Ok one last test for tonight:

http://www.procooling.com/~phaestus/distribfold.jpg

Still not nearly as tight as CPUBurn but nowhere near as erratic as Prime95. I believe distributed folding is quite ram dependent; I will add another 256mb stick tomorrow and see if it makes any difference.

(Much) more to come

It takes a surprisingly long time for your CPU to return to ambient temperatures after you shut down the box. Not altogether surprising as the copper heatsink SHOULD become a heat source after the CPU is off. The temp jump at boot tho is almost instantaneous. Article writing up in progress as I type.

I've been beating my head against this current measurement problem all night.

I was actually hoping the THG tester had found a clamp-on that might be able to do the job accurately, but considering that clamp on is a hungarian made device with a maximum frequency spec of 1kHz, I doubt that's the case. You need something with reasonable accuracy up to about 2 MHz to have any confidence of accurate results.

It's looking to me, like the most practical means of getting reasonably accurate results, is a modified version of my second idea. (Shunts in series with the switching inductors.) I can design a little circuit that will put out a DC voltage proportional to the current draw. Let me think about how to do this...

Tanks Since87; seems like this is the last piece of the puzzle for getting a real cross calibration. I have thought of two more tests that I will do over the week:

Why not set the system up and watch the settling of arctic silver over a couple of days of CPUBurn running? I should have the required resolution.

I have another interesting test that involves extreme motherboard airflow to see how big of an effect that cooling the rest of the mobo's components has upon die temps. This is I suppose another indirect way of looking at the secondary cooling and heat sources. This one I'll start tonight.

Does the software you are using, with the MAX6655 development board, allow you to log the values for the three voltage inputs simultaneously with the temperature?

I was thinking you could monitor Vcore with the Vin3 input of the MAX 6655, and connect a signal proportional to the current draw to Vin2.

You could then take the logged voltage and 'current' measurements, and multiply them together, and then multiply by a scaling factor to get the output power of the Vcore regulator. Accuracy wouldn't be great, because the voltage inputs to the MAX6655 aren't that accurate, but the ability to see how power consumption and temperature track would be cool.

Alternatively, I could design a circuit to just put out a voltage proportional to Watts. (I'm not sure how accurately I can do an analog multiply with a reasonably cheap circuit. I believe about 1% accuracy would be fairly easy to achieve. Better than that is probably reasonably easy with some calibration of the circuit.) This Watts output could then be connected to the MAX6655 and/or a voltmeter if greater accuracy was desired.

Yes it logs the voltages as well; they have to fall within a defined range I believe for the IC but as long as they do then no problems. Being able to log them simultaneously would be a huge advantage. Good idea!

That is 60cycle electrical interference from electrical outlets, flourescent lights, computer monitor etc... try turning them off and see if the interference persists. You can also plug into a dedicated ground to help diminish it.

Didn't really read and pay attention did you? :p

Should mention that while testing is in progress that all lights in room are turned off along with the monitor of the testbed PC.

But does it matter if all the lights are turned off etc? If it does, then the measurement results are questionable with a running PC.

Verifying insensitivity to EMI would be worthwhile.

Well the light I use is a 100W table lamp; I figure it CAN'T be good to have that pointing in the same general location. Same with the monitor. Concerned with the lights and monitors as heat sources rather than EMI specifically. Other than the stir plate (which was a rather extreme case) I have never noticed any other interferences.

OK check this out
 

This evening's test was to study the effect of cooling all of the other mobo components and hot spots on CPU temperature. To do this, I took a Comair Rotron 172mm fan (it's a 24VDC fan that I am running at 12VDC but it generates a fair airflow still) and pointed it over the CPU socket, mosfets, and other hot bits in that area. I first let the CPU heat up to equilibrium with CPUBurn High priority and then turned on the fan. Here's the graph:

http://www.procooling.com/~phaestus/mobocooling.jpg

If I were to plot the raw data, you'd see that the air temperature at the heatsink intake was barely changed at all (it actually rose in temperature, but less than 0.75 C) while the CPU temperature drops dramatically. These are those secondary cooling paths and secondary heat sources that everyone is talking about.

