pelt pumping power

[athlonnerd]

since87 brought to lite on the formums here that running multiple pelts at way low power is a vary efficient way to cool something, becasue of the non linearitry of pelt out put vs. pumping power. what i was wandering is if yall had any ideas of how to actually measure the nonlinearity, so we can get some numbers for commonly used pelts in overclocking. it would require one to measure hom much heat the pelt is moving.

i had an idea of having a copnstand heat load, attached to cold side of pelts, with a heatsink attached also. then cool the hot side of pelt and blow air over the heatsinc.

assuming that *c/w is practiacl, adn thus if you double the heat load, the delta t will aslo double.

you ncould then:

run the set up with pelt V=o, so that all the heat load is passing through the heatinc, so you get a temp reading

runt eh pelt at 1v,then 2v, etc.

the graph of temp vs. trial should not be linear.

what do yall think, would this be an accrate way to do things. if so, i might do this, im very interested in some real world numbers on pelt efficiency.

 

[athlonnerd]

something like this, of coarse, everything will be insulated with poly styrene or something, except the heat sink. readings will be taken from the copper block

 

[Since87]

This seems like a fairly complicated system to me. I haven't worked out the math to know what all the issues would be. I'm not sure how'd you'd know how much of the heat was travelling through the pelt and how much was traveling through the heatsink. Also, some heat is going to conduct through the pelt and the coldplate-waterblock screws at 0V across the pelt.

I think I'd eliminate the heatsink from the picture all together. I'd start with the pelt near Vmax, and use a heatload of 70-100 Watts to simulate some hypothetical CPU. I'd measure the the temperature of both the heat die and the waterblock as close to the pelt surfaces as possible. Then record both temperatures for Vmax, Vmax-1V, Vmax-2V...

Do this until the difference between the two temperatures is zero.
This would give you the upper part of a deltaT vs voltage curve. The lower part of the curve wouldn't be of as much interest, but if you wanted to get a curve for lower voltages you could use a lower heatload like 25 Watts or something. In fact, a set of deltaT vs voltage curves, at different heatloads, would be very interesting.

The paper I've based my pelt calculations on is somewhat old, so I'd be interested in seeing if newer pelts match up with what that paper says.

 

[athlonnerd]

well, having heat leak throug the pelt would only aid in making the setup more real world. as fotr the heat sinc, having no heat sink would work, but you would be dealing with extreme temperatures, seeing as everything is insulated. i does not have to be a heat sinc, just something wiht a constant *c/w.

 

[Since87]

The reason I was suggesting running the test from Vmax down to the point of zero temperature differential across the pelt, was that this would insure that you didn't get extreme temperatures. At the voltage where the the temperature differential is zero, the operating Q is equal to the power dissipated in the heat die. Measurements below that voltage are measurements of the pelt in an operating region no one would ever want to use, so I don't see much need to get that data.

The water cooling system should have a very constant C/W. The pump heat will skew the results a little bit, but if the pump is inline, and you can put a fan on it, the pump's heat will be small enough it can be ignored.

 

[athlonnerd]

sorry, i didnt under stand what you meant the first time. it would indeed be interesting to obtain this data. itl be a while befor i start though. im kinda short on cash right now, and am being bombarded form every directoin with homework lately, and i have a robotics competition in 2 days, for which we have no working bot