Like Intel, AMD benefitted from more sales and a somewhat better selling price last quarter.
The company as a whole essentially broke even, the CPU sector a little above, the rest of the company a bit more below.
It is very likely AMD will make a profit, perhaps $100-$150 million, next quarter, primarily due to an increased number of Hammers available for sale.
However, if you read the tea leaves, there’s some troubling indicators for next year.
90nm Delays?
It looks like 90nm chips will be delayed somewhat. They had been expected the end of the second quarter. AMD now says that production starts will begin the middle of the year, which was when these things were supposed to show up.
AMD also made some strange comments about letting the market decide when it wanted 90nm chips. Well, unless things are going very, very badly, there will be 90nm Prescotts early next year.
More importantly, making 130nm Hammers is hardly good for AMD’s production capacity. These chips chews up nearly 200 sq. mm of space, or almost twice as much space as Athlon XPs. To get Hammers down to around XP size (and thus be able to make about as many Hammers as XP from a single wafer), AMD needs to go to 90nm as soon as possible.
AMD illustrated this problem by saying that while half their wafer starts would be Hammer wafers in Q1 2004, they didn’t expect Hammer crossover (i.e., selling more Hammers than XPs) until the end of 2004. This is a far more modest target than we saw in the purported roadmaps, which had crossover in Q2.
All this points to a somewhat delayed 90nm introduction, maybe August or September.
The Long Beta
Microsoft said that Windows for AMD64 won’t be out until the latter half of 2004. Public betas ought to be available soon, though.
Those who have advanced copies of this beta have generally reported that the OS itself works pretty decently, but drivers are a problem. It’s a much bigger problem than usual because close doesn’t count here; apparently you can’t use older 32-bit drivers at all. Either MS or the manufacturer have to provide a new driver for all the devices out there, and nobody seems to be in any rush to do so.
Not the kind of thing OEMs want to hear.
What we’ll probably see is something rather similiar to what happened with Windows 2000. There will be an extended beta period, with drivers added in dribs and drabs as time goes on. Keep an eye on what the beta folks are doing and getting from MS. When MS comes out with a release candidate, figure the final is three/four months away.
Clamming Up
Right now, both major CPU companies are clamming up.
Intel is playing, “Yes, we have no bananas,” with Prescott. There’s absolutely nothing wrong; there just aren’t any.
This is pretty bad, but Intel is a font of knowledge compared to their competitors. Sure, we’re clueless about Prescott, but we do have a pretty good about pricing and socket T and what we can expect if Intel takes care of its Prescott problems.
For AMD, well, you’d probably be better off calling the Psychic Network than asking anybody, including Hector Ruiz, any major aspect of what AMD will be doing six months from now. It’s bad enough when we don’t know. It’s worse when you don’t think he knows, either. The best bet is to assume that the AMD execs are going to run the company by the seat of their pants, just more so than usual.
Now, more than ever, the best attitude to take is, “I can’t waste money/get blindsided if I don’t spend it.” If these companies can’t trust you with their “secrets,” then you ought not trust them with your money.
On the Intel side, the most prudent thing to do is pretend Prescott isn’t going to exist and that Tejas is the next big upgrade over what’s out there now (which probably will be true even if Intel fixes its problems), and decide based on that.
On the AMD side, hold out for at the very least a socket 939 board the maker says in writing will support second-generation Hammers. Sorry, but I’m not big on socket 754 for serious systems (unless they get dirt cheap, which I very much doubt over the next six-nine months). I suspect the performance gap between socket 754 and socket 939 systems is going to be rather bigger than it is today between 754 and 940. It would be wise to wait long enough to see whether I’m right or wrong on this before laying out big bucks for an Athlon64.
Stick with what you have, or do a minimal upgrade if you’re hurting, and if that means you won’t be able to upgrade again until 2005, that could well be a blessing in disguise.
Because 2004 is starting to look like the Year of FUD. Well, “fear” is stretching it.
Let’s call it the Year of UD instead.
Be the first to comment