Fridge is where we all want to be really...
Not that my plan has made any progress yet. I'm still at the point of learning the thermofluids required to allow me to start the design process.
Phase change doesn't get rid of heat either. That's why there's a damned great radiator thing with fans attached involved. No cooling solution ever gets rid of it, it just moves it to a more convenient place.
There's more to peltiers than the wattage. Most critical is how much heat it moves across for a given wattage of electricity. Say you could move 80W, but it'll cost you 60W to do so. One side goes down by 80, the other goes up by 140. When undervolted the efficiency can be pretty good, at 12V it's shocking. Say, 5 or 6V, both being conveniently available from a (good) atx psu. Say one of them can move 50W fairly efficiently, and you have to drag 300W out of the loop. Fairly clearly you'd need six of them. This is why you can't just mount them between processor and waterblock anymore, processors are too powerful for a reasonably sized pelt.
So, if you run one loop through processor, chipset, graphics. And another through some radiators. You can then join them together via a heat exchanger, which is where the pelts sit. So one loop gets somewhat colder, while the other gets a lot hotter. As radiator efficiency improves with temperature, having one loop at 60/70 degrees while the other is at 10 isn't so bad. As long as the pump and tubing can cope.
Next up, condensation. The solution I like is controlling the peltiers to keep the cold loop temperature at ambient/above dew point. Two thermometers, one at ambient, one at loop temperature, and run the pelts at variable voltage to keep the two equal. Fairly simple electronics, I think.
I think it's possible to build the exchanger into the 5.25" form factor. This depends a bit on available pelts and required wattage. However if possible, a (very heavy, and internally insulated) 5.25" box, with four G1/4" threaded holes on the back, and power connectors in the shape of 8 pin pci-e leads, would find great application in powerful matx systems.
I think it'll be about £200 in materials. Probably going to take me a few years to get the design down though. So, I'm rambling. Going back to Rogers & Mayhew I think. Cheers Super, you've got me thinking again
