Project Megachill

“That brought my rig up to 4 peltiers, 6 radiators, 4 pumps and 9 fans.” Extreme cooling – Ji Kuang

Hi all! Its that time of the year again, the urge to get better cooling for my PC hits me once again.

Perhaps some of you may remember my previous cooling rig of dual 172 watt peltiers on my Thunderbird 1333. You can see the article HERE.

I’ve gotten some mails since then regarding the insaneness of it. Well I guess its definitely not something for everyone, but I’m pretty satisfied with it all. The ghettoness of it, especially the yellow duct tape, gets mentioned a lot, and I agree, but looking good has never been any part of my objectives. More effort could be spent to make it look better, but that would mean less time and effort to spend on performance, so I choose to concentrate on the latter.

Just a few months back, I switched over from the Thunderbird on an SDRAM board to an Athlon XP on a DDR board. Being the bargain seeker, I bought the InnoBD 7300D KT266A motherboard for around US$80. HERE

Next I found a seller of an Athlon XP 1600+ for around US$100. The seller is the second owner of this chip and that makes me the third. The first owner had already unlocked it so its less trouble for me. From what the first owner said, it could do 1900 Mhz, 2V watercooled. Its an AGOGA “0202 Y”. Neat chip I thought so I bought it.

After I checked that everything worked fine, I went on to do the necessary condensation prevention on the CPU and motherboard in preparation for the residence of the dual 172 watt peltiers. On the XP chip, I coated all the bridges and transistors behind the core with nail varnish, so that condensation would not short anything.

I also painstakingly made an insulation tape shim that prevents chipping and reduces air-pockets, like I did with the Thunderbird. Its basically just layers of tape to near the height of the core. I prefer it over a neoprene shim because firstly, neoprene expands sideways as you compress it and I might have some of it accidentally covering my core. Also, it exerts a force against my mounting and I might not get good contact or even worse, the lack of any contact.

The tape shim works great for me and its dead fixed on the chip so I don’t have to worry about it moving as I mount my block. A probe is installed on the CPU ceramic plate that would give me assurance that my peltiers are working every time I switch on the PC. The rest of the air pockets were filled with thermal grease and silicone grease.

On the motherboard, I did the standard procedure of applying silicone behind the board and around the Ziff socket. I also applied silicone on the transistors inside of the socket. For the raised side of the socket, I filed it down so that even if the block accidentally overlaps onto this area, there will not be bad contact.

I packed the pin holes with Castrol Moly dielectric grease and also coated the pins with it with a brush. I cut a piece of neoprene to cover the surroundings of the socket. Its basically a repetition of what I did in my previous article.

For the mounting aspect, I was able to use the central hold down mechanism once again with this new board. This was used in addition to 4 nylon bolts and springs.


It serves to centralize the pressure on the core so I can afford not to be so precise with the 4 nuts. Without it, I had to tweak each of the nuts bit by bit to get the lowest temperature. Not as ideal. To prevent me from breaking any lugs from over-tightening this time, I used a spring below the thumbscrew.

After mounting the block + colldplate + peltiers, I removed it again to check for Arctic Silver 3 imprint. This gives me assurance that there is good contact. Still, it gives me a heart attack every time I switch on the PC for the first time after a fresh install. Its very easy to have no contact with the chip and end up with a fried chip.



Here you see it mounted on the board in the case. I also changed the wire connectors from molex to the screw on type as the molex can’t take the high current and became blackened. My board was also modded to give a higher Vcore voltage as it was limited to 1.85v. I have it up to 2.3v. My DDR voltage was also modded with a variable resistor so I can overclock my ram better than the 2.7v limit of this board.

The board is a little troublesome to use as it is largely dipswitch based. Only the core voltage and FSB is adjustable through the bios, the multiplier and DDR voltage has to be adjusted through the dipswitches on board. After playing about with FSB, voltage and multipliers, I got a good speed of 2079 Mhz that can run prime 95.


It runs a load temp of 21C, 2079Mhz 2.18v with 31C radiator intake air, and 34C water and 2 172 watt peltiers at 22V each. From what I’ve seen, it beats a Vapochill in terms of performance.
The max I could push it was 2135 Mhz


However, it was not stable. Still, I was very satisfied with what I got.

Now, I was limited to a stable FSB of around 166 Mhz. I think it was either my ram or my board. I am using Samsung PC2700 third party ram. My NorthBridge area was crammed with capacitors so super cooling that part would be difficult.

