There is some info here:
http://www.qdi.nl/support/CPUQDISocketA.htm
On any processor parameter, there will be variation from unit to unit.
In fact, on the production line, some of the processors are "harvested"
for mobile use, due to their lower dissipation and ability to run
with less Vcore and still pass all test vectors.
If your previous processor was on the "good side" of the distribution
and the new processor is on the "bad side" of the distribution, then
the new dissipation might be higher. If the die area of the processor
is different, then if you are using the same heatsink, the thermal
resistance to the processor will be higher, if the new processor
has a smaller die.
You should also check the measured Vcore, as power is proportional
to F*C*V**2, where the "V-squared" term is the Vcore voltage. Some
Asus boards are notorious for overvolting, even when the Vcore
is set to "Auto".
Finally, you should be using the delta between CPU temp and case
temp as a metric. Case_temp + Power*Thermal_resistance = CPU_Temp.
If the case_temp is not the same in your two examples, then
you need to factor that in. You should be comparing
CPU_Temp - Case_temp for the two computers, as a means of
roughly comparing performance. As I said above, if you were using
the same HSF for both tests, a difference in die area can influence
how effective the HSF is, and you might need to correct the
results slightly for that.
In the P4 world, life is much easier, as you can simply stick
an ammeter in the +12V feed, and measure the power that way.
You can also do this on an AMD board, or any other board for that
matter, by removing the "high side" inductor in the Vcore circuit
and placing the ammeter in series with it. There is a fair
amount of sawtooth ripple in the signal, so in fact an oscilloscope
and precision shunt resistor is probably a better way of getting
at the truth. I use a DC clamp on ammeter with Hall probe to do this
measurement, and even though my meter is "True RMS", it doesn't
have the bandwidth to do the measurement properly. A typical meter
might have a 50KHz bandwidth, and the switching waveforms run from
200KHz and up. That is why snapping a pic of the current waveform
with a scope makes the job of measurement easier to verify for
correctness.
HTH,
Paul
