Watercooling the Cyberstorm PPC

Please note that information for this article was provided by Steve Giovenella. If you have any comments or queries, please refer them to him.

For quite some time now, PC overclockers have been exploring new techniques in cooling their hardware. One popular technique is watercooling, in which the traditional heatsink and fan is replaced by a watercooling "block". This block is attached to the chip being cooled in a similar manner to the usual heatsink and contains ducts in which the cooling fluid passes through. An inlet and outlet tube is attached to the block. These tubes run via a pump to a radiator outside the computer's case. The idea is that the fluid passes through the block, cooling the chip and is then passed through the radiator, which cools the fluid in an endless cycle.

There are quite a few advantages to this scheme. Water cooling can potentially deal with greater amounts of heat than traditional heatsinks and can cool to a lower temperature - using the right fluid and by specially cooling the radiator, subzero temperatures can even be attained! Alternately, the greater cooling power makes it easier to use a Peltier cooler on the chip. Other advantages include no dust problem and less acoustic noise (assuming a quiet pump is used).

There are also disadvantages. A leak would be obviously catastrophic - and not just from the physical effects of the liquid running all over your motherboard! The loss of cooling caused by the leak would almost certainly fry the chip being cooled as there is no residual cooling effect of a normal heatsink. A pump failure would have a similar effect as well. The use of a non conducting fluid would avoid water damage in the event of a leak, but would not help avoid the loss of cooling effect. The moral of course, is that watercooling is to be done at your own risk.

This article deals specifically with cooling the Cyberstorm PPC and is still incomplete. Further updates will follow.

As you can see, there are two cooling blocks, one for the PPC processor and one for the 68K processor. Of course, as the 68K processor runs very cool, it is not vital to watercool it (unless you plan some very serious overclocking!). The two blocks are connected in series in the picture. Note also the heatsink on the SCSI controller chip and the fan on the Permedia chip on the attached Cybervision card. Future experiments may involve adding water cooling to this very hot running chip. Also note the fan on the left positioned to blow air across the memory, as outlined on the overclocking page. This is also an area for possible future experimentation with water cooling, especially the chips underneath the memory, which run very hot when overclocked.

I have a thin copper plate under the aluminium waterblock (on the PPC), as I've read its not good to put aluminium and copper in contact on the chip's surface for corrosion/electrical reasons. This serves the purpose until I go with copper blocks on both chips.

Copper is better than aluminium for watercooling, and should be used exclusively where possible (to keep corrosion down). Aluminium is ideal for dissipating heat (air cooling via heatsinks), but copper is ideal for conducting heat (chip to water). I use an aluminium radiator, located outside my CPU case, to dissipate the heat from the water system.

Another important thing, these blocks are solid metal and they weigh a lot, probably a pound or more for the copper block alone. You really have to be careful not to chip/crack the PPC chip under the weight, especially when connecting/disconnecting the water tubing. I recommend using a little bit of silicon adhesive to glue some small rubber spacers all around the PPC. Their height should allow them to squish just enough to allow solid contact under a little pressure, but still support most of the weight of the waterblock. There are watercooling specialty stores that sell thin copper spacers, shaped like hollow squares, to build up the entire chip top surface to one uniform height. All were sold for common PC processors though, and I don't know exactly how the 604e measures up size-wise. I think my solution is good too, and there were a lot of rubber grommets and gaskets at the local plumbing supply to dice up, at minimum cost.

I've also modified the mounting assembly since I took the pic, I used two slighlty unequal thickness rubber gaskets as spacers on the top of both waterblocks (instead of the temporary washers shown) to make the two heights identical. I also have since made the aluminium mounting bars exactly the same on both sides. The idea being to make the mounting assembly perfectly square, so as to evenly distribute the heavy weight over both processor surfaces. The rubber gaskets on top also keep the blocks from shifting during installation. The aluminium mounting bars are attached using two PPC fan clip holes between the PPC and the 68K, using thin bolts (shown) through thin nylon washers, which are under the card. Be very careful to keep anything on the underside of the card as low profile as possible, as chips on the Amiga motherboard can bump them just enough to interfere with a good CS-PPC to motherboard connection, and you know how sensitive these cards are. I couldn't figure out any other way to attach waterblocks to the card, but it all worked very well.

My next project will be to add a PPC peltier assembly to see how far I can push the PPC's MHz under supercooling. Peltiers only cost about $25-$35, and can produce sub-zero temperatures at the chips surface, but you really have to take care to insulate and seal off any open air contact with any cold surfaces, or water will condense and possibly short out the card. Another problem is that they produce tremendous heat on the hot side, but now that watercooling is in place that'll be a piece of cake to deal with.

I'm still researching how the 604e's surface mounted nature will affect this, I'm afraid the pins underneath will collect a tiny amount of water condensation, even if I seal the perimeter of the chip with silicon first, and then use foam gaskets around that. On the other hand, if a silicon seal around a surface mounted chip is plenty to keep additional air from bringing in water via humidity, then it would make the peltier solution extrememly easy for CS-PPC users. Problem is, most of the expertise thats been developed in the last few years pertains to common pc socket mounted processors. So, still some technical pondering going on, my gut impression is that it can be done rather easily, reliably, and very cheaply if a little care is taken.

For more information, contact Steve Giovenella,

Back to Amiga Page

Introduced August 17th 2001. Updated August 17th 2001. Version 1.0