the second half of the year.
The Palomino will go to 2200+ (1.8GHz). In the second quarter, Thoroughbred will debut as a Model 2400 (1.93GHz?). Third quarter, there will be a Model 2600 (2.06GHz?). Then it’s Hammer, Hammer, Hammer.
The Sons of Athlon won’t vanish, though: they’ll stay on the roadmap into 2004 and apparently will go through .09 micron migration and end as about a 4000+.
How overclockable will the Thoroughbreds be? Good question. A few days ago, Intel announced that they had dissected an Athlon and found signs of .13 micron technology already there. AMD semi-admitted to that, but denied that they were sacrificing tomorrow to save today.
After looking at this roadmap, seems like Intel is a little closer to the truth. The ramping up stalls during the Throughbred period, only to pick up again with Barton.
We’d probably have to torture something to get this confirmed :), but I suspect AMD has been doing this for a while. They borrowed Palomino technology to make the later TBirds, and apparently borrowed from Thoroughbred technology to improve the Palominos.
Intel isn’t promising much during the same period, either, so this “stall” may be real, or just a sandbag. I will be less than shocked if AMD advances some Barton technology into the later Thoroughbreds.
This slide shows a Model 3400 Hammer at .13 micron debuting at the end of 2002, a Model 4000 coming the first half of 2003, then a .09 micron Hammer at 4400 the second half of 2003.
3400, 4000, 4400 of what?
Surprisingly, the answer looks to be “MHz.”
If you go back to the infamous slide 35, it shows the estimated PR rating of the Hammer chip.
Remember, the initial model is 3400. If that were PR, the slide would show a yellow dot around 3400, but it doesn’t. Instead, it shows a dot around 4300. That why I think the Model 3400 Hammer is probably a 3.4GHz chip.
Maybe it’s already here with these AGNGA chips, maybe it’s to come, but AMD will have to improve the initial Palomino chips to even meet their roadmap. Patience will be rewarded.
From an overclocking standpoint, the Thoroughbreds look to be a bit of a question mark; they may or may not crank up as much as we would normally have thought due to .13 micron.
AMD obviously wants you to pick up the Hammer in 2003, and it does look to be pretty appealing based on the slides. Whether or not they can deliver on time is another story (and the roadmap indicates initial debut September-December 2002.
I see three reasonable AMD upgrade paths at this point:
Early: Palomino now or shortly thereafter, Barton/.13 Hammer, .09 Hammer
Middle: Thoroughbred, .09 Hammer
Late: Barton, .09 Hammer