I have a piece of thin aluminum plate (one side of a type of extruded aluminum enclosure really, but the edges are not thermally well-coupled to the other sides of the box) about 25 cm long by 5 cm wide by 1 mm thick, and a 40 mm by 40 mm copper water block thermal epoxied to it. The lone cooling block looks rather like this:
Connected to a copper processor water block thermal pasted to the CPU that looks rather like this:
About 18" total of 3/8" PVC tubing connects everything up to a small off-the-shelf water/coolant pump of dimensions about 4" x 3" x 3".
With the CPU idling at around 75 watts dissipation the loop reaches thermal equilibrium in a couple minutes, with the pump flailing away at
2500 RPM, and the CPU temp reported at around 50-51 degrees C/125 F, too hot, but better than I was expecting for not much coolant block contact area at all, and not hot enough that the processor enters thermal shutdown immediately so hey that's something!There's probably enough data available from this simple setup, along with measuring the plate and block temp, to construct a thermal resistance model of a somewhat more elaborate system good enough for rock and roll.
I think three or four of them distributed strategically around the enclosure surface might be enough to keep it cool enough under moderate workload to not have to resort to TECs or fancy techniques. Maybe just a lil bitty BLDC fan mounted to one side of the final cooling block.
I'm using the "long cure" variant of this brand thermal epoxy:
it's pricey, takes about 2-3 days to fully cure but seems to transfer heat across an interface amazingly well.