Intel announced that it was cancelling its 4.0GHz Prescott.
It said it was doing so because it was now more interested in doing other things with CPU.
Ha. Ha. Ha.
An honest explanation would have been, “We’re going to stop killing ourselves trying to make this POS go any faster.”
This no doubt is a wise move. There’s no point in throwing resources at a brick wall, especially when it’s only going to get you a little speed bump in a cursed generation of processors.
But let no one think this was a voluntary move on Intel’s part. They’re doing it because they can’t deliver.
Misery Has Company
Is this a bad thing for Intel?
I’ll tell you what would make this a bad thing for Intel. It would be a pretty bad thing for Intel if they hit a wall its competitors managed to swerve around.
If AMD were able to crank out 90nm Hammers that could do PR equivalents of 5000+ MHz, that would be pretty bad news for Intel.
Fortunately for Intel, they can’t, despite the most fervent wishes of a few commentators out there.
When you’re in a horse race, falling in a ditch and being unable to get out is usually pretty bad, except when everyone else falls into the ditch with you.
That’s essentially the situation right now. It’s pretty silly crowing about Intel not being able to make a 4GHz processor at 90nm when your champion isn’t currently able to make and sell a 3800+ processor.
No, it isn’t due to “marketing” or some other cockamamie reason AMDroids dream up. The initial 90nm chips have been tested. Even overclocked, they don’t run much faster than their 130nm ancestors. There’s no hidden reservoir of performance in these chips.
Yes, AMD will probably be able to improve on that somewhat, but that’s about it.
Even if you assume the best for AMD, they’re only promising at most a 2.8GHz Hammer by the end of next year (and it may well not be even that, at least not at the A64 level; it may be 2.6 with 1Mb cache).
Yes, using SOI has helped AMD. Yes, it will give AMD a bit of a performance edge for a while.
However, that edge will be considerably less than that enjoyed by AMD during the Thunderbird level, and what happened to Intel during that time? AMD took 5% market share away from them.
5%. Not exactly world conquest numbers.
If AMD could only get 5% when their best processors had a big lead over Intel, what do you think is likely to happen when they get maybe half that lead?
The point to all this is not to knock AMD (or for that matter Intel or IBM). The point is to note that this is not a situation where the AMD horse just jumps the ditch and races to the finish line.
It’s more like AMD has fallen into the same ditch as Intel and IBM and may be able to crawl out of the ditch and creep a little further ahead with a couple broken legs and “win.”
They may well be good enough for you to go green in your next computer system, but it’s not going to make a blue world go green, just tint it a bit that way.
x86-64? Intel is already selling them. When and if they need to make more, they will. No competitive advantage in the desktop space.
There are those who say Intel’s version of x86-64 is inferior to AMD’s, but the only convincing evidence I’ve seen on this would only affect servers with more than 4Gb of RAM in them. Certainly a significant factor there, but that’s it.
Intel plans to increase cache sizes in lieu of increased processor speed. This will certainly mean less processors per wafer, and make their CPUs more expensive to make.
Is this any big deal? No, for the following reasons:
1) Intel makes billions, AMD doesn’t. Last quarter, the division of Intel that makes processors made 2,800 million dollars while making somewhere around 40 million processors, or roughly $70 a CPU.
AMD, on the other hand, made 89 million dollars selling 7-8 million processors last quarter, or about $12 a processor.
Clearly Intel has a much bigger cushion to play with when it comes to increased cost of processors.
2) More cache doesn’t cost a lot more cash It costs Intel about $25 to make a Northwood. This is quite a bit less than it used to cost Intel to make a processor; back in the PII era, cost per processor was estimated to be about $60.
Intel being Intel, they wanted to cut costs even more, so going to 90nm and using 300nm wafers probably was supposed to reduce the cost per CPU to maybe around half that $25 figure.
Doubling the cache on a Prescott will certainly increase the cost of one, but we’re talking about going from $12 back to $20-25, or no more than Northwoods cost.
