I’ve been watercooling for 2 and a half years now. My first watercooled rig is an 80 watt peltier cooled by a cross-drilled waterblock on an Intel P3 600E @ 900mhz with a full load temp of 15C. When I went onto Thunderbirds, the cross-drilled
waterblock followed but I abandoned the 80 watt peltier because it was insufficient to keep up with the hot CPUs.
The block on the right is the cross-drilled block I used to use, seen here without the brass barbs.
From the cross-drilled waterblock, I
went onto a Danger Den Maze2 and then a home-made waterblock with 2 inlets, cooled
with my first evaporative cooler. This rig was cooling a Thunderbird 1000mhz
AXIA "B" chip which was overclocked to 1600mhz at 2.33v.
This was my previous rig and you can see more details of it here.
I ran this rig for about half a year before deciding to make another cooling rig upgrade. There weren’t many choices left. I didn’t think a waterblock upgrade was going to help much, since the block was besting a Danger Den Maze2. Therefore my next step would have to be some sort of supercooling that could get temperature of my CPU well below ambient.
I considered a few options: I could go with a phase-change system with a compressor ripped from an old fridge or air conditioner. Or I could go with direct peltier-cooling, or with peltier chilled
water, or a combination of both.
I decided on direct peltier-cooling. This was because I did not have the expertise on making a phase-change system as it required some skills in setting up and I don’t have enough confidence in doing a
good job. Water-chilling isn’t as efficient as a direct peltier cooled CPU, so I decided to go with direct peltier. I could add a water chiller later if I wanted to.
Next, I had to decide on the peltier that I was going to use. A highly overclocked Thunderbird at high voltages can put out near to 140-150 watts of heat. Clearly, a 156 watt peltier wasn’t going to cut it. A 172 watt peltier was pushing it also. Therefore, I decided to run 2 172 watt peltiers side by side.
This should keep a highly overclocked Thunderbird at a cool temperature. Next, I had to acquire the equipment to make this plan a
reality. Some of the pictures are pretty low quality due to the bad focus my camera gives me, so please bear with them.
The first thing that I contemplated was the waterblock that I was going to use. My previous waterblock was certainly out because of its design and the 4 holes. The only commercial waterblock designed to house dual peltiers is the Danger Den Maze2-2. I am a lover of high flow-rate and with dual peltiers, it was even more important to have fast flow.
Even though the new Maze2-2 has improved flow rate, I wanted something even faster. There was nothing that fitted this
out there, so I decided to make it. I had to come up with a design for the waterblock. Something with low flow rate cut and huge surface area would be good, something in the style of heater cores that perform so well compared to winding radiators. This is what I came up with:
It’s basically a watersink that is multichanelled. I decided to go with 5/8" fittings for this, firstly because my pump has a 5/8" output and I was confident in coming up with a 5/8" fitted evaporative cooler or radiator setup. Thus, the smallest fitting in my loop would be 5/8" and this would surely help flow rate.
This is the design of the top plate to cover the milled block. It measures 80 x 50 x 3. All measurements are in mm.
The hose barbs are without screw threads and they were to go into the plate and soldered underneath. This would make it look neater compared to if it was soldered on the outside.
This is the design of my coldplate. It measures 80 x 50 x 6.5.
Side A (The side facing CPU)
Note that the 6 holes for the screws on the cold plate are counter sunk for easier insulation.
I decided to use a 6.5mm thick cold plate because anything thicker wasn’t going to improve heat dissipation by much (according to this
article) and was going to add a lot of weight to my block. To get enough clamping pressure on the peltiers, I went with 6 screws instead of the usual 4 screws like in the Maze2-2. However, after the block was made, I realized that having the 2 center screws would make it impossible to hold a 50x50mm peltier.
This waterblock coldplate design can hold 2 40x40mm peltiers (such as
the 120 and 172watt peltiers). If you want to make one that can hold a 50x50mm peltier, then don’t incorporate the 2 center screws. Some people have said screwing down the Maze2 coldplate causes it to warp. With 2 screws in the middle, it is certainly not happening to mine.