All I did was apply Arctic Silver 3 on it and increased the pressure of the heatsink on it by placing plastic spacers in between the mounting pin. I also changed the fan on it to a bigger one. Now, to cool my ram to try to improve my FSB overclock. I’ve always been interested to find out what super-cooling ram would give so I decided to give it a shot.

I remembered reading an article somewhere that rams are the same kind of devices as CPUs and performance improves as temperature drops. On top of that, it seems that those people with lower ambient temperature seem to get better speed from their ram so temperature does matter.

Well now I could just buy myself some new ram, but for the challenge, I decided to chill my ram. I would not choose just cooling the ram with a ram waterblock because it doesn’t get so hot so watercooling it would not be much different from using ramsinks. I had a 65 watt peltier waiting to be used so it was commissioned with the duty of chilling my ram.


As my RAM is single-sided, it would be easier to peltier cool it. The peltier is to be cooled by water of course. The question is which waterblock to use. For a light rig so that my ram holders don’t break under the weight of the block, I decided to use an aluminum block. I could custom make one, but I had a better idea.

There is an aluminum block that caught attention locally due to its cheap price. Together with a pump, it was sold for US$25. Now that is an amazingly cheap deal. Here’s the block.


The brand name is Elen and it’s a Chinese company.

Well I got hold of a secondhand block from a guy for just US$8. Can’t beat that price. The first thing I did when I got the block was to remove the quick connects that take a hose size of a whooping ¼”.


Next, I screwed on two 3/8″ hose barbs.



Ji Kuang

As you can see, the side caps are epoxied on. To ensure leak free operation, I applied JB weld here.


This should cool the 65 watt peltier well. But what’s going to spread the cold onto the ram chips? I needed an aluminum cold spreader. I bought some aluminum sheets and plates. When clamping the plate with the block, it needs to be rather thick to avoid the cold plate from bowing.

However, a thick plate is not ideal as the ram chips stretches across some distance and 8 of these chip will probably have slightly different heights. The plate should be able to bend a bit to accommodate the difference and maintain good solid contact. So the cold plate is to be made up of 2 parts a thick part for the clamping area and thin part elsewhere.


To bond them together, I used Arctic Silver Epoxy. I used a F-clamp to clamp the 2 pieces together as the epoxy cures.


The excess area of the plate was to be trimmed.


It is all good and well that the Elen block comes with 4 pretapped screw holes, for whatever purpose I do not know.


So I drilled 4 holes on my cold plate and had one side countersunk.



The side facing the ram was lapped afterwards.



I often use JB weld to reinforce the peltier wire joints.



Ji Kuang

Next up would be mounting the dude onto the ram. Here’s the mini C-clamp that was custom made to hold the waterblock onto the ram.



Thick perspex was used because I wanted something light and couldn’t find any PVC around. It is fairly strong, strong enough for what I need it for.
This was for the clamping force on the center, at the block. The 2 sides I needed clamps as well so that there is good contact with the ram chips. Here are the clips I considered using.



Domestic stuff comes in handy sometimes.
Here I did a test clamp.



Next of course would be insulating the ram. This was crucial as the ram was not going to produce much heat and temperature would thus be freezing.
I coated almost everything on the ram with nail varnish, especially the pins of the ram chips.


For the RAM chiller side, its basically silicone and neoprene on the parts that were going to be exposed to air. It was easy to get the neoprene to stick well by sticking them before the silicone cured.




Ji Kuang

Before clamping on the ram chiller, I applied Arctic silver 3 on the ram chips, and also removed the stickers on them.



After clamping on the chiller, I squeezed silicone grease into the air spaces through the tiny openings and then covered with more neoprene.


Good seal I hope. Now I want to go onto the other side of the cooling rig, the video card.

I acquired a Radeon 8500 AGP recently, and decided to use an 80 watt peltier to cool the GPU. Before operating on the GPU, I put on some ramsinks with Arctic Silver epoxy. I covered the ram pins with nail varnish in case anything got shorted.



Now what was I going to use to cool the 80 watt peltier? I decided to use a cross-drilled copper block I have to cool the peltier.


A cold plate was used not just because I want to get the cooling from the whole peltier, but more so to raise the block above the GPU so that the hose barbs would clear the ramsinks.



The block is fitted with 3/8″ fittings.

Now this is a very heavy mass of copper. I needed to clamp it onto my GPU. The big drawback of the Radeon 8500 is the lack of holes for the mounting of heatsinks or waterblocks. So a custom made G clamp is in order,




Ji Kuang

I tested the strength of the clamping device on a table first.


