The 3% Wonder
I’m getting some emails which indicate some ignorance of what x86-64 entails.
From a circuitry standpoint, x86-64 is, and never has been a big deal.
All x86-64 processors have is a little extra circuitry to handle 64-bit instructions. This only adds 2-3% to die space. If you don’t believe me, let AMD tell you that (page 15).
That doesn’t mean it’s useless, quite the contrary. x86-64 circuitry gives a tremendous amount of bang for the silicon buck.
Taking out x86-64 from Hammer processors makes no sense from a cost-cutting or die-size perspective. Hammer’s memory controller chews up much more die space than that.
The only plausible reason to rip-out/deactivate x86-64 circuitry from the already greatly gelded mini-Hammer is not to save money making those, but to make more money on those processors that do have it.
The idea that people will be shaking in their boots running 32-bit applications on a 64-bit capable CPU is just silly, especially around here.
Trying to get a big premium (as in hundreds of dollars) for that extra 3-4 mm may prove to be just as silly.
This approach may make some sense at the very bottom of the totem pole, down around the Duron level, but otherwise, it looks like a very short-sighted racket.
AMD is not just trying to get a new processor line going. It is also trying to get x86-64 going as a new hardware standard. To do that, you need software, and to get people to port software over, you need a user base, as big a user base as you can get as quickly as possible.
It takes time to get software companies to be convinced of that, and more time for them to do it. That’s true even for Intel, which is as sure a bet as you get in this business. You didn’t see tons of SSE2 software around the day the PIV came out; it came over time.
Companies are going to need a lot more convincing to go x86-64, and the best way to do that is to show those companies customers. Lots of them.
You don’t do that by making x86-64 an elite chip. You do that, you don’t have a mainstream processor, you have an Alpha. True, you’ll get a few workstation-type programs converted quickly, but how much else?
Stretch, Don’t Snap
AMD makes too little money from its CPUs right now. They need to make more. I’ve said that several times in the past.
But it is foolish to go from one extreme to another. You try to get $100 from the person who was laying out $50, or $150 from the person previously spending $100. Trying to get that kind of premium is OK.
Trying to get $300 or $400 or $650 is not. Especially when the new chips don’t do much better than the $50-100 ones, and especially when the x86-64 Windows software that is the only thing that will make these CPUs shine is scarce.
AMD seems to think people are foaming at the mouth for x86-64, and there are a few that are do, and are willing to lay hundreds more out for x86-64. But how many? Would you?
Maybe this works for a quarter. Then what?
The Old Celeron Scenario
Here’s what may be an underlying fear in AMD corporate headquarters:
What’s the performance difference between a little Athlon64 and a big Opteron-like one?
Assuming the “little one” has 256K cache, the performance difference between 256K and 1Mb cache is about 10-12%. If the little one has 512K cache instead (as rumored as of late), the difference shrinks to about 5%.
Single vs. dual channel memory? Figure about another 7%. Do you see why AMD got interested in dual-channel systems all of a sudden?
So depending on the amount of cache the “little” Hammer has, the performance difference on average will range anywhere from a bit over 10% to a bit less than 20% from its “big brother.” That’s not much given that the cost differential between the two platforms will be hundreds of percent.
The only other way to widen the gap anymore is to get rid of x86-64 in the “little” Hammer and make the “big” Hammer an elite expensive chip.
That will widen the gap more than enough, but it flushes x86-64 as a mainstream OS down the drain.
What AMD Should Do
AMD really has two tasks. They need to get a less-than-optimal first generation of a platform off to a good start, and to do that, they need a good start to a mainstream x86-64 user base, so software writers will write sooner rather than later. That will mean a bigger base of x86-64 software, which will only make Hammer better as time goes by, and greatly encouraged purchases of second-generation Hammers.
x86-64 is AMD’s only chance to get any real performance edge on Intel. It’s not just the CPU; the software is even more important.
I don’t fear what AMD charges initially for the product; what I fear is what they may keep charging for the product next year.
Nor is this, like some think, a tantrum of “I want a cheap Hammer, and NOW.” Personally, not only am I willing to wait; I’m more than willing to go without. I’m not hurting. I don’t see any big advantage from the first generation that can’t wait until the second. It’s going to be pretty hard to persuade me to buy the first generation at a low price.
And most of you feel the same way.
What I fear is that early next year, after AMD actually starts cranking out significant numbers of these processors, they’re going to try to keep sky-high prices, and there’s going to be a lot of high-priced CPUs gathering dust.
Should that happen, and should Hammer get dismissed as a flop, it’s going to be pretty hard for AMD not to go down the tubes.
Once AMD can finally make a lot of Hammers (presumably early next year), what they ought to do is put out some slower processors and make the low-end “little one” around $100, and the low-end “big one” about $200 or a bit more. They should all be x86-64 enabled.
That ought to be enough to get enough people to buy these processors and establish an initial user base to encourage the average software writer to jump on the x86-64 until
What AMD Will Probably Do Instead
AMD will introduce the processors, and it will seem like about twelve people will actually pay for them in this neck of the woods. Actually, those who have a real, pressing reason to do so will buy them, and that could account for a few hundred thousand sales and give AMD a bit of a blip-up for the last quarter.
Encouraged by this, AMD will keep prices pretty much up, and after the early adopters have adopted; it’s crash and burn time. Supply will far exceed demand, and then it will become a question of when AMD will do another policy back-flip and cut prices/introduce lower-speed cheaper processors.
Unfortunately, if it takes them a quarter or so to realize that, by that point second-generation processors and more importantly second-generation motherboards will be imminent, so buying at that point will be buying into near-instant obsolesence.
The longer they take, the more damage they’ll do to their long-run prospects. These CPUs need x86-64 software to get an edge over Intel, but mainstream software won’t get developed for workstation-priced systems. So when the second-generation comes along, and AMD says, “OK, now we need x86-64 games and software,” it won’t be there, and AMD will have flushed a much-needed advantage next summer down the drain in return for a relatively few quicky sales this fall.
Can they recover from that? Maybe, maybe not.
The biggest problem for AMD is that for whatever reason, they don’t think strategically. Far too many of their decisions seem forced or last-minute; they think about today’s problem and situation, and we’ll worry about tomorrow tomorrow.
In the case of x86-64, it seems to be a matter of “let’s milk an extra quart from the cow today,” and risking the loss of gallons later on.