There’s two roadmaps over at the Inquirer showing what Intel plans to do over the next year.
The pricing roadmap shows the EE standing in for the Prescott the rest of this year (no surprise there). Prescott looks likely to debut early next year (mid-January is the usual time Intel historically makes its first product announcement for the year), with a big price cut following on February 15.
From the perspective of availability, this roadmap ought to be viewed with some skepticism. It basically says, “No Prescotts for another three months, then bingo, bango, we’ll have a flood of them.” It really looks more like, “We hope we can get this fixed in time.”
If you read the tea leaves perhaps a bit too closely, the impression one gets is that Intel isn’t having problems making Prescotts, but rather getting them to run fast.
The couple engineering samples that have popped up here and there are for 2.8GHz Prescotts, which is supposed to the slowest Prescott to be released. When Intel releases engineering samples for validation purposes, they normally don’t do that. Rather, the engineering samples are released at high speeds (and are unlocked so engineers can change the multiplier to validate at lower speeds.)
There’s also been a potential sighting of the yield game. What’s the “yield game?”
The yield game is a semantic exercise in which one tries to confuse overall yield with desirable yield.
For instance, if Intel makes a Northwood wafer, and all the processors that work are capable of 3.2GHz, that’s not only a good yield, it’s a desirable one, since Intel can sell them anyway they like.
On the other hand, if Intel makes a Prescott wafer, and the same percentage of processors work, but only 2% work at 3.4GHz, it’s a good yield, but not a desirable one, since Intel wants most of these processors to work at that speed or better.
AMD had the same problem with Hammers. They could get them to work; they just couldn’t get them to run fast until about the middle of this year.
The second roadmap tends to reinforce this belief. It shows Intel not reaching 4GHz until the end of 2004, when Tejas should show up. This shrinks even further Intel’s rather modest plans for Prescott.
Consider this: From June 2004 to perhaps October 2004, the flagship Intel chip will increase only about 20% in speed, even with a process shrink. That’s unheard of.
This isn’t terribly good news for the average computer buyers, but it’s death to overclockers. I think that 20% figure is a good ballpark figure on what overclockers might expect from Prescott over Northwood.
One Cripple Or Two?
Under normal circumstance, this would all be good news to AMD. However, there’s good reason to doubt that AMD’s process technology is all that healthy, either, given the miniscule number of Athlon64s they plan to make the rest of the year.
There are other explanations for the low numbers, though. AMD may well not be able to make a lot of Athlons and shift to 90nm process technology AND make a lot of Hammers at the same time.
If that’s the case, then the question becomes, “How soon and how well can AMD make 90nm Hammers?” If they can, they ought to get an edge with their 90nm chips at least until Tejas even if Intel fixes Prescott enough to get it out the door.
If they can’t, then we’re likely to find two aemi-incapacitated CPU companies waving their crutches at each other.
This really isn’t a knock on either company. Making CPUs faster is becoming harder and harder, and it’s understandable that there’s going to be problems when new, unproven technologies have to start being used to keep progressing.
What it does mean, though, is that unless second-generation Hammers are extremely robust, it’s likely we’ll have to scale back expectations for some time to come, and not assume the level of progress we’ve gotten used to.