Outline of my plan

My cooler is going to use two peltiers, mounted back to back, with the hot side of the peltiers cooled by water.  I will fabricate a copper heat exchange unit which will be mounted on the peltiers, and will pump water through it to remove it from the system.  The water will be cooled in a radiator, or an oil cooler.  The system will be thermostatically controlled.

The processor that I'll overclock is an early Pentium II 350.  The fact that it's early is important; it means that in conjunction with an Abit BH6 or BX6-2 motherboard I can unlock the multiplier lock that is otherwise inflicted on the chip.

What bits do I need?

I have some of the parts already; others are about to be ordered.  The main parts are:
 

Peltiers		I'm using two, but I forget the exact model names.  Their approximate specifications are 12 V, 8 A, dissipating about 50 W.  I'm getting them from Marlow, as they're the cheapest UK supplier I can find, and they seem to be quite helpful. There are two possible ways to mount them.  I could have them running side by side.  This will give me the stablest temperature -- they'd be able to cope with almost any load thrown at them -- but won't be very cold.  Or I could mount them back-to-back, using the cold side of one on the hot side of the other.  This gives the maximum cooling effect, and I think this is the method I'll use.
Water pump	I'll use a fish tank pump, as it'll be cheap and reliable -- any 100% duty cycle pump could be used, and these are the easiest to get hold of.
Power supply	I'm using a pair of 200 W AT PSUs.  I have to use two because one alone provides slightly less current than I need.  I'll wire them up so that they use a common switch, and will mount them together such that their fans blow through the radiator.
Heat exchanger	I will make one, milled from a solid block of copper.  The block I'm using is about 6×2×½inches³, and cost about £8.  Copper is a better thermal conductor than aluminium, but is harder to mill.  I could use aluminium, but don't want to because of its inferior thermal conducting properties.  If the worst comes to the worst, though, I may have to.
Radiator	For the radiator I'm using the heating matrix (I think that's what they're called) from a Mark II Ford Escort XR2.  It's a nice size, and has lots of fins.  It was also pretty cheap, and in good condition, so it should be just about perfect. The external dimensions of the radiator are approximately 23×16×5cm³.
Motherboard	I have got an Abit BX6 Revision 2, as it's got the best spread of features of any motherboard.  I got it from Simply Computers in the UK, as they were the cheapest I could find and give free delivery for orders made over the Internet.  Unfortunately, I have one of the later variants of the board, and cannot read the thermal diode temperature, until I solder a resistor and capacitor onto the board.
Pipes	I should be able to use cheapo nylon pipes, that should be fairly flexible; other parts may have copper pipe.
Insulation	I will have to build an insulated box to put around the processor, to stop it icing up.  I should be able to get away with cheap foam insulation.
Processor	The processor I'm using at the moment is a first generation Deschutes-core Pentium II, originally sold to run at 350 MHz.  This is good and bad.  It's good because the chip doesn't have an effective multiplier lock, which makes overclocking easier, but is bad because it's only got slow cache RAM on it.  At the moment, it's got a socking great heatsink on it, and has a fan formerly from a PSU to blow on it.  Click on the picture to see it properly. If the processor ends up limiting the speed then I might replace it with a Socket 370 Celeron on a Slotket.  I'll have to make some modifications to the cooler to enable this; I'll have to redo the thermal plate.  However, the design of the cooler should mean that's the only change I'll have to make.  I'll get one with a high multiplier; a 433 MHz (or greater) one.
Control logic	I am hoping to implement some kind of a contoller, so that the voltage across the peltiers increses as the temperature rises, and also so that I am prevented from turning on the PC if the temperature is to high.  I'd also like to have a system the PC gets turned off if it begins to overheat.  Most of this is unproblematic; the thing I'm not so sure about is the voltage regulator.  I'll have to use op-amps and stuff, and I really don't know how to go about such a thing.

I will put up a rough schematic some time; these will show the approximate design of the cooler.

Back to Top

What to do with the bits

I am removing the PSU from its metal box, so that I can slightly rearrange its design.  The fan that is part of the PSU will be moved, so that it be used both to cool the heatsinks inside the PSU and the radiator; I will probably need an additional fan for this task, but I don't know quite how effective the radiator will be as I don't have it yet.

The heat exchanger will be a solid block of copper, with a deep channel milled into it through which water will pass.  It will be sealed with a copper sheet that I will solder in place, which should provide a nice water tight seal.

I am also going to have to strip the plastic case from my processor.  I'm a little wary of this, but as long as I'm careful it should go okay.  The reason I want to open the processor is because I want to replace the aluminium thermal plate that comes with the processor with something else.  The supplied plate touches only the processor core -- I want to be able to cool the cache too.  The chips on the little circuit board in the processor cartridge aren't all the same height.  I need to have a flat sheet with small protrusions in order to touch all the parts that need cooling.  Also, I will be able to make the plate out of copper, which is a better thermal conductor, so should perform better.

I rather want to embed a thermistor into the cooler, so I can read the temperature, and have the voltage applied to the peltier varied accordingly.  I am devising a system similar to that which Kryotech uses, whereby you cannot turn the PC on until it is sufficiently cold.  Though this shouldn't be too tricky to implement, some of the repercussions of the system may need working around.  I also want the system to power down automatically if there is prolonged overheating.  This shouldn't be too hard, with the ATX PSU; all I'll have to do is send the required signal to the motherboard, and it'll shut down and turn off by itself.

I'd quite like to drill a small hole in the fascia of my case, and insert into it a suspend switch, so that I can force a suspend.  There's no real reason to do this at the moment, but if hibernate to disk starts to work in Windows 98 (and it should do, eventually) then it may be quite useful.

Back to Top

What I've done so far

I've replaced the old Dell non-ATX PSU with a new el cheapo 250 W ATX one, and have replaced the non-ATX Intel/Dell motherboard with an Abit BX6-2.  At the moment, I've only overclocked to 4 by 103 MHz, which gives a noticable speed boost.  It does this at 1.99 V (or thereabouts; it's set to be 1.95 V, but is coming out a bit high).  It is rock solid; the case temperature reaches a maximum of 28° C, and the heatsink reaches 26° C, which is rather unlikely and probably not a true reflection of the processor temperature.  When I've left the system off a while, to let it cool down, I can use 4.5 × 100 MHz, for 450 MHz, but it heats up and stops working, so is no good.  Perhaps more voltage would help, but I'm loathe to do such a thing until I get the cooler organized.  All I have for cooling at the moment is that heatsink and fan that's attached to the processor.  The fan is stuck on by a cunning combination of Sellotape and a blanking plate bent into a U-shape.

I've also rewired the non-standard switch and LED connectors so that they use the standard connectors.  I might put up a rough schematic, as I'm sure I can't be the only person to have been irritated by this problem.

When I get round to it, I'll order the peltiers.  The power supplies are sitting in my room, looking menacing.  I'll repackage them soon, so that they can use a common power lead.

I've tried using a 112 MHz FSB, and tested extensively at 3.5 × 112 MHz = 392 MHz.  Everything works fine -- even a really old hard disk, which is slightly surprising.

I've also tried using the 117 MHz FSB setting with a PCI multiplier of 1/4.  This too works -- the video card doesn't (seem to) mind at all.  I think it needs a fan on it anyway, though.  I've not tried the 1/3 PCI multiplier yet, as I'm worried about what it might do to my hard drives.

Back to Top