This would be a big deal if CPUs sold for $29, but when they sell for an average $150 a pop, this is hardly a big deal. So Intel makes $2.3 billion rather than $2.7 billion. You might not like that as a stockholder, but it’s hardly going to break the company.
Yes, this calculation leaves out a very big factor, the cost of building fabs. If Intel has to build an extra fab to accomodate bigger chips, that’s not chump change.
However, when you make over $10 billion a year making CPUs, a few extra billion is not going to break you (and I doubt it will come to that; it’s not like Intel is converting their whole product line over to 2Mb cache chips).
AMD, on the other hand, doesn’t have that kind of cushion, but it’s important to understand the main reason why. The big factor isn’t whether or not an AMD CPU costs five dollars less to make. The big factor is that AMD CPUs sell for about five dozen dollars less than Intel’s.
So long as there’s this price gap, talk about cost per chip being a major competitive advantage is laughable.
It’s like saying Joe’s Diner has a cost advantage over Starbucks for a cup of coffee. It’s absolutely true, and absolutely irrelevant because Starbucks can get people to pay a lot more than Joe’s.
Even if Joe’s managed to get their prices up to Starbucks levels, that would be good for the owners of Joe’s, but it would hardly drive Starbucks into bankruptcy.
No, the only situation where this would matter would be if Starbucks suddenly couldn’t get any more money for their coffee than Joe’s. Then, and only then, does cost advantage begins to matter.
What Is Going To Happen?
Over the course of the next year, Intel will do whatever it takes to keep any performance competition close between it and AMD, and unless AMD exceeds its own expectations, Intel will succeed in doing this.
Initially, Intel will add cache to some models, which will keep things close.
Later, they’ll universally adopt x86-64 as needed, again, to keep it close with AMD.
That will be good enough to keep AMD market share gains to a few percentage points. Yes, that will be good for AMD, maybe they’ll make a couple hundred million a quarter for a few quarters.
AMD will likely do better percentage-wise in the server market, though again, Intel will do what it can to keep matters close. Opterons do look relatively better compared to Xeons, so AMD is likely to do better here than on the desktop, though the world won’t turn green here, either. Maybe they’ll get 20% marketshare in a year’s time.
In neither case are we going to see a breakthrough, where AMD becomes more like Pepsi to Intel’s Coke. AMD just doesn’t have the degree of advantage which would permit this to happen, certainly not within the next year or so.
Get Out of The Geek Ghetto
If you’re approaching, at, or beyond tirade mode, consider this:
We’re not at all trying to tell you not to buy a Hammer. We’re just telling you why the world isn’t going to do so.
It is a common human flaw to think that if you think a certain way, everyone thinks the same way.
I often get emails from people who get irritated at one thing or another, and instead of saying that “I am irritated at this,” they say, “We are all irritated at this.” More often than not, I end up informing them, “You’re the only person so far who has complained about this.”
I don’t doubt most of the geek community loves Hammers and Opterons, and will buy them. All we’re saying here is “You’re not the world, the wide, wide world out there doesn’t think this you, and there’s more of them than there are of you.”
Not that geeks have any monopoly on mental ghettoes. For instance, in the current U.S. presidential election, there is a chunk of New York (Manhattan specifically) called the Upper West Side, where you probably could launch bombing raids and level city blocks without touching a Bush supporter.
Many people living there will say that they don’t know a single person who’ll vote for W, and I wouldn’t doubt them. Senator Kerry will probably end up with 80%+ of their votes.
However, that doesn’t mean Kerry is going to end up with 90%, or even 51% of the total vote. That’s because this isn’t an election for the Presidency of the Upper West Side, it’s an election for the Presidency of the United States, and there are lots of places where people like W a lot better than they do on the Upper West Side.
Those Upper West Side may well consider Bush supporters idiots or worse, but that doesn’t matter one bit or invalidate a single Bush vote.
We have a similiar situation here. You may think AMD is the only possible choice, and by your criteria or even mine, you could well be right.
But there’s a big world out there that doesn’t look upon this the same way, and their votes count, too. Don’t confuse what you want to happen with what is likely to happen.