I was going to use the 4 holes mounting technique and thus I had to incorporate 4 bolts to the coldplate. Thus, the side of the coldplate facing the CPU had to be drilled to around 4mm deep and tapped so that 4 bolts can be screwed in.
This is the design of the waterblock itself. It measures 80 x 50 x 19.
Side A (The side facing the hotside)
I decided that the base was to be milled to 2mm deep because I’ve had better temperatures with a thinner base plate before. Some people argue that a thicker base plate is better but this is not what I found from my own experience. Also, with the same thickness of block, a thinner base would mean higher surface area.
Special thanks to weazel666 from the
tekforums for helping me with the graphics.
Now that the design was on paper, I went shopping for the parts to make them. Here are the materials to make the block and coldplate.
These materials cost me around US$6. Together with the design, I passed them over to a machinist who helped me make my block. I did not do it myself this time because I wanted to save myself some time and effort.
While I was waiting for the waterblock to be made, I was rather worried that he might get some parts wrong, especially the distance of the 6 screws from the edge of the block and I would end up with a block that was unable to house the 2 peltiers. Thank God the end product was exactly as I had wanted. I paid him about US$27 for his service – well worth it, I feel. Here’s the picture of it before I polished it up. Beside it is my previous waterblock:
Here are some more pictures of it.
Here it is with the 2 peltiers with a layer of Arctic Silver II on both hot side and cold side of the peltiers:
Before I screwed the cold plate onto the waterblock, I dipped the metal screws into some silicone grease to prevent thermal leak between the cold plate and the block. I also cut some insulation tape to act as insulative washers before I screwed the metal screws into the counter-sunk cold plate.
I fitted the cold plate with nylon bolts instead of metal ones because I didn’t want the cold plate to lose its coldness to the bolts. Also, if I had used metal bolts, I would have to insulate them from condensation. From my experience, nylon bolts can offer strong retention power if they are used with tough springs and washers.
Yes, they bend easily, but the weight of the block and cold plate should not be resting on the bolts but on the springs, washers
and the screw threads.
Now, I had to decide on the setup to cool the water. I could use my bong, as evaporative coolers can take care of the huge heat load the 2 peltiers and CPU put out.
However, with the high water temps, evaporative rate is going to be very high and the constant refills could become a chore in the long run. Therefore I decided to go with a radiator setup. However, not many radiators can cool the water effectively and the big ones were going to cut down flow rate a lot.
Thus, I decided to go with a dual loop setup since I have an extra pump collecting dust. One loop will be the main waterblock loop on a 3000L/hr pump and the other will be a loop cycling reservoir water on a 1200L/hr pump. I decided to go with a multi radiator setup this time. Here’s the 3000L/hr pump I was going to use, modded to be inline with some plumbers goop and a hose barb.
Here’s the 1200L/hr submersible pump also modded to be inline with plumbers goop and hose barb, so that the pumps do not heat up the water much with their heat:
Time to pick the radiators I was going to use.
I was using this dual loop setup to cool my Thunderbird 1333 at 1780 MHz, AYHJA at 2.16V and water temperatures stabilized at 3-4C above ambient. I burned in with CPU K7Burn program running my chip at 1780 MHz, 2.25V with the 2 172 watt peltiers at 20.5V and the water temperature rose to 5-6C above ambient.
I wanted to lower the temperature of the water and, in so doing up, the peltier voltage to try to get a drop in CPU temperature. Therefore I did what most people think is an overkill: I added another loop.
This is the 2200L/hr pump I used for the third loop, modded to be inline as well with some plumber’s goop and JB weld like substance. The reason why my pumps are all submersible is because they cost much less than pumps that are able to work inline. I bought the 3000L/hr pump for US$25, the 1200L/hr pump for US$13 and the 2200L/hr pump for US$16.