One would expect this effect to be larger with watercooling since there is no longer any airflow at all to cool the hot parts. The magnitude of the effect is probably pretty motherboard specific as well; I have never seen this large of a difference before on my A7V-133 with diode readers.

Nice! Thanks pHaestus!

Do you think that putting a small neoprene "gasket" around the edge of the CPU, would reduce this effect? (not that it would normally be advisable!)

This is a real temp though, isn't it, so air-assisted watercooling may have some potential, no?

Tons of possibilities with your setup ;)

Interesting.
For a wb probably would expect a larger effect on the "secondary convective resistance"( Rs ?) .
However would expect this to translate to a smaller change in temp due to the lower "C/W" of the water-sink.Maybe:-
http://www.jr001b4751.pwp.blueyonder.co.uk/SecW3.jpg

Further work on an un-powered cpu with a hsf and wb(using different coolant temps) maybe revealing.

Good thinking.

Andrew at Millennium-thermal suggested once that business cards between the heatsink and socket are useful to exclude some secondary cooling paths. It is very close to time to get my wc testbed up and running for some comparisons. I will play around with the air cooling stuff more tonight and tomorrow and then Saturday and Sunday set up a wb.

SO many testing ideas leap to mind. For example, I really wish I had a TBredB and a Barton and a Morgan all to play with (and look at cache and die size effects) but money is a bit tight at the moment for such extraneous consumables. May buy a 1700+ TBredB just for the hell of it though.

I am not exactly sure what you mean with different cooling temps and the heatsink Les. You mean a temp controlled chamber with different ambient temps? I guess I may have to start pricing something like that out if I want to get serious...

Continuing with fannage variation work on hsf (including un-powered cpu)
Fannage on wb (including un-powered cpu)
Then varying temp of coolant in the wb(un-powered cpu).
All exploratory rather than definitive.

Certainly not suggesting expensive equipment.

Here is another wacky idea.

What about making an incubator using a plywood box, heat lamps or light bulbs of some sort to generate heat, and then have those connected to dimmer switches to adjust their output. Put the motherboard and system into the box. If my wireless access point can penetrate it then I can skip the monitor, kb, and mouse and just radmin into it to test. Maybe converting an ice chest or styrofoam container would work better than plywood for the wireless. Something like this seems reasonable:

http://www.melodious.com/gate/howto/coop/incubator.htm

I suspect that it would take some fiddling to get right, but I could presumably collect temperatures at a more controlled ambient this way. Just trying to think of ways to get an isothermal environment...

If I can find the stuff to make a decent res for my Little Giant pump I will start setting up a cooling loop tonight. Don't be too surprised if I am testing with the liquidcc surge waterblock for the first few months (it has a hole predrilled for a thermocouple in the baseplate and this same block has been tested extensively by Bill already).

A couple of thoughts re The Surge
Although would probably expect the "C/W(TIM)" to scale from the the 100sq mm Die to the 128(?)sq mm CPU would have reservations re the "T/W".
Better "T/W" could maybe expected from the narrower channeled "SlitEdge" ( http://www.overclockers.com/articles748/
However it is probably a little early to worry about such details.

one step ahead of you; Bill, Joe Citrella, and I are all getting identical slit edges for cross correlation thanks to Brian at BeCooling

A guess at the correlation between Bill's 100sq mm Die and Joe C's 128sq mm Die:
http://www.jr001b4751.pwp.blueyonder.../Phaestus1.jpg

Based on Heat Spreading calculations using guessed relevant Dimensions (not all given in article). Used Bill's "C/W(TIM)"=0.1 and scaled to 0.078125(1/1.28)
for JoeC's(no Spreading).

Looks like it is going to be fun.

EDIT Corrected crappy arithmetic for "C/W(Tim)".

OK this is going to require some (much?) more experimental design. I repeated the test as follows: unpowered CPU, no fan --> unpowered CPU + Rotron fan --> boot to CPUBurn with fan--> turn off fan.

