On the CPU front, the future is dual.
Intel laid out its medium-term roadmap the other day and said enough to indicate that multicore technology will become a reality for folks like us in 2006.
They also made it clear that the idea is to make two relatively slower processors do more work than one relatively fast one.
In the nearer term, Prescott will be it for quite some time to come. Northwoods will be phased out towards the end of the year, and Prescott will go to a 65nm shrink.
Prescotts with x86-64 circuitry turned on ought to become available in August on the upper-end, at prices no higher than other Prescotts.
At the end of the year, we’ll see 2MB cache Prescotts running at 1066MHz FSB.
Perhaps more importantly in the long run, Intel has tossed the Prescott and the Dothan people into a virtual sandbox and told them to play together.
Initially, we’re probably just looking at tweaks here and there, but we may well see Prescott becoming rather more Dothan-like as time progresses, and the Dothan design morphing in the background to become part of the first Intel desktop multicores.
(Incidentally, a desktop Dothan would hold up pretty decently even today against today’s best processors, though probably not tomorrow’s. Intel doesn’t plan very much scaling of Dothan in the year ahead, which is usually a good indicator that there isn’t a whole lot of scalability left in the chip.
Will the Ugly Duckling Turn Into A Swan?
Don’t count on it. All the Prescott changes look more like a complete makeover to make Prescott presentable enough to keep people from turning to AMD in the short term.
That may be good enough for folks buying Dell boxes, but for overclockers, Prescott has two huge problems: heat and lack of scalability because of the heat. Until that changes dramatically, all these changes are like dunking yourself in perfume rather than bathing. It’s just a cover-up.
Beyond that, all this talk about multi-core in 2005 is again meant to keep people from straying. Intel may well get out some multicores in 2005 in 90nm, but from both a manufacturing and software perspective, it’s not going to be practical to make multicores in the millions and millions until 65nm and Longhorn.
AMD had actually planned on doing much the same thing as Intel when it comes to multicores. The only differences will be that multicore 90nm chips will be a bit more practical for them due to SOI, which is a good thing since they’ll get to 65nm later than Intel.
In the shorter run, the unanswered question for us is “How far can 90nm Hammers scale? Will AMD have the same sort of scaling problems IBM apparently is having with its 90nm SOI processors?”
AMD hasn’t promised a whole lot out of 90nm, but then they don’t really need to to beat a prettied-up Prescott. If they can execute their roadmap and deliver what they’ve promised more or less on time without Hammers turning into a furnace, that ought to be enough to make it the overclockers’ default choice for 2005.
Do I have doubts about that? Yes. I don’t doubt AMD can make 90nm Hammers now; I have doubts about them being able to make fast ones quite yet.
This is partly due to IBM’s apparent problems, partly due to the reluctance of AMD execs to talk them up in their analyst conferences. Put very simply, when AMD has a hot product in the bank, they hype it in these conferences. When they don’t talk about it, and say as little as possible when asked, they’re having problems.
There will be another analyst conference call in mid-July. If things are going well, Hector and Company ought to be bursting at the seams wanting to tell the world about it. If they’re not, they’ll be like a defendant being interrogated with his lawyer present.
We shall see.
One thing the Prescott prettying-up does do is take away AMD x86-64 advantage. If Intel isn’t going to charge any more for it, it’s going to be hard for AMD to try to.
This doesn’t mean AMD has to drop its prices dramatically, but they’d better not treat their 90nm processors like they were gold, because if they do, Intel will scarf up the gold from the average computer buyer with lower-priced 90nm, x86-64-enabled Prescotts. when
When 90nm comes out, AMD needs to have a range of 90nm Hammers that match each and every enhanced Prescott out there. So long as they do that, they’ll still have a price advantage in the OEM market over Intel, since the rest of the LGA 775 platforms will be considerably more expensive with pricier DDR2, and heftier, more expensive infrastructure to handle Prescott heat.
But if they insist on not selling a 90nm Hammer for less than $400; they might as well be working for Intel for the effect that will have in the overall computer market.
As Mike Magee, another long-time observer of the compnay, put it the other day:
“Let us never underestimate AMD’s ability to shoot itself in the foot.”