After the modification with the peltier-cooler and the GlobalWin FDP-32 heatsink, I still wasn't satisfied with the cooling performance. So I decided to build myself a water-cooler with which I was going to cool the peltier. I had seen this at a friend of mine who cooled the cold-plate on his Celeron 400A with the same peltier and a home-build water-cooler down to -25° C (-13° F) while the CPU was at idle and rain was turned on. While playing Unreal Tournament, the CPU temperature went up to approximately 5° C (41° F). The water returning from the cooler is cooled by letting it flow through the radiator of a Subaru Jumbo, while letting the fan on this thing turn at a silent 3 V. He had his Celeron overclocked to 600MHz, the power output at 2.3 V was 47 W (formula: P_actual = P_default x (V_core / V_default)² x (f_overclock / f_default) ). I also calculated P_actual for my K6-2 450 running at 560 at 2.4 V, P_actual = 18.5 x (2.4 / 2.2)² x (560 / 450) = 27.4 W. This power-output was considerably lower than that of the Celeron, so in theory I should be able to cool my K6-2 under load better than the Celeron.

And this is how I did it:

I started by designing a water-cooler which had to be easy to make and have good cooling performance (you can download the Acad drawing here). I had a 40 x 10 x 1000 mm copper-rod lying around, which I wanted to use. In the drawing you can see that I did not drill the holes right next to each other on the centerline, but I shifted them up and down around  the centerline. I did this so I could drill more holes in the cooler whereby the water would have to take a longer way through the cooler and take up more heat in the process. Of course the amount of heat absorbed by the water is also dependent on the flow-speed of the water. If it flows too slow the warmer water will not be able to take up any more heat and an upstream portion of the cooler

So after designing it, I cut 3 pieces from the copper-rod. 1 piece of 40 x 40 x 10 mm (the cooling-part) and 2 pieces of 40 x 3 x 10 mm (these 2 will seal the cooler at both ends after the holes have been drilled). Then I drilled 6 4 mm holes through the copper. At one side I drilled the 2 outside holes to 5 mm, 10 mm deep so the would accept the copper tubes. Then one of the small end-plates was also drilled with 2 5 mm holes which fit exactly over the tubes.

The parts can not be soldered together yet, because first the holes have to be connected to create a single continues duct through the copper. the connections were made on a milling-machine, but can also be done by hand with a boring-machine.

Now the parts can be soldered together. This was done with Thin-Flo paste 175 (Castolin Eutectic). This chemical is smeared on the contacting parts which are then lightly pressed together. with a blow-torch the parts are then heated, which makes the chemical to transform from a dark-gray clay-like material to a silver-isch soldering material which in its heated state flows in and nicely fills any gaps still left between the parts. The cooler was tested with compressed (3 bar) air for leaks.

So now that it was ready I was very anxious to test it out. I "ripped" the FDP-32 of the peltier and installed the water cooler. Since I only wanted to test it out without turning on the computer I did not insulate it for condensation. After switching on the power unit (see PART 1) and the little 10 W aquarium circulation pump, I cranked the peltier up to full power (15.8 V and 5.5 A) and down went the temperature. Within approximately 25 seconds the temperature of the cold-plate was down to -29° C (-20° F). Now that was not bad, but of course the CPU did not produce any heat.

So after some careful planning I started to assemble the system properly (with the right amounts of silver-grease between the parts) and also isolate it with green compression foam, which I had left over from my work at Fokker. This foam (forgive me for not knowing the exact name anymore) is able to withstand very high temperatures (up to 200° C (392 ° F)) without melting or oxidizing. I cut pieces so as to enclose the cpu and cooling-parts completely without leaving any air between them. I also put a piece of foam in the cpu-foot under the cpu to prevent any condensation from forming on the underside of the cpu. The foam parts were connected by double-sided tape and are also "glued" to the motherboard with this tape, but can easily be removed while providing an excellent air-seal. The cooling unit was fastened to the motherboard with the clamp that came with the FDP-32. I had to modify it because it wasn't long enough anymore. To do this I just made 2 small hooks from piano-wire which I attached to the clamp. As a finishing touch I made a foam cover which I also attached with the double-sided tape.

So now everything was ready for a first test run. I switched on the pump and the power unit, waited until the cold-plate was -20° C (-4° F) and swithched on the computer. And there it went for the first time with the water cooler at 2.4 V. It posted, booted, started Windows and crashed. What the f*ck was that. That was not supposed to happen. I switched the computer off and on again, now it didn't even get past the memory-check, it just crashed. I didn't understand it, I was certain there could be no condensation in the foam. So I decided that 2.4 V was not enough (since the cpu only ran at 560 MHz at 2.7 V with just a Alpha heatsink). I tried every voltage from 2.2 to 2.9 nothing worked.

And then it hit me, the top of the cpu got so cold that in turn also the little metal pins (you know the 237 something connections) got below freezing. Now that was not a problem because it was all isolated. But the pins cooled the cpu-foot on the motherboard below freezing (no problem, still isolated) but the foot was soldered at the backside of the motherboard. Oops not isolated, I checked and there they were, the most beautiful frozen drops of condensation I ever saw. :-) So I cut a piece of foam and put it under the cpu on the backside of the motherboard and my baby has been purring at 560MHz at 2.4 V very happily ever since.

The temperatures with cpu idle and rain on are: cpu -13° C (8.6° F), cold-plate -23° C (-9.4° F) and hot-plate 25° C (77° F).

While running Unreal Tournament: cpu 3° C (37° F), cold-plate -11° C (12° F) and hot-plate 28° C (82° F).

Unfortunately, I can only pump around the same 5 liters of water continuously, without cooling it. This will change when I also install a small radiator + fan from a car or motorbike. Since the cpu is able to run at 560 MHz at 2.4 V now I will attempt to get it up to 616 (112 x 5.5) 616 MHz at a maximum of 2.7 V.