Joe: If this is the case, 3.6 GHz overclock would suggest 2.4 GHz top speed. Anyway, I am just speculating here. I would welcome input here.
You basically can't use LN cooling experiments for squat.
Cooling the chip that much dramatically changes many of the physical properties of the chip.
One obvious reason why cooling is so important when overclocking a chip is that transistor speed is temperature dependent. With a 2% decrease in CMOS transistor switching speed for every 10°C of temperature increase, the effect is not trivial (the numbers ripped from a Test & Measurement World brochure I had lying around).
This effect is actually pretty simple and one could possibly compensate for it when comparing the overclocking results at LN-temperatures.
However, there are many much (much) more complicated effects.
One such effect that could be important in the case of P4 is the incredible increase in thermal conductivity¹. The effect could be that hot spots that would otherwise present a hard limit for the chip would no longer play the same part.
Resistivity is another. At "small" temperature changes, the resistivity of a bulk metal varies linearly with temperature, i.e. a decrease in temperature lowers the resistivity. While the Al "wiring" in a P4 chip may not behave as the bulk metal, the effect would almost certainly still be there (and, quite possibly, be more significant).
Even more complicated, carrier mobility in a semiconducter is also temperature dependent. Along with the temperature effects for metals, this could result in a reduced clock skew at lower temperatures (although I certainly don't know whether this is a significant effect at all).
In other words: If you could be sure that the speed bottle necks in your processors were the same, you might be able to attain relevant data from LN overclocking. However, when comparing two chips as different as an Athlon and a P4 (on two vastly different processes, btw!), there's just no way to extract useful information from LN overclocking results.
-fyo
¹ isonics.com |
