I know im a bit late but i thought id add to the list of water facts on this post. As stated before water expands when it is frozen, in actual fact it starts expanding when the temp drops below 4'c then at 0'c its volume greatly increases however after this it starts to contract again.
This is due to the orientation of waters molecules caused by its hydrogen bonds, but i wont get into that :rolleyes:
IMO this would have been quite an important fact if someone had decided to try using this method with phase change cooling because although my knowledge of how much the volume of ice changes after 0'c is very limited im sure it would only require a very small contraction to cause some major problems (even worse then its less then average thermal transfer problems). Think cavity with "loosely" fitted block of ice in the middle! your not going to get the best thermal transfer when your compound isn’t even touching the CPU :eek:

Heres a shoddy graph just incase my shoddy explanation didn’t make much sense :D

http://www.freewebs.com/sadfes/water%20dvt.jpg

Water expands when it cools.. fill an ice cube tray 3/4 full (deep) with water, freeze it, and the ice will have expanded to fill the area.

I disagree. Water should be better at filling in the micro-crevices due to it's low viscosity, leaving less air bubbles. Also because of the ultra thin layer water will produce, it's thermal conductivity shouldn't matter much. My personal tests showed ~3C improvement compared to multiple mounts of AS3 (using cheap testing equipment of course)

I sealed the top of the processor package using a wide piece of clear tape (cutting a hole for the die) then I made a bead of dielectric grease circling the die to create an air tight seal between the processor and the bottom of the waterblock, preventing the water T.I.M. from leaking or evaporating.

I didn't do any long term tests using water as I wanted to try Fluorinert FC40 as a substitute since it's non conductive. Unfortunatly I had no luck in getting a sample.

I see one potential advantage
 

The only thing that I see that could be considered beneficial for ice to be a TIM joint is its high specific heat.

Basically, as the processor goes from idle to heavy use and back down, it generates thermal transients, which is undesirable from a stability standpoint. Phase changers and aircoolers have a really high DT between these states because of the low specific heat capacity of their cooling mediums. Watercooling maintains a much narrower DT for idle and work states than either of those approaches due to water's high heat capacity.

If there was ice as a TIM, and the phase changer could keep it as ice, then the delta T between idle and work state temps of the CPU would be much narrower I think. However absolute temperatures overall would be higher, so I don't know if that would be a worthwhile trade. Interesting idea, good luck if you do it. Sounds like the expansion/contraction issues, plus how to keep it ice when the computer is turned off, are some serious problems in application of the idea though!

Hmm, so how did you apply the water? Did you just put some drops on the die and seal around it, then mount? Let us know how exactly you did this.

I plan on doing the mod to my NF7-S that allows for the reading of the internal diode temp probe. So that will give me some descent temps readings IF i decide to even bother trying it out. I with mercury wasn't so toxic :) .

After I made the bead of noncondutcive grease encircling the die, I put a single drop of water on the center of the die then carefully mounted the waterblock. You don't want to make the bead too thick nor too close to the die. I practiced with a 2 inch square of plexiglass over the cpu (removed from the socket) to make sure the seal was consistant and that the water didn't squirt past the seal when I applied pressure.

P.S. I should mention that I did this all a couple of years ago with a palamino processor which didn't have anything mounted on the surface of the cpu, unlike the newer athlons. The clear tape I used may not work so well for you.

Water contains a lot of dissolved gases which will tend to collect on the surfaces of your cpu/block and cause much lower performance though. Just try leaving a glass of water for a day or two - you'll notice a lot of air collects on the surfaces of the glass.

I'm sure if I could fit an entire glass of water on my cpu that gas bubbles may become a problem (j/k). Even when using a single drop, most of the water is squished out. I don't see gas being a problem but I do see pumpout being a potential problem.

The low viscosity will leave a very thin layer, but the high surface tension (and the fact that neither silicon nor copper wet really well) will mean a tendency to bridge the micro-crevices instead of filling them.

Now if you'd used glycol, with its excellent wetting and hardcore capillary action...

I would think that the mounting pressure would be enough to overcome any surface tension as my results seem to indicate but perhaps glycol would be even better.

I think most of the mounting force would be transmitting by the areas of direct block-die contact. Water would squirt out rather than be pressurized. And I think that's why you got good results -- minimal TIM, maximum contact.

The way to find out is to test with a range of mounting force. Compression of the trapped air (less volume, higher thermal conductivity) is one of the ways TIM joints improve with greater force (I don't think water would be as subject to the 'more pressure = thinner TIM' mechanism like high viscosity goops).

I just wanted to say that most of the air in which ever liquid you use will float upward w/ some vibration. So that would leave the air all on once side of the die w/ a mid tower case (depending how the socket is positioned). And we could allow enough room between the die and the seal for the air to remain w/ out affecting thermal contact. It will be the difference between thermal contact and thickness of TIN VS. thermal conductivity.

So if/when I try this out any other liquids worth trying?