Two Products In One
Intel ought to have two product introductions within the next three months. They’ll introduce Prescott in February, then they’ll introduce Prescott in April.
No, unfortunately, I’m not repeating myself.
We’re supposed to see a Prescott for socket 478 come within a month, then we’re supposed to see a socket T Prescott come roughly around April Fools’ Day.
For an overclocker, it is likely that they might as well be considered two different products.
Socket 478: Power-Capped?
Socket 478 Prescotts will initially get a lot of attention simply because people with Canterwood/Springdale motherboards will want to get a processor upgrade.
However, at least a major reason for the delay in Prescott is Intel’s apparent inability to get Prescotts to run within the power environment of those mobos. It only stands to reason that if Intel can’t get Prescotts to run fast in an 875/865 environments due to power restrictions, overclockers are hardly going to be able to get them to run faster by adding more power.
There are some indications that socket 478 Prescotts will only be sold in the 2.8-3.4GHz range, with anything above that being a socket T processor.
This is hardly promising for an overclocker.
So unless Intel has pulled off a miracle, the only people who ought to be interested in a socket 478 Prescott are those who already have Canterwood/Springdale boards, and maybe not even them.
Socket T: Electrically Sound, But?
No, if we’re going to break any serious new ground, that will almost most certainly require a socket T platform.
Until we get some tested results in, it’s probably not unreasonable to expect 4-4.2GHz out of these. That ought to be enough to equal or edge ahead of a 130nm FX, and beat an Athlon64.
Unfortunately, it doesn’t take very much improvement from a current O/Cd Northwood system to do that, and the total cost of ownership for a socket T Prescott system could be somewhat harrowing for the benefits you get from it.
If you want the best Intel has to offer next April, that means a new CPU, a new mobo, DDR-II RAM and a new PCI-Express video card. For practical purposes, that means the real cost of a socket T system depends on how much recent, good-enough stuff you’d have to replace from your current system.
You’ll probably be able to duck having to buy DDR-II, and maybe you’ll be able to duck replacing your AGP card, but even if you do both, we’re still talking about $350 for a system probably not much better than a current O/Cd Northwood system.
Another major issue with socket T Prescotts will be “How much power will they chew up?”
The real answer at this point is, “We don’t know enough about it yet” and if you think you do, you’re just wrong.
What we do know enough about at this point is that this is very likely to be an important issue, and could end up being a huge problem, especially for overclockers.
The item that we don’t know which we need to know is whether or not Prescott acts like other PIV processors when it comes to power/frequency. There’s been some speculation that due to changes made in making the CPUs, the power requirement of Prescott jump up much more with increased frequency than has been the case in the past.
To put it simply, if Intel’s spec sheets show that a 3.4GHz Prescott chews up 10 or 15 watts more power than a 3.2GHz, we overclocker have one big problem on our hands, much more so than if the difference is just five or six watts.
Another possibility is that the power curve on Prescotts starts at a much higher point than one would normally expect. For instance, historically, one would expect a 2.8GHz Prescott to chew up much less power (say a third less, including the extra cache) than a 2.8GHz Northwood. If that’s dramatically not the case, again, we’re going to have some real problems.
It could be both.
What we really don’t know is why Intel has had such problems, and the degree to which Intel has gotten a handle on those problems.
Put another way, if it takes 120-130 watts to overclock a socket T Prescott to 4.0-4.2GHz, that isn’t too bad. If it takes 160-180 watts, that’s bad.
We just won’t have the kind of information needed to start coming up with decent estimates until the socket T Prescott datasheets come out, probably late March.
Can’t We Just Wait Until 2005?
For those asking that question, in one word, yes.
Ask yourself how many hardware sites would say that publicly. 🙂
The problem with my job is that some people won’t take “yes” for an answer. 🙂
They tell me, “This is all well and good, Ed, but I need something now. What should I buy?”
I feel like saying, “Where were you six-nine months ago when I told you to buy?”
I don’t tell them that, but the problem is there just isn’t going to be any terribly good choices in 2004. Here’s what it boils down to:
Athlon64: Not cheap, doesn’t overclock much, hardly better than O/Cd Northwood systems, maybe 20% better than cheap XP systems, the first platform with a negative upgrade plan.
Socket 939 Too expensive, doesn’t overclock much, only somewhat better than O/Cd Northwood system, maybe 25% better than cheap XP systems.
We know this is going to be the case for most if not all of 2004.
So now we look at Intel and Prescott. For socket 478, it’s guilty until proven innocent, we’re as sure as we can be at this point that this is going to be bad news for overclockers.
So we look at socket T, and we say:
Socket T: Will probably be a little better than socket 939 and somewhat better than socket 754. Could be really expensive for some, and not all that better than current Northwood systems. Will probably have significant heat problems, could have terrible ones.
This is hardly a ringing endorsement. It’s more like “This may suck less than the other choices.” When your other options do suck, it is wise to wait to hear about the one that may not.
But then again, if the figurative gun isn’t pointed at your head, and the gunholder isn’t saying, “Choose,” there are certainly plenty of good reasons to say “No, thanks” to all three.
The Overclocker Perspective
This year is going to demonstrate the difference between an overclocker and a nonoverclocker.
The Athlon64 is a good example of this. The 130nm Athlon64 is a lousy overclocking chip. It is that way for a very simple, objective reason: it doesn’t overclock very much (which is hardly surprising given its 130nm manufacture and its high rated speed), and its future upgrade path is exceedingly poor.
Prescott looks like a better bet not because I hate AMD, not because I’m on Intel’s payroll, but simply due to the fact that it’s a 90nm processor which ought to make it a better processor than any 130nm processor.
It says a lot about Intel’s problems with this chip that this isn’t the no-brainer it would be under normal circumstances.
It says a lot about Intel’s problems that if I had to bet today, odds are at least 50-50 that even a socket T Prescott is going to join the ranks of the “not-quite-good-enough” along with the 130nm AMD products.
Prescott just represents the last glimmer of hope for overclockers until at least towards the end of the year, and probably 2005. By then, we’ll have 90nm Hammers and hopefully some more reasonable pricing from AMD, and Tejas from Intel.
That just a better environment to pick from than what we’re going to have this year.