The Inquirer has an article about the revised power requirements for boards that are supposed to support Prescott.
For those of you without current 865/875 mobos who are wondering, “What’s the big deal about this?” perhaps a few graphs will help illustrate the problem.
Here’s a graph showing the Thermal Design Power of Willamette and early Northwood processors:
This shows a power consumption pattern historically typical of Intel.
Power consumption scales upward fairly consistently until you get to the tail end, where it spikes up a bit. A die shrink drops power consumption quite a bit, and then the upward progression resumes.
Let’s see what happens when we substitute some later Northwoods, and extend the scaling to 3GHz.
You can see the power curve moves upward a bit from 2.53 to 2.8, much like it does going from 1.8 to 2.0, and then jumps a lot going from 2.8 to 3.06.
It takes about 20% more wattage to go from 2.8 to 3.06, which is less than 10% more speed.
How do the hyperthreaded CPUs fare?
How Do The HT Chips Do?
The HT chips chews up a bit more wattage than non-HT chips, but that’s to be expected, and the power increase is along the lines of 5%. Here’s the TDF for them.
We see another, similiar jump, though the TDFs for the HT chips are a good deal more erratic than for the other PIV. The 2.8 chews up hardly any more power than the 2.6 (and the 3.2 only a smidgen more than the 3).
Anomolies aside, there does seem to be a rather unusual boost in power required once you get past the 2.8GHz point.
In and of itself, that doesn’t necessarily mean much. After all, the 2GHz Willamette needs significantly more power, too. This could just mean Intel pushed the high-end Northwoods a bit more than the high-end Willamettes.
But then we have this graph:
The purple line shows the TDF for the 2GHz Willamette and the 3GHz Northwood, then takes the original TDF for the 865/875 and assumes that this would be what it would take to run a 4GHz Prescott. Looks like a nice smooth line, doesn’t it?
The red line shows the same TDF for the Willamette and Northwood, and takes the revised TDF for the upcoming 865/875 boards. Big difference, isn’t there?
In all likelihood, the projected 4GHz figures for both purple and yellow line are a bit too high, since Intel probably wants to allow for some leeway. However, the gap between the two figures is probably pretty accurate, and quite significant.
Keep something in mind, the wattage figures jump up more even though the percentage ramp decreases quite a bit. Going from a 2GHz Willamette to a 3GHz Northwood is a 50% rampup in frequency. Going from 3GHz to 4GHz is only a 33% rampup in frequency.
In short, you’ll need a good deal more power for a good deal less speed increase.
One Last Piece of the Puzzle Wanted
We have a rough idea how much power a fast (4GHz) Prescott will require. What we don’t know yet is how much power a slow one (say a 2.8 or 3.0) will chew up.
Why is that important? That will give us a big clue on how hard these chips will be to cool when overclocked.
We know a 4GHz Prescott will chew up proportionately more power than previous generations. Will that be true of all Prescotts, or just the faster ones?
Some examples to illustrate the point:
Example 1: A die shrink usually gets you about a 30% power reduction from a previous generation’s chip running at the same speed. If you use that rough rule of thumb for the 2.8 and 3.0 Prescotts, you get wattages of about 49 and 56 watts.
That looks pretty good until you realize that a 4GHz is going to chew up almost 100 watts. That will mean ever 200 extra MHz will chew up an extra 7-8 watts rather than Assume an attempted overclock to 4.5GHz with a modest 10% increase in voltage, and even if you’re optimistic, it’s hard to see how the thing isn’t going to chew up at least 130 watts. Be somewhat pessimistic or more adventurous with the voltage, and 150 watts plus is quite possible.
Given that Prescott will be not too much over 100 sq. mm, we’re talking about twice the wattage per sq. mm that the cool-by-comparison Thunderbirds used to chew up.
On top of that, trying to get 130-150 or more watts into the CPU could quite possibly make voltage regulators do new, exciting things like melt a few layers deeper into the motherboard. SPMDS (Sudden Prescott Motherboard Death Syndrome) could become a new acronym; cooling of the voltage regulators will probably become essential for the non-suicidal overclocker.
This is not what the average “buy a retail CPU and get a big overclock by changing a setting” Intel overclocker wants to hear.
Example 2: Let’s presume a die shrink does relatively little good. Let’s say the 3.0 Prescott chews up 70 or 75 watts rather than 56. That won’t be all too good for regular Intel customers, but since the scaling upward for frequency increase will be flatter, overclockers will find their total wattage to be somewhat less than the scary figures mentioned above.
It still won’t be easy, though.