Then I got the bright idea that maybe the temperature drop was only due to the increased airflow to the heatsink directly. Made a "roof-shaped" cardboard deflector to put over the 80mm fan. Repeated testing, and now CPU temperatures rise when the Rotron is turned on. This seems to be because the hs/fan's intake air temps also rise when the Rotron is on.

Looks like the important thing here is whether the supply of fresh air to hs/fan is improved or inhibited and not the airflow to the mobo area.

This is not an easy thing to just "test" Maybe a bunch of 40mm fans arranged over all the hot spots would be more useful?

Does sound like an experimentalist's field-day.
All results are revealing.
Have corrected crappy arithmetic in previous post.

I've been thinking more about how to measure the current through the switching inductors.

It's turning out to be an uglier problem than I thought at first. The big problem is the resistors to go in series with the inductors.

A length of copper would be great, except for one problem. Copper has a Temperature Coefficient of Resistance (TCR) of 0.4%/C. In other words, unless the copper is kept at a stable temperature, the resistance won't be known and it will cause error in the current measurement. The most practical replacement for copper is probably Nichrome with a TCR of 0.017%/C. However Nichrome has 67 times the resistivity of copper, which means that the shunt resistor would need to be physically large.

Looking through the Digi-Key catalog I found some 0.005 Ohm Nichrome resistors that would do the job. (Ohmite 620HR005) It would take four of these total. Two resistors paralled, in series with each inductor to provide a voltage proportional to the current flow to feed into the rest of the circuit.

The resistors are 1.675" long, and it might be a cramped fit to get them installed, but once the resistors were in place the rest of the circuit could be put in a more convenient location.

So are you up for doing a hack job on your motherboard?

Sure. Just set up my Little Giant Pump and reservoir today. Baby steps....baby steps. Hacking up motherboards sounds like much more fun than plumbing though. I think I am about to commandeer a piece of plywood I saw in the garage rafters and just mount all the valves and instruments and tubing on it. That way I can set it up like a table when in use and put it in the corner when not needed.

http://www.intel.com/design/pentium4/guides/252161.htm

3.3.4

anyone know where I can get an unpopulated (bare) P4 mobo ??

and no, I can provide no lead on getting the die

Do you have access to one of these TTV's?

Did I miss something, or is Intel actually doing thermal testing with the heatsource distributed uniformly over the whole IHS surface area?

If you can come up with a dead motherboard, I'll strip the necessary components.

I decide to test some of the CPU current measuring concepts on an old socket-7 board. Between it having single phase regulation and being disposable, I figured I couldn't lose.

First problem was the spacing of the toroids's leads. I had a devil of a time heating both at once. Eventually, I melted a big pool of solder onto the board, which conducted enough heat to the loosen the leads simultaneously.

Next difficulty was room. For my shunt I had paralleled five 0.1 Ohm resistors and there simply wasn't enough space between the socket and the nearby capacitors to place the shunt and toroid side by side.

I was forced to mount the resistors vertically, with the inductor on top and a lead returning to the board. Soldering it in was a colossal pain - the extension lead was short enough that the solder would flow at both ends. In the future, I would get some enameled wire and rewind the inductor with long leads.

Next step is amplify and digitize the reading. Last step is to add isolation, for reading from a second PC.

In the meantime, some pointless data! Using CPUburn/BurnK6 launched at high priority, a DMM, and a lot jumper changes. Current scales nicely with both Vcore and multiplier, doesn't it.

Good work.

I'm surprised your DMM did so well with the ripple that must have been there with a single phase supply.

The green line looks a bit off. I'd expect it to split the difference between the other two lines. Still, those are pretty good results.

Well, there was quite a bit of ripple. The DC reading randomly wandered in a 50-80 mA range, plus a higher frequency 30-40 mA AC component. Lets hear it for eyeball averaging.

Plus, this board was one of the early victims of exploding cap syndrome - the replacement in the regulator has triple the capacitance and half the ESR. Throw in that CPUburn is a steady running program...
The graph was made with only two significant digits. Give me a few days, and an ADC, and it'll look better. Current vs. time is gonna be sweet.