Here’s my collection of radiators:
Most of them are used and I got them for free from a neighbour who deals with auto parts. I can just visit his shop and get 2-3 radiators per week. Some of them I bought from motorbike shops for cheap. Look at the big black one here:
It can hold 8 120mm fans with some room left for a shroud. However, it is not the multichanelled heater core type but a flat-tubed winder, so I gave it a pass as flow rates would be hugely impacted.
This is one of the radiators I’m using on my third loop. It’s designed in the style of heater cores with flat parallel tubes.
This is the other heater core I was using on my third loop. It developed a leak one day at the copper elbow epoxy connection, so I removed it.
Fitted with 2 strong 120mm fans, it cools very well. Here’s the flat-tubed winder I chose for my second loop:
Its a little old but still works very well. Here are some of the motor bike radiators I bought.
They are multichanelled, fit a 120 mm fan well and would fit nicely on top of a casing if one wanted an in-casing watercooled setup. The best thing about these is that they are very flow-rate friendly and flow rate is hardly affected at all as water goes through them.
I decided to use one in combination with the flat-tubed winder for my secondary loop. Another one, I used on my primary loop. Here’s my favorite heater core that I decided to use on my primary loop, the
The amount of surface area in it is huge and the fins are nicely packed to expand surface area. As this heater core is rather thick, I decided to use 2 120 mm fans on it, one for intake, the other for output in the style of a wind tunnel. Here’s the secondary loop. The 2 radiators were connected with some hoses and a copper 90 degree elbow.
Here’s the third loop before one of the heater cores had a leak:
I used a total of 7 120mm fans, 5 radiators and 3 pumps. Overkill perhaps, but that’s the extent I go to for 2-3C. The fans on the secondary loop are not so strong – I just used the 120mm fans I had lying around. All of them are shrouded to the radiators. Most of the shrouds are made from baking trays while two are made from shoe boxes. The flow rate through the waterblock was very good.
My radiators are connected in series. Some people feel that connecting radiators in series isn’t as effective as connecting them in parallel because firstly in parallel, the temperature of the water entering both radiators would be equally high and thus heat transfer would be faster.
In series, heat transfer after the first radiator would be slower, since the water temp is already lowered. Secondly, flow rate is said to increase with parallel. However, because of the many "Y" connectors that I have to use if I had wanted
to go parallel, the setup would get more complicated – hose bending and elbows would have to increase. Thus, I decided to stick with series.
In the later part of testing when the rig was up, I found that the secondary loop really helped. When I turned off the pump for the secondary loop, water temps rose by 3C within 30 minutes. The third loop which was added later on also helped. Also, the air from the secondary loop was very warm and slow so I suspect that water temps may improve if I replace the fans with stronger fans.
I think this setup can cool at least another 172 watt peltier, perhaps even more. Almost my whole desk is taken up by my cooling equipment but that’s OK because I don’t use it anyway.
I decided to use a rather big reservoir so that the hot water from the block does not mix easily with the intake before it goes through the recycling loop. I drilled 2 holes in the tupperware that I chose and fitted them with hose barbs. I used a rubber O-ring in between the hose barb that was screwed to a PVC end cap.
This will ensure that the reservoir does not leak. The other hose barb was plumbers gooped because I ran out of 1/2" PVC end cap. The cover is also done in the same way using hose barbs and PVC end caps.
Before I set up the rig on my PC, I did a dry run to test the efficiency of the cooling rig. Within 2 minutes, the cold plate temperature dropped to -15C and it was continuing to drop slowly.
Notice the frost on the cold plate. Cool. It was time to insulate my equipment.
I had some experience in insulation from my previous peltier cooled P3, but that was on a slotket. It is a very important step in peltier cooling your CPU and not only helps prevent condensation but also helps your temperatures as you do not lose the cold to the surroundings.
The basic rule is to eliminate airpockets and to coat transistors etc, with silicone. For a nice walkthrough,
you can read this article.
