Question is we've always asked ourselves. How much power does X PART develop. While taking a shit today, i realized that measurement of power via the measurement of the time required to heat up a certain object, the delta and the final temperature at the end of that time could technically show how much power that item develops when it's working. Don't ask me how i thought about this when taking a dump, a smart man keeps his muses secret. Anyways, so what power do we usually want to know? Obviously the power of stuff that consumes the most power, namely processors and graphics cards. Graphics cards are easy because you can use a shunt and use Ohm's law to derive something, it's just hard as you'll have to desolder the PCI-E slot's power connectors or isolate them.

Thus we've come to the processor.

The goal of this thread is to figure out 2 thangs
1) The variability of leakage power from the same stepping and of processor (for example, 20 E6600 Core2Duo's temperature variance between units), this is for academic purposes.
2) A real quantitative figure of power use by processors under various conditions. This is a measurement of the power consumption of the proc alone, thus even if you change mobos, the figure stays the same.

Anyone wanting to join this hunt for power measurement can request their name be added. I'm looking for extreme ocers, engineering educated, math and physics inclined folk, plus God forbid it, more EE goons. So here's the list so far.

1) Empire23

If you have nothing good or useful to post here, don't post it


Anyways let's get to what i've figured so far.


1) Measurement Method 1

Q = m c ΔT

Q = Power Needed

m = Mass

c = Specific Heat capacity

ΔT = Difference of temperature

Asuume room temp is 26 Degrees C and the Target/Final temp is 90 C

Assume a processor is perfectly insulated and perfectly coupled to a 1kg block of copper.

The mass would be 1kg. Coppers Specific Heat Capacity is 385 j/grams. Thermal delta is 90-26 = 74c

Thus Q = 1x385x74 = 28490J

Under the assumption that to get 1kg copper to a temp of 90C one has to inject 29k joules of power

1 watt hour is about 3600 joules.

X watt = X Joule / Seconds

here's where we add the time. If it takes 5 minutes or 300 seconds to heat up till 90c

Thus 28490 / 300 = 94 watts.

I think this is correct, i'm not a materials or mechnanical engineer for the love of God. So i hope some engineering folks from other fields chip in with their much needed expertise

Problems as conveyed by fishychickens

1) Repeatability
2) Insulation
3) Mounting standards
4) Thermal Interfacing and Coupling
5) How people in HW are now complete idiots



2) Measurement Method 2

We can use the EE method of discovering power chomped by assuming the P = I V

All motherboards have Voltage sensors, and even a retard can clip a set of voltage probes across an inductor on a motherboard right? Point is that V is easily measured.

Problem is finding I or Current in Amperes

There are 2 methods of finding it which aren't too bad.

1) You break the circuit and add the ammeter in series with one phase.


2) You use a current probe that uses the Hall Effect.

This has everything going for it.

1) It's damned accurate
2) It's repeatable
3) You don't need to overheat anything
4) Probably easier

But there are 2 problems here as i've myself realized.

1) Some Processor voltage regulators play around with the phases and shut them down when not needed, so we MUST know the pattern and we must know how the load is distributed
2) I'm assuming that all regulators here are pass style MOSFET switching regulators.



As with all research ventures, you'll have to know your measurement limitations and what equipment can be shored up.


Things we might need
1) Processors (duh)
2) Motherboard with easily de-solderable inductors
3) Balls





I'm not expecting freaking geniuses here but i think community participation is good as in the HTT Reseach thread. Not everyone was an elite analog/signalling engineer but community participation did help alot out in figuring out on how the damn thing worked. So if you think you're good, lend a hand, we could actually compile a list that's worth a damn!

You want me to derma my setup as guinea pig?

I'm in for my own Q6600 G0.
Empire23 has his own Q6600 B3.

Anyway, i do have a lapped Pentium D 820 unused, would it be useful? [Since they are known for their uber heat]

See how Tan Sri, now just need to ask you how viable is measurement method one.

Question is, assuming we have 2 materials, copper tube and water in the tube. How to calculate leh?

I'm sure you still have a little left from your thermodynamics class

in what way can i help mr-empire23, do put my name in..

i lost 90% of my EE knowledge after i left the school for 6months of on job training
can i join? :3

This post has been edited by Beach_Boy®: Dec 29 2007, 08:33 PM

I'm not sure different steppings can be used. They must be as similar as possible and if can the same VID. Because we want to make leakage power the only variable here.

