Hammer Gets Hammered

AMD has imagined the following for the last few years:

Intel continues to blunder along with their non-x86 64-bit Itaniums and McKinleys which cost (or will cost) an arm and a leg and need (for practical purposes) all new software. Along comes AMD like Mighty Mouse with
a relatively cheap extended-x86 64-bit Hammer which can run current software.

Everybody outside of a few geek purists says to hell with Intel, and AMD lives happily ever after.

It wasn’t called “Hammer” because Jerry Sanders really wants to be Bob Vela; it’s supposed to hammer Intel.

AMD strongly believes that, enough to pretty much bet the company on it in 2003 and 2004.

As I’ve mentioned before, there’s much to be said for that approach. Intel’s 64-bit processors have turned into monstrosities hardly suitable for conversion to the desktop.

There will be those who’ll say Intel’s 64-bit approach is “better,” and so long as you’re not paying for it, it certainly is.

In the real world, though, if you have to choose between a $5,000 processor with additional software costs and a $250 processor that is 80% as good without additional software costs, guess what well over 90% of people are going to choose?

In the real world, good enough and cheap beats the expensive best every time.

What The Hell Is A Yamhill?

There appears to be at least a few folks at Intel who have a least a few top-notch brain cells functioning. They see this could mean trouble. So they’ve come up with Plan B (the Inquirer first came up with the story), and it is called Yamhill.

Yamhill is essentially an implementation of Hammer design into the next generation of PIVs (the 0.10 micron ones) that are supposed to be out the latter part of 2003.

Hammer-compatible X86-64 instructions will be built into the chips, and Intel will decide down the road whether to turn it on or not. If Hammer does really well, they turn it on. If it doesn’t, they don’t.

(Well, something tells me that feature is going to get turned on whether Intel wants to or not.)

Two Markets

Intel began designing these 64-bit chips to get into the more profitable heavy-end server industry, and that’s what these chips look like. They have everything and the kitchen sink, with a price tag to match.

Is there a market for such chips? Probably, but it’s not very big compared to the desktop market, which is far more attuned to cost.

Intel would no doubt much prefer to have one basic CPU design, and getting rid of x86 means getting AMD off its back.

But McKinley is just not a desktop chip, and it just doesn’t offer the average person real benefits. At the least, you have to give people a good reason to toss away their x86 investment, and this doesn’t.

The Hammers are light, lean and mean. McKinley’s are just the opposite. In a time when increasing power on the desktop becomes more and more debatable, a CPU like this is a processor without a cause on the desktop.

McKinley (or any foreseeable descendant) will also get bit on the rear by the MHz myth. It has an 8-stage pipeline. This means forget about big GHz numbers. It will probably require .10 micron just to get McKinley’s GHz numbers up to around current PIV, much less what we’ll see, say, two years from now.

You try marketing a 2.5GHz Son of McKinley vs. a 5GHz Clawhammer in 2004 to Joe SixPack.

I think Intel is going to be stuck with two processor lines for a while. They’ll probably be led kicking, dragging and screaming to turn on x86-64, but I think they’ll have to.

Whither AMD?

This is pretty bad news for AMD, easily the worst it has had the last couple years. Just by the time Hammer will start to get going, Intel will have the ability to match its major selling point any time it feels like it. When it’s even, AMD doesn’t win.

Hammer will still retain some advantages over a converted PIV, but they will be relatively no more than they have right now. That will mean AMD won’t be playing David. They’ll probably hold their own, but that’s about it.

AMD has to pray that Intel has another Rambus-like fit of multi-year extended stupidity which lasts so long that they become Pepsi to Intel’s Coke in the meantime. It could happen, but I don’t think Intel is going to be quite that stupid. They’ll get to the x86-64 party late, but not too late.

Email Ed

Be the first to comment

Leave a Reply