First thing I did was to insulate the CPU itself. My previous CPU was an AXIA 1000 MHz Thunderbird. I sold this away and bought a 1333 MHz AYHJA chip. The funny thing is that I bought it at the same price I sold mine. Anyway, it came with the core lapped by the previous owner (totally bare core without any markings) and the 4 rubber feet removed.
I inspected the core and not satisfied with its
flatness, so I lapped it a little more with some 1200 grit sandpaper. Next, I coated the bridges with a thin layer of silicone:
This chip came factory unlocked, so I didn’t have to worry about the L1 bridges. I also coated all the metal resistors on the chip with silicone. Next, I taped the surrounding area with layers of insulation tape until near the height of the core.
This will reduce the amount of air-pockets around the core and act as a shim
so that my core does not get chipped without the 4 feet. It really does work
because I’ve inspected my core and found no chips even when I applied great and
uneven pressure on the waterblock. Without it, I’m pretty sure I would have
ended up with a dead chip.
My thermal probe was taped on the ceramic
plate of the CPU. It won’t give me accurate CPU temperatures, but it will let me
know if my peltiers are working every time I start up the machine. I usually wait
for the probe to read 0C before starting up the machine. I have never even
tested this chip before. This would be the first time I was going to run it.
Next, I went on to insulate my motherboard. It is an Abit KT7A Raid. First, I coated the PCB around the socket with silicone. I flipped it over and coated the back of the motherboard near the CPU area with silicone also.
I acquired slabs of neoprene around 15 mm thick from a friend who deals with air condtioners.
I used a paper template to cut out a piece of neoprene that fitted snugly over the socket.
I coated the center of the socket with silicone, using a tape to cover my socket holes and the motherboard’s thermal probe so that they do not get siliconed accidentally. I then cut a piece of neoprene to fit over the center socket hole with a slit for the thermal probe to stick out.
I bent this probe upwards and coated it with some thermal grease. This should give me a rather accurate reading of CPU core temperature as it was giving me readings very close to external probe when I was using straight water cooling.
Now we come to the controversial part: What can we use to pack the socket pin holes? Some people have used Vaseline without any problems. I have tried both Vaseline and silicone grease.
I found that Vaseline turns green when I took out the CPU.
Condensation perhaps, but another thing to note is that vaseline itself contains water. I also found a few pins near the ceramic plate had turned black. Finally, I went searching around for some dielectric grease and the closest I could find was Castrol’s Moly grease, which was black in color and supposedly dielectric.
The thing that drew me to buy it was because it offers excellent protection against corrosion, something the gold-plated pins of the CPU needed. I took off the socket holder and packed the holes with Moly grease.
It was more liquidy than the silicone grease I had and so it filled the holes easily. I also coated the pins with the Moly grease with a brush. I slapped the CPU on and some excess Moly grease got squished out. Good!
Now that this was settled, I went onto apply silicone to the area around the Zif socket to block out any air holes. I also coated the metal handle of the socket with silicone. The raised platform of the socket at one side was filed down as this area is a little higher than my core and in case any part of the cold plate or its insulation went over to this area, the block would still sit flat on the core.
Next, it was time to insulate the coldplate.
First, I filled in the space between cold plate and waterblock with
silicone. It is not as permanent as it seems because the clear silicone I used can be easily cut and peeled off. I also coated the sides of the cold plate with
silicone. Although it makes my block look ugly, it serves as a very good prevention against condensation.
Next, I wrapped a layer of neoprene around this:
The cold plate face around the CPU core was also coated with a thin layer of silicone. I then slipped 2 pieces of neoprene into the nylon bolts. In this picture, the two pieces of neoprene are sandwiched between the cold plate and the main socket neoprene:
Time to mount this onto my CPU.