Well, it could help statistically so it would probably useful. The Pentium D we could probably sacrifice it for prototype run



Can't just put ye name in. Take a look at the problems i mentioned above and try your hand at a solution.

Can, but take a jab at the issues faced above, hell i'm open to any suggestions.

viabillity like all engineering problems would depend on accuracy levels over a number of sample taken. The problem is unless we have a way to measure raw data outputed from the Tjunction probe, it's quite hard to narrow down variances due to calbration and register reads from the respective probing chips specified as according to particular manufacturer. Which was why the MBM5 plugin to read the Tjunction for my C2D caused 50% CPU cycles due to manner in reading the registers. That i if we use Method #2. Method #1 like fishychicky stated is quite variable because of those exact reasons.

1. How do you create a viable controlled environment, unless god forbid you decide to build a pressurised vacumm chamber which would 97% eliminate external heat variances.

2. TIM is not an issue as we would just need to set an all encompassing parameter involved. Once set and we maintain to within a tolerance limit of application, our results would be within an acceptable limit +/-

3. Mounting standards can be set accordingly. All we need is a heatsink (No, I'm not going to mention waterblock here, rest assured) which utilises a screw and spring method. Then we'd need to find a mini torque wrench so as to be able to make the mounting screws setting to be as even as possible.

How do we go from here is limited by out imagination and budget as I'd prefer a near refrigeration level chamber cooling the proc bare without heatsink and utilising IR based heat radiation analysis

EDIT: BTW have you looked through again Kris@freecableguy's article concerning on Intel's power delivery specifications?

http://www.thetechrepository.com/showthread.php?t=126

This post has been edited by almostthere: Dec 30 2007, 01:39 PM

The thermal method is pretty much out of the question, for obvious reasons.
The electrical method holds much more promise. You don't even have to get a 100% correct figure for the cpu, the more important thing i believe is repeatability and accuracy of measurement, because the comparison is the important part here, a single power figure in isolation doesn't tell you much. Take a look at some mainboards, maybe they use some inputs on the 24pin connector just for the cpu power. Isolating the power from those pins alone should be accurate enough for this purpose, and isn't even hard to do.

1. Method 1 has too many losses associated with it, both mechanical and radiation wise. There are alot of holes in the system which are hard to account for at this time. IR would be best, but in this case an averaged figure would be flawed and one would have to result to thermography, which for now is out price wise.



Accuracy will be slightly flawed i garner due to a few things if one were to measure from the inductor or power circuitry.

Regulator are straightforward in a way, you can either fully parallel them and use a single feedback amp to control them or use a more efficient method that reduces losses, if one were to use more than two phases, a cascade of error amps could be used or, independent controlled switching. I'm not sure it's a good example, but here's a simple illustration of what i mean.


The problem is that we seriously don't know how the circuit is laid out. For one, i don't thing Asus publishes datasheets for it's mobos lol. The problem becomes more complex as we factor in ways that manufacturers design control ICs.

Example : Board has 4 phases, each can do 10 amps

1) Due to heat limitations the phase use isn't 100 percent linear, for example, if you need 20 amps the controller might divided the juice by 5 amp per phase.
2) Or Due to losses caused by coil hysterisis or parasitics, if 20 amps is required the controller might do say 8 + 8 + 4 + 0 amps on the phases

We're under the assumption that things are working 100 percent in parallel. BUT here's a popular controller.


What we can see controlling the PWM MOSFETS is tristate AND logic (One is fault and one is the control signal, last input is ON/OFF). To complicate matters, there's phase interleaving. f***.

The problem with your assertion of dedicated power pins is hard to use, because firstly, no mobo maker is going to publish the specs of which and what power plane goes where, they just abide by the Intel specs and they generally operate in the grey area. Worse off is that the mobo most likely sips power off the same planes, with the 680i taking up to 30 watts...hell i'm not taking chances with that kind of variation.

I think it would be quite logical that some pins are used just to power the cpu. Maybe asking nicely will get you answers. Also, there are mainboards which consume a small amount of power, so even if you can't completely isolate the cpu, you should be able to get within ~5W i think, which is plenty fine for comparison and it will be easy to do. Also if the dedicated pins idea holds true, you can even measure the efficiency of the power conversion across different mainboards.

Of course, i am not a big fan of the complex approach.