Using undervolted TECs in cooling: experiment

[zds]

In short: by utilizing thermo-electric cooling modules properly, you can get your chips run cooler than without them by using exactly the same heatsink and fan.

Here is my take in this, in form of practical experiment:

http://zds.iki.fi/zds/projectlog/2007/07/02/120

Please note that by utilizing the TECs the wrong way you can fry both your chip _and_ the expensive TEC module, so if you try to do this, make sure you get your math right and test it first with some less valuable equipment than a microprocessor.

[lm]

Perhaps I missed something, or did you do those tests outside a case?

Any energy the TEC itself uses is going to heat up the air inside the case and making system cooling harder as a whole.

[Trekmeister]

My experience in peltier elements is limited to the occasional mini-fridge, so it is very limited to say the least. However I was under the impression you need to feed 20W into the TEC in order to remove 20W from the cold side to the hot side. You only pump a few watts into the TEC and still load it with 20W of heat. Or is my knowledge even worse than I thought?

Oh and for the power fluctuations, perhaps you see the large drop in voltage on the 5V line because you don't put any load at the 12V? Did you try putting a load on the 12V output while doing you tests? You could always try a larger resistor for the heat load and put it on the 12V output instead.

[jaganath]

It's not unusual for domestic heat pumps to have a ratio of heat moved to input power >1:1, maybe this is true for TEC also.

[zds]

Perhaps I missed something, or did you do those tests outside a case?

Any energy the TEC itself uses is going to heat up the air inside the case and making system cooling harder as a whole.

The test was totally synthetic in nature, so no case was present. One example of how you might do it in real life is this:



Here is depicted quite typical small HTPC case, where heatpipes transfer the heat to heatsinks on the sides. The important difference is that between the heatpipe fastenings and heatsinks is six TEC modules and some insulation. This means the only part that would get hotter than usually would be the heatsinks, while everything inside would get cooler than usual.

Another way to do this would be to use water cooling and then place radiator so that the heat would go straight out of the case.

[zds]

My experience in peltier elements is limited to the occasional mini-fridge, so it is very limited to say the least. However I was under the impression you need to feed 20W into the TEC in order to remove 20W from the cold side to the hot side. You only pump a few watts into the TEC and still load it with 20W of heat. Or is my knowledge even worse than I thought?

If you look at the numbers in my experiment, you see that in the latter TEC test CoP (Coefficient of Performance) value was near 4. This means for every watt put to peltier element it pumped four watts of heat. So you get the target heat pumped with adding just 25% more heat to the system.

With undervolting TEC more and placing multiple modules in parallel you can get even better CoP. To my knowledge there is no theoretical upper limit to the amount of heat you can pump per watt, but in practice CoP 5 is already darn good. And in case of TECs you need to seriously over-rate their capacities to get there and thus it also costs a few bucks.

Oh and for the power fluctuations, perhaps you see the large drop in voltage on the 5V line because you don't put any load at the 12V? Did you try putting a load on the 12V output while doing you tests? You could always try a larger resistor for the heat load and put it on the 12V output instead.

Gotta experiment with that.. the PSU I used is pre-P4 era, so it has it's most juice in the +5V and +3.3V lines, so it _should_ be able to cope with load only there..

[zds]

My experience in peltier elements is limited to the occasional mini-fridge, so it is very limited to say the least. However I was under the impression you need to feed 20W into the TEC in order to remove 20W from the cold side to the hot side. You only pump a few watts into the TEC and still load it with 20W of heat. Or is my knowledge even worse than I thought?

To clarify still a bit: the trick here is to move the heat so that the heatsink gets a lot hotter than without the TEC while still the overall heat load is not significantly more. And as heat transport from heatsink to air is approximately linearly proportional to temperature gradient between air and heatsink, the hotter the heatsink is, the faster the heat moves to air.

Thus by artificially keeping heatsink hotter without actually having so much more heat load we can keep the target object cooler than otherwise.

Even if you are not into heatpipes and/or water cooling, in chassis like Antec P180/P182 you could maybe install the TEC module(s) between the heatsink and CPU and then have fans evacuate the heat before it spreads to the rest of the chassis.