The block and cold plate was to be held down by a screw clamp and 4 nylon bolts with springs. This way of clamping works great because you distribute the weight of the block and cold plate among the 2 hold-downs; with the metal thumbscrew over the center of the CPU, it should sit properly on the core. Here is the hold-down I made with a metal thumb screw, brass rod and an aluminum bar:
It uses 4 lugs of the socket. Due to one time of over-tightening, I broke one lug in addition to another one I broke a long time ago, so now it’s only held down by the 4 bolts. For more details on how to make it, refer to this
Before I mounted the cold plate block onto the CPU, I filled up the
airpockets near the core with thermal grease and with silicone grease further out. Thermal grease was used near the core in case the grease got squished over the core, there would still be good thermal transfer. Other airpockets around socket were filled up with silicone grease.
A rather thick and even layer of Arctic Silver II was spread over the core. Some have said that Arctic Silver II changes or foils at low temperatures. This has not been confirmed yet and its performance at low temperatures is still rather unknown. With the neoprene in place, I mounted the block, screwing down on the block with the screw clamp.
I took it out once to see if there was a nice imprint of Arctic Silver on the cold plate – there was. After the block was clamped down, I inserted a large piece of neoprene with 4 holes behind the motherboard. A hole had to be cut in the motherboard chassis of course. Next I put in an acrylic plate through the 4 nylon bolts that I made to squeeze the neoprene against the back of the motherboard.
This is the acrylic sheet I made:
It was held in place with washers, springs and nuts. I turned the nuts until the springs were very compressed and the neoprene was squeezed very tightly. Good. There should be lots of hold strength and pressure. I used a tape to tighten the squeeze of the neoprene; I also added some more neoprene around, held with zip ties.
Finally, I set it up on my PC. Here are some pics of the system running:
I waited for the thermal probe to read -10C before I fired up my PC. This was the first time I was running this chip after I had bought it, so I had to find some time to find its sweet spot. The probe shows this after a while – the left hand probe shows water
temperature before the waterblock and the right-hand probe shows the CPU ceramic plate temperature:
At default voltage and speed, it idles well below zero, but who runs them this way? These are the temps and overclock I got with the AYHJA 1333 MHz Thunderbird:
23C at full load, 1790 MHz at 2.25V, with peltiers running at 22V each and water temperatures 3-4C above ambient. At this setting, the CPU is producing about 150 watts of heat. Ambient is around 28C in an air-conditioned room – the heat output is heating up my room! Not too bad!
I have to get myself a power supply that can give me a
better 5V line. Maybe then I can get it to run stable at 1.8 GHz.
I’ve run my CPU at different settings to analyse the performance of my peltier rig. It makes it possible to compare the performance of my peltier rig with others. It also gives me an estimate of the temperature I would get if I were to switch to an Athlon XP.
Delta A refers to the difference in temperature between ambient and CPU temp. Delta W refers to the difference between water
temp and CPU temp. The heat output is estimated using this program. Tests were ran at different settings running Prime 95 for around an hour each
and water and CPU temps have stabilized. Ambient is 28C.
Water temps have hardly risen throughout the different heat output,
indicating that my radiator rig is performing well. If I were to be using an Athlon XP at 1800 MHz at 1.80v (70watts) at an
ambient of 22C, my CPU temp would be roughly -6C at full load.
If I were to run an Athlon XP at 2044 MHz at 1.84v (80watts) at an ambient of 22C, my CPU temp would be roughly -4C at full load.
Pretty vapochill-like results eh?
Peltiers can give good results on the high heat output AMD CPUs as they did with the Intel CPUs. For a substantial drop in temperatures, dual peltiers are highly recommended. I’m very pleased with the results the dual peltiers are giving me. For those deciding between one 172 watt peltier or two, I would advise to choose the latter if you have enough cooling power for the hot-side.
There are still a few things that I can improve on, like connecting the radiators in parallel andchanging the weaker fans on the radiators to stronger ones. The rig looks very ghetto, but I’ve always been more of a performance kind of guy, so it’s OK with me.
This cooling rig is certainly following me when I
switch to an Athlon XP on a DDR platform. Once I do, I will be sharing the results with you guys. Until then, Cya!