XBit Labs reports that 90nm Hammers are going to be furnaces, too.
The key item in the piece for those wanting to buy a Newcastle when they first come out is this:
“According to currently available details, maximum current (IDD) of future 64-bit processors from AMD will be 80A, while maximum thermal power will be 105W. These are mandatory requirements for mainboards in the second half of the year.”
What that means is that a socket 939 motherboard you may buy in April or May may not suitable for overclocking a 90nm Hammer in November or December.
It’s really the Prescott problem all over again, except this time, we know about it before people buy the mobos.
A Deeper Problem
There appears to be a deeper problem, though.
Neither Intel nor AMD seems to have been able to capture the normal wattage reduction they’ve gotten up to now from a process shrink, strained silicon, SOI, whatever else notwithstanding, and that appears to severely retard the amount of scaling from the process shrink.
We’ve spoken a number of times about Intel’s apparent inability to scale Prescott, but AMD isn’t making any big claims about its 90nm CPUs, either. So far, 2.8GHz is as far as they’ll say they’ll go.
They’ll probably end up with more than that in 2005, but it hardly looks like an aggressive ramp. If we use Opteron roadmaps as a guide, they imply that 3GHz looks to be the last gasp for Hammer before the K9 takes over.
Given that the end point of 130nm Hammers looks to be 2.4GHz, that’s not much of a ramp (though better than what Prescott apparently is going to do).
The point is not to go “nyah, nyah” at either company, but to point out that both companies are having real problems with 90nm technology compared to earlier generations. The usual isn’t happening.
And that means a slowdown in progress, no matter whom you like, and a slowdown means companies start doing things they never would have otherwise done.
For instance, a slowed-down silicon progression means that items like x86-64 become relatively more important. If Intel can’t just crank up the cycles at 90nm like it usually can, it just isn’t going to beat a 2.8 or 3GHz Hammer running x86-64.
So if you can’t beat them, join them. The increasing likelihood that Intel will at least announce something like x86-64 fairly soon probably has much to do with this slowdown.
(The accompanying comment that it will take Intel most of a year to get things ready is either a) an excuse to wait until Tejas to do this and/or b) Intel plans on having a version of x86-64 that is not compatible with AMD64. It’s probably both.)
It also might help to explain why both Intel and AMD are suddenly so interested in dual-core processors in 2005. I suspect that each of the cores is going to be a relatively slow one (at least compared to expectations).
What To Do
On the AMD side, you’ll probably want to wait until there are socket 939 mobos that are swear-to-God-and-cross-our-hearts compatible with 90nm Hammers. Given the relative early warning mobo folks are getting from AMD, this may mean no additional wait at all.
On the Intel side, there is even less reason than ever to want a Prescott or even early socket T. If Tejas ends up being the first Intel desktop 64-bit processor, you might as well plan at the moment to get the whole package meant to run it. Given that you’ll have to toss out just about everything except your drives to handle Tejas (including the case; BTX is a new ballgame), go Northwood now, and maybe add a Prescott in a year.