On the block side, condensation prevention consisted of the usual silicone and neoprene.



On the Radeon side, I coated the surrounding areas with nail varnish. The core was lapped a little after the heatsink was removed.


One important area to watch out for is the GPU platform. If you notice carefully, the GPU chip is raised a bit above the PCB, and in between, there are metal connectors below. You have to seal up this area! I used silicone grease to fill the air gaps and silicone to seal around the perimeter. Next a layer of silicone grease, then a very thin neoprene shim just to take care of the airpockets.


Arctic silver 3 on the core. Behind the card, I coated the area with nail varnish, then a layer of silicone.


Then a piece of neoprene and a metal plate. The neoprene both serves as cushion and prevents condensation.


And then the mounting of the block onto the card.




The clamp eats up 3 PCI slots leaving me with 2. Just enough for my Network card and soundcard.

Ji Kuang

Next, before adding the new cooling to my present rig, I decided to give them a test run. The ram block and GPU block were to be run off a separate additional loop with a separate reservoir because it would make things easier for me when I need to change anything.

The radiator I picked for this loop is this radiator I had.




The test run on a 1500L/hr submersible pump.




The peltiers were receiving around 12V each from two AT power supplies.

Thick frost on the GPU side after a while.



The probe read -0.1C.

On the ram chiller side, its frosty as well, though losing a lot of cold as its completely uninsulated.


1.5C on this side. The radiator was used without fans and water temp raised to 40C. I was pretty pleased.
Now it was time to mount the cooling onto my rig.

Ji Kuang

On goes the RAM chiller.




The 2 metal clips help clamp the chiller onto the ram and also clip onto the ram holders to ensure maximum support.



And then the Radeon…



An inline thermal probe was installed to monitor water temperature just before the ram block.


And my home-made reservoir.


The pump is submersed and motorcar coolant was added.

Everything hooked up:






Ji Kuang

And two 120mm fans cooling the radiator with a shroud made from an aluminum sheet.


That brought my rig up to 4 peltiers, 6 radiators, 4 pumps and 9 fans. The loops go like this:

  1. Reservoir 1 to 3000L/hr pump to 2 radiators in series to CPU block to reservoir 1.
  2. Reservoir 1 to 1200L/hr pump to 2 radiators in series to reservoir 1.
  3. Reservoir 1 to 2200L/hr pump to 1 radiator to reservoir 1.
  4. Reservoir 2 to 1500L/hr pump to 1 radiator to ram block to GPU block to reservoir 2.

The water temperature for the CPU side is typically 3C above ambient while the water temperature for the ram and GPU side is typically 2C above ambient.

So what did it get me? Well, I found out that its definitely my motherboard limiting me as the ram chilling did not make any difference. The ram was really cold as the metal clips were cold even being separated by silicone and neoprene.

I do not believe that it does not let me run my ram 1 Mhz more. I’m still with the belief that ram chilling makes a difference like a CPU and hope that some memory experts/makers would step up to answer this question that I failed to answer due to a difficult motherboard.

Since the ram chilling was not making a difference for me, I removed it from the loop and installed just a fan to cool the ram. This will give my GPU side improved flow rate and slightly colder water.



And so, on my Radeon, I was able to overclock it to 340 Mhz core, 1.86 core voltage modded by pencil, and 340 Mhz memory, 3.98 mem voltage. And here’s a 3Dmark I benched, 11170.


Mad Onion

And here are some final pictures.






Alright now I know some of you may be thinking I could have gotten a better motherboard with the money I spent. Well, that’s not the whole point of me doing all the things I did, I am more interested in trying out things not commonly done before like ram chilling. And I often get asked how much I spent on my cooling rig. Here’s a rough estimate:

  • Peltiers ~ $130
  • Power supplies ~ $90
  • Blocks/cold plates/mounting stuff ~ $60
  • Pumps ~ $60
  • Radiators ~ all free except two ~ $30
  • Fans ~ $45
  • Miscellaneous ~ $40
  • Total ~ $455
  • Man hours involved ~ many late nights

Another typical response is, “Dude, you must own a power plant!” Heh, I do use up quite some electricity and it adds up to about $30 more per month. The rig is not 24/7 but almost. Again I will stress that this isn’t about looking good as you can tell from my yellowed 5 years casing that I don’t really care how my machine looks, as long as it is a good performer.

Am I insane? Well, to most people! And that’s my latest cooling gizmo! Cya!

Ji Kuang

Be the first to comment

Leave a Reply