Xbit Labs reports that due to problems Intel is having with strained silicon, we are likely to see just a paper launch of Prescott in December, with no real availability until 2004.
This is not terribly surprising given the various indicators (or mostly, lack of such) over the past few months.
However, up to now, we haven’t known if this is a small problem, or a big one.
A small problem would have been if Intel could get the processors out, but they couldn’t overclock very well. Bad for us, but not too big a deal for Intel.
A big problem would be if Intel can’t make enough processors that cut it even at default ranges for anybody. That would affect Intel far more.
This news item would seem to say that Intel has the latter problem.
Another indicator which tends to back this up is the emergence of engineering sample Prescotts here and there. They have not been 3.4GHz samples. Rather, they’ve been 2.8GHz samples.
One could read too much into this, but this might mean that Intel has the same problem with Prescott than AMD had with Hammer; they can make them go, but they can’t make (enough of) them go fast.
What Does This Mean For AMD?
Well, it’s hardly bad news, but it’s probably not as good news as you might think, for the moment. The reason for that is AMD isn’t going to be making very many Hammers anyway for the rest of the year.
It probably does help to make sure AMD will end up selling the Hammers they do make at the price they want the rest of the year. That could be enough to give AMD a breakeven/slightly profitable quarter the last quarter of the year.
But then again, they’d probably would have done that anyway.
It’s doubtful that a lot of would-be Intel buyers (especially at the corporate level) would at least initially shift over to AMD due to this delay. They’d either buy a Northwood, buy an Extreme Edition, or just wait a bit.
No, this is one of those problems that will take a while to have an effect, and even if it does, it will be limited by a number of factors.
First, AMD can’t significantly outclass Intel until they start making 90nm processors and lots of them. That won’t be for some time to come.
Second, AMD itself won’t start making Hammer in serious quantities until the second quarter of 2004, and even then, we’re only talking about 15% of world CPU production.
Third, one ought to keep in mind what happened the last time AMD had an edge on Intel (after the 1.13 PIII fiasco). The edge was bigger and lasted a much longer time than is likely here. What impact did it have?
AMD’s share of world CPU production went from about 18% to 23% of production, and they had a number of fairly profitable quarters.
Now AMD could certainly use a repeat of that, and there’s nothing wrong with that, but it won’t be worldshattering for the computer market as a whole.
That’s the difference between Prescott delays and Hammer delays. Yes, both companies were or are not being very honest about their problems. But Intel’s existence isn’t threatened by Prescott delays the way AMD’s was with Hammer’s. Intel could even skip Prescott and go to Tejas a year from now and not suffer too much damage (indeed, if you’re a PIV owner, you ought to be thinking along these lines already).
However, hard as it might be to believe, AMD could squander this gift, and here’s how.
A Possible Intel Trap
Let’s assume things are really bad over at the Intel fab plant, and they can’t get this fixed until next June or so.
Intel can make Extreme Edition their flagship chip rather than Prescott, and keep the rest of their late-October pricing plan in place. The next price cut isn’t due until February, maybe they squeeze a 3.4 out of Northwood, maybe they don’t.
If they don’t, that sets a possible trap for AMD. If AMD keeps setting its prices by Intel standards; they’re going to find themselves making millions of processors with a $400+ price point.
Sounds good until you realize there aren’t all that many buyers willing to pay that kind of money for any CPU, much less one from AMD.
In the past, Intel has trapped AMD on the low side when they have had the performance advantage by strategically pricing their chips minimize AMD’s revenues.
You can play that game another way, too. Intel can actually trap AMD on the high side, too, by grabbing the middle price range and leaving AMD the relatively sparse high-end.
I realize that sounds bizarre, and probably sounds like a far-fetched excuse to many. It only works if all of the following happens:
a) AMD goes into mass production.
b) AMD keeps its pricing sky-high, and doesn’t introduce moderately-priced Hammers into the picture one way or the other (either by lowering prices or introducing lower-speed models).
c) AMD ends up making far more sky-high priced CPUs than the market is willing to buy, and is slow to respond to lack of sales.
The trap only works if AMD execs say, “Well, if we get $400 a piece for 500,000 processors, we can get the same for 5,000,000 processors” and there’s only a million people out there willing to pay that kind of money.
If Rolls Royce made ten times as many Rolls Royces, and didn’t change the price, you’d have a lot of unsold Rolls Royces out there. That’s the danger.
Obviously, this would require considerable cooperation on AMD’s part. They could easily break out of it.
The question is not “Can they break out of it?” but “Will they break out of it?” Will they get so enraptured from getting a lot of money for a relative handful of desktop chips that they’ll think it can go on forever?
Do I think AMD can be that stupid? Yes, I think AMD can be that stupid. Well, stupid may not be the right word for it.
If you’ve been starving to death, and suddenly you getting ten times the money for a CPU that you got before, it’s hard to walk away from it voluntarily. But that’s what AMD is going to have to do some months down the road.
Given their track record, what’s more likely is that they’ll keep prices high until they have a ton of unsold processors, then abruptly change course.
One Item We Don’t Know
We know how many Hammers AMD expects to make over the next nine months. What we don’t know is what kind of Hammers they’re going to be.
The vast majority of Hammers AMD will make will be socket 754 CPUs, and what we don’t know is how much cache is going to come with how many of those processors.
In all likelihood, there will be “heavy” and “light” socket 754 Hammers. “Heavy” Hammers will have 1Mb of cache. “Light” Hammers will have 256K. According to the AMD tech docs, there can even be a “medium” Hammer with 512K cache.
We suspect “light” Hammers won’t perform terribly well compared to “heavies;” it won’t be a Celeron/PIV situation, but the difference will be significant. We suspect that “mediums” would also trail, but not by nearly as much.
It may well be that AMD plans on maximizing revenues by making relatively few “heavies” at continued high prices when they crank up, and a lot of “lights” at lower prices. Then the question becomes “Will the “lights” sell, and at what price point?”
This could be very tricky for AMD. You’ll have two kinds of “heavy” Hammers, one or maybe two versions of Hammer Lite, Bartons and TBred/Thortons. I guess AMD felt it wasn’t getting enough competition from Intel. 🙂
Getting the pricing and production mix on this correctly is going to be very difficult, especially given the relatively narrow differences in performance compared to the likely differences in price.
I don’t even want to add the prospects of 90nm processors and when they might show up in the mix, my head hurts enough looking at this as is. Doesn’t yours?
Intel’s having problems. They are embarrasssing, and the longer they last, the more the embarrassment will turn into pain. But unless this lasts several years, it’s hardly fatal or even especially health-threatening.
What is a minor loss for Intel can become a big gain for AMD. Not big as in toppling Intel, but big as in getting back to the Thunderbird days. However, it’s not a given, and AMD could forfeit much or most of its opportunity due to its unfocused confused-at-best, non-existent-at-worst strategy during this period.
For you, Prescott delays probably means CPUs will cost more early next year than they would have otherwise. This makes an even stronger case to update now with cheap, available products, and sit 2004 out.