In relation with my Mahogany Wooden Case project, it was time to make some changes on the watercooler as well. Not that the old one wasn't good, quite on the contrary, but there were a few things that I found really annoying about it. What was a really good plan turned into a grotesque mix of computer hardware, autoparts and fiberboards -- It had turned into what one of my lecturers used to refer to as "optical noise". When my first watercooler was published on my homepage, it got the name "The FrankenCooler", and from the picture below, you can see why:
Fig. 1 The FrankenCooler I
My goal with the new and improved cooler was to get a cooler that was more efficient, easier to do maintainance on and more aestetic. It is basically the same cooler as the old one, but with a few changes that makes a huge difference. With the new woodencase that was made with supercooling in mind, it was a whole lot easier to get it the way I really wanted instead of having to settle with compromise solutions all the time.
Like mentioned before, the starting point with this project was the new woodencase I made. In the picture below, you can see how the case is divided in two sections; one for the cooling(right) and one for the electronics(left). As you can see, there's plenty of space even for a freon compressor, so I'll never run into space problems, that's for sure.
Fig. 2 The case with the two sections.
I started out with the radiator. I quickly found that the easiest way of doing this would be to make a duct with the radiator and fans attatched to it. I know it looks a bit "ghetto" using cardboard and duct tape, but I didn't have the tools to make one in aluminum, which I really wanted to. Anyway, in the pictures below you'll see how I did it, and it should be pretty self explanatory:
Fig. 3 Fans and radiator mounted on dampening Aeroflex.
Fig. 4 The cardboard being taped on to the fans and radiator.
Fig. 5 The cardboard duct is done.
I started out taping both the radiator and the two Nidec 4.5" fans to a piece of Aeroflex (there's actually four Nidec fans in the system, the other two are used for casecooling). The Aeroflex work as a buffer and will prevent resonance from the fans to travel into the woodencase, thus making it very silent. I cut the cardboard to fit around the radiator and the fans and then wrapped duct tape around it. Fast, simple and effective.
Fig. 6 The fans are slightly angeled towards the center of the radiator.
On the picture above, taken from the back of the case, you can see how the fans are slightly angeled towards the center of the radiator, as the two fans are a little bit wider than the radiator. The duct took me about ten minutes to finish, and it's much more efficient than the insane fiberboard duct used in the first watercooler. The fans are running at 5V instead of 12V, and this make them run very silent, yet they still move an adequate ammount of air.
This was a new approach for making a waterblock. In stead of the standard crossdrilled, I decided to solder on a brassbox on top of a copper baseplase. Inside the waterblock there are lamells of thin copper which serve as turbulators and they also increase the surface area inside the block.
Fig. 7 The waterblock before lapping.
This is what it looks like in the beginning. The base needs to be lapped and the rest of the block needs to be painted. I chose high gloss Tamyia black modelpaint for this job. You want to apply the paint in layers, or you'll have problems with the paint dripping off and this doesn't look good.
Fig. 8 The waterblock after lapping.
Fig. 9 The waterblock painted.
Testing the cooler
In every project involving watercooling, it is very important to run it outside your comuter to check for leaks etc. Make no mistake; it is not fun to discover that there's about 10 litres of water gushing inside your computer.
Fig. 10 The watercooler being tested for leaks.
In the picture above you can see the test bench. The radiator is inside the computer case, but exctept for that, everything you need is in the pic; the Eheim 1048 pump, a 250W variable 12-24V powersupply for the peltiers, waterblock with two 90W peltiers, hoses and a red bottle that I've used as a reservoir.
Notice the two following pictures. These are before and after powering up the watercooler, and, as you see on the last picture, there are quite a bit of ice forming on the coldplate. The temerature measured on the coldplate with no load was aproximately -25C.
Fig. 11 Before firing up the cooler.
Fig. 12 After firing up the cooler.
Whenever you have a cooling system that go below your ambient temperatures, or actually the dewpoint in the room, there will be some condence. The dewpoint in a room with a temperature of 20C may vary from 13C to 17C depending on the relative humidity in the surrounding air. Needless to say, when operating at sub-zero temperatures, you will have serious problems if you don't insulate properly.
This is the third major change to the original cooler. I used Aeroflex in the first version as well, but the sandwich I made at that time was held together by two sheets of aluminum that was zip-tied together without any form of clamping device:
Fig. 13 The old insulation.
Even though this solution was more than adequate for the purpose, it didn't look all that good, and I was also certain that with a few changes, I could make the insulation more efficient, thus being able to achieve even lover temperatures. After a tip from Michiel Mitchel in Canada, I started looking for electronic project boxes, and I did find a box that would fit perfectly.The holes for the slotket were made using a saw, and for the fittings I used a drill.
Fig. 14 Materials used for the new CPU enclosure.
As mentioned earlier, I used Aeroflex for insulation. I