There seem to be a bunch of shadowboxing taking place between Intel and AMD.
If you look around, you’ll find resellers saying they’ll have Athlon 64 3400+ Real Soon Now. You’ll find other resellers saying they’ll have 3.4GHz PIVs out one of these days, too.
No doubt some people will make a fuss about it, but to an overclocker, this is just yawn-inducing.
Overclockers are not interested in rating changes by a CPU company. Rather, they are interested in capacity changes and price changes.
If Intel puts out a 3.4GHz Northie or AMD puts out a 2.2GHz 64, well, this audience did that months ago. Big deal.
Looking For Inner Change
What is important to an overclocker is not whether the rating on the chip has officially changed, but rather whether the chip itself has changed, a new stepping, a new process shrink.
So far, there’s been no indication that something like this has occurred. That doesn’t mean there hasn’t been one, but you have to look into the company’s technical documents to find that out.
AMD generally changes its document to reflect such changes at the time of product introduction. Intel sometimes gives a little advance warning in theirs in some of their regular updates in their stable of products. Within the next two weeks, we ought to know if these two processors represent anything actually new.
Even when that happens, though, the overclocker isn’t terribly concerned about a new stepping for the top-end CPU. What he wants to know is how he can identify the new version so he can get the same product in a lower-rated and less expensive version.
In all likelihood, we’ll probably see 130nm Hammers get at least one new stepping over the next six months (though that doesn’t mean we’ll see one very shortly).
What Intel will do is much iffier. Right now, they’re trying to squeeze out every little last bit out of the Northwoods, and when you get to that point, there’s usually very little advantage left for overclockers.
So a new AMD Hammer stepping is likely to matter, while a new Intel stepping probably won’t.
Not All Steppings Are Created Equal
This brings us to a very important point about the benefits of steppings. Not all steppings are created equal. AMD is nearing the end of the 130nm process generation, and Intel is (for practical purposes) already there. You can’t expect huge jumps from either until they go to a 90nm process shrink. AMD probably can squeeze another couple MHz out of this generation of Hammer (which they’ll probably do by the time socket 939 comes out), with Intel probably not doing any more than we overclockers already have done.
On the Intel side, the overclocker isn’t interested in a heavily pushed Northwood. He would normally be interested in a low-end Prescott, simply because process shrinks have given a big boost to the current level of performance.
The reason why people are concerned about Prescott is that all outward signes indicate that what usually happens isn’t happening with these chips.
On the AMD side, under normal circumstances, the stepping wouldn’t be all that important, but because it will probably come at the time when desktop Hammers become reasonably priced, and because 90nm Hammers probably won’t become reasonably priced until early 2005; it takes on more importance.
Taking A Bank Shot
That doesn’t mean an overclocker doesn’t look forward to a higher-rated CPU come out. He does, but he does for an entirely different reason than many. A new product usually means it takes the top price on the pricing totem pole. That an OCer couldn’t care less about. A new CPU on top usually knocks down the prices of the most of the older CPUs on that totem pole (except towards the bottom). That is what an overclocker cares about, and that’s how a new chip actually affects him. Very indirectly, like a two-carom bank shot in pool.
So when he sees a 3400+ for $440 sale at Zipzoomfly; he doesn’t really see that price tag in his mind’s eye. Instead, he sees a $270 price tag on a 3200+ coming soon, or perhaps hopes for an even lower price tag on a 3000+ real soon (he’ll get the first, but probably won’t get the second).
An Intel overclocker isn’t really interested in the $450-500 price tag of a 3.4GHz Northwood, either. He’s rather more interested in the prospect of a $175 2.8GHz Prescott next month.
The Real Dates To Watch
Over the next three months, some introductions will happen which will be important. It’s just that 3400/3.4GHz won’t be one of them.
In early February, we’re supposed to see Prescott finally come out, followed shortly thereafter by a price cut. Will it come out officially, or will Intel postpone, again? Will it REALLY come out as in “give me money, and you can have one?” How well will it overclock? Will it be the overclocking dog we suspect it will be, or will it be able to keep up or exceed overclocked Newcastles?
OK, most people reading this won’t care too much about that (unless Prescott does rather better than expected). The AMD overclockers will be quite interested, though, in any AMD price cuts triggered by the Intel price cuts.
At the end of March, we’ll see the Hammer socket 939 platform unveiled and processors for it, including Newcastles. I would also bet that if Intel plans on committing to some form of x86-64 in the future, we’ll hear about that at about the same time so as to rain on AMD’s parade.
Those are the real dates to watch.
A Few Words On Prescott Rumors
There’s some rumors about that Prescott is undergoing some major revisions.
It’s very hard to comment on the likelihood of Intel doing such things when you don’t really have more than a vague notion of the problems Intel is facing.
This isn’t to say they’re false, but that doesn’t mean that they’re true, either.
On the one hand, if heat truly is a killer problem (as opposed to problems by earlier motherboards handling more power), these are plausible ways Intel could handle the problem.
On the other hand, these steps are rather large, fairly desperate means of handling them, measures generally reserved to major revisions of a CPU line, which Prescott isn’t supposed to be.
It would be rather more believable if these changes were included in Tejas (but then again, it’s possible Intel is cannibalizing Tejas to get Prescott out the door).
As someone who’s done a bit of speculating in his time :), I do have to say that I’d like a little more to hang my hat on.
However, even if these things are true, that’s not necessarily bad. For instance, some people have a bad habit of obsessing about the penalties of branch misprediction in CPUs with long pipelines. They’re hardly good for you, but with that negative from an extended pipeline comes a plus: the ability to ramp speeds higher.
CPU design is a series of trade-offs. You try to make the tradeoffs that give you the best net plus, and there is more than one way to measure a “best net plus.”
Right now, we don’t have the foggiest notion of what Intel is really doing, and for practical purposes, we don’t care. We’re concerned about what a CPU does, not how it does it.
Overclockers have an even broader perspective on the matter. If Intel makes changes that drop baseline performance 5%, but increases overclockability 20%, I’ll take that tradeoff any day.
The point is not to say that any changes Intel may or may not be making are bound to be good, or bound to be bad. It’s just to point out that CPUs are like a forest, and finding a dead tree there doesn’t mean the forest is necessarily doomed.
We just don’t know.