I am yet to build a thermoelectric cooler, but I've been looking into this off and on for a year or so. I'm going to build a heat exchanger based on this, but it's going to take me a couple of years. Reasonable chance it'll be my final year project at uni. These are my thoughts on it to date. Perhaps the rambling will be useful.
The old school approach of strapping a tec between processor and heatsink/water block doesn't work so well anymore as to deal with the heat load the tec has to be bigger than the space available in the socket. Imaginative designs for the block (someone on here suggested a cube with pelts on the edges) may resolve this.
The most sensible application seems to be building a chiller based on them. This isn't particularly easy, but if you're the same guy who took a milling machine to his ek supreme then I think you know a guy who could help. To start with I suggest you read
this catalogue.
***********************************************************************
What it boils down to is multiple large peltiers running undervolted is the way to go. If you run them at 12 or 15V then they move more heat, but they also waste a lot more and the cooling loop has to be massive. By using several at 5, 6 or 7V they're more efficient so you get a larger heat exchanger but a smaller cooling loop, and use a less ridiculous amount of electricity. With care you can get the cop (coefficient of performance) (ratio of heat moved to heat generated) to about 3, but realistically it's more likely to be around 2. If you don't undervolt, it might be 0.5 or so.
Xtreme systems suggests that a nominally 226W tec will draw 9.5A at 6V, this is a 50x50x3mm one. There is considerable advice in favour of using a dedicated meanwell psu for this, I'm ignoring this at present. Choosing 6V purely because 12V is readily available from pci-e lines, which can carry
at least 150W. Don't take my word for it until I've actually done the calculations and preferably built the prototype, but it looks like it'll be possible to run a pair of 50mm peltiers from a single 8 pin cable.
One significant advantage is that you no longer have to care about the hot loop temperature, as there's no electronics in it. Neither feser nor thermochill were prepared to quote a maximum operating temperature for me, however laing were pleased to say the ddc is rated up to 62 degrees, and tygon is a bit above this. Peltiers will start to degrade around 80. This effectively limits the hot loop temperature to 60. Radiator efficiency improves fairly linearly with temperature, so this helps keep the number of radiators sane. I've tried hard to get a result where tecs allow fewer radiators than conventional water cooling for matx systems, I don't think it's possible. The temperature difference between 30 and 60 degrees looks great in centigrade but pretty negligible in Kelvin. Shame.
***********************************************************************
A choice you'll need to make is whether you want to go below dew point or not. Peltiers are capable of producing roughly a 35 degree difference between hot and cold loops, where you just add more peltiers to the block to deal with more heat. Beyond this you have to stack them, at which point you can take temperatures pretty much as low as you like but you will not like the electricity bill.
The sensible (relative term) use for them is to keep the cooling loop at ambient or just below, ensuring it is above dew point and so avoiding all condensation related issues. You can control the peltiers by various circuitry, the idea is pretty much you severely undervolt them when idle and run them at higher voltages when there is more heat to remove, it's probably possible to keep the water temperature very close to constant independent of load. All that happens as processor load increases is the hot loop gets hotter and the electricity meter spins faster. You have to use enough of them to deal with the maximum load they're likely to see. Parasitic load is zero as it isn't below ambient, so the calculation is easy enough. Including a 120 radiator in the cold loop will make it just about impossible for condensation to occur and provide a degree of robustness against tec failure, it'll bring temps down until you hit ambient and bring them up if you go below.
The daft use is to take things below ambient. This is daft pretty much because phase does it cheaper. Peltiers can do it quieter (assuming compressors are loud and you don't end up running hundreds of fans), in a smaller space (again assuming the radiators aren't too ridiculous), don't feature toxic gases at various pressures and can probably be assembled only using a milling machine. Peltiers can also lend themselves to a modular design.
***********************************************************************
My initial idea was to fit as many pelts as possible within a 5.25" form factor. I reckon you can get six in there at a stretch, with four a more sensible option. Only aiming for ambient cooling, automatic control based on the difference between two thermometers, one in the cold loop and one in the case. Probably with a control which doesn't let the hot loop go over 60 degrees based on a third thermometer, running the system off a few pci-e lines and a molex for the control circuitry.
However now I'm more inclined towards mass production of a smaller unit. A heat exchanger with two 50mm pelts, run off a single pci-e line. Possibly with internal control circuity, more likely an external control box with multiple pci-e lines going in and some thick gauge wire coming out with variable voltage across it. The intention is to prototype this then go through a series of designs attempting to perfect it/get it working, at the end of which any number of them can be flexibly combined.
***********************************************************************
So yes, it could be a very worthwhile upgrade. You'll basically have to design and machine the things yourself though unless you find someone willing to do this for you, there isn't really anything available on the market yet.
