Direct Die Cooling (Project O-Ring, Part 4)

When it comes to direct water to die cooling, we at the abode like to think of ourselves as pioneers. While we were not the first to eliminate the barrier between water and CPU, we knew of no one that had previously tried it with the exposed core of the "flip chips". Some of you may remember our early attempts at direct-to-die (D2D) cooling on the Intel FCPGA chips and later, Spode's socket A "Sunny D" setup. These early attempts were enough to get us exited about the performance of this method of cooling and warrant further testing. The one major drawback of our designs was the method of attachment. In order to seal the block to the CPU it was necessary to use a sealant such as silicone or epoxy. The socket A chips presented a new problem of exposed components on top of the packaging that would need to be sealed as well. The problem with gluing the thing together is in trying to disassemble the works when you want to change anything. This proved fatal to many a Duron and several Celerons as well. The goal of this project is to eliminate the need to use adhesive to make the seal to the CPU.

We had made quite a few of these and pretty much had the best design for the internals figured out, so the main goal was to come up with a method to seal the block/CPU together and be able to remove it easily. This is very similar to my earlier attempts as far as construction. A cube of acrylic bored down the center, however this one has a couple changes.

The first difference is the use of the 4 socket A bolt holes on the mother board to attach it. I am surprised more makers aren't utilizing this mounting system - it would seem far superior to the flimsy plastic socket tabs. I like to use the nylon fasteners to eliminate the possibility of the bolts shorting something out.

Here's the big difference. I machined a groove around the outside of the cavity and siliconed a rubber gasket into the groove. The opening is just large enough to clear the resistors on top of the AMD chips, and seals against the packaging. Spode discovered that my swirling coolant in the first design was less than efficient, now the delivery tube lands the cool water directly on top of the die to reflect Spode's improved design.

Applying a little silicone grease to the gasket will make me feel better about possible seepage.

First we need to seal up the resistors/bridges on the front of the CPU, and then its time for the testing! After bolting it to the motherboard with #6 nylon fasteners and bleeding the air out of the system, it was time to power up! Thankfully it worked and didn't drown my system yet again, I don't think my GeForce2 could stand another bath :P

The temps were consistent with our previous attempts and there was no sign of leakage. SUCCESS!! .......well , almost that is. Unfortunately after running for some time the water had penetrated the sealant on top of the CPU and shorted out the bridges. Fortunately this didn't kill anything. Taking the block off was a no brainer, and I was testing different sealers for the CPU in no time!

After trying silicone, polysulfide, polyeurathane, grease, enamel and a host of other things, I was never able to find a sealant that would hold up over time, they would all eventually leak and short the CPU. I feel this may have had to do with the ceramic packaging of the Durons absorbing water, and hope to resolve this problem with the new packaging material used on the Athlon XP CPU's.

One of our major fears of this method of cooling was dispelled when I had a pump failure while running D2D. I came home and heard a bubbling sound and realized the water had stopped in the block. As crazy as it sounds, it appeared to be boiling the water in the block, and pushing steam up the hoses. To my amazement when I took the block off and tried to boot it with a HSF it fired right up! Apparently the water sitting in the block in contact with the CPU was enough to keep it from smoking!

After the good doctor had finished with this cooler, he sent it my way to have a look at it. I set up on the same system I was running my Sunny D cooler and althought I can't find my results anywhere due to harddrive failure, from memory I remember them being HALF that of the Sunny D cooler. Wether this was due to increased flow through the block (and I'm using a pretty crap pump) or just a better design I don't know.

There is still some thought that direct die cooling gives misleading results becuase it's using in socket probs it is just cooling the ceramic. I think this is wrong, becuause ceramic is a bad conductor of heat anyway, if you take a chip straight out of a machine and feel the base, the section directly under the core is always hottest, it doesn't even out - the thermal transfer is poor. If cooling the ceramic around the chip is supposed to affect how the temperatures are, then when you takea chip out, the entire chip would be warm - unless it's been on a while, it isn't. Thats just my .2$. When we get the dough and the time to test a palomino with internal temperature probe, we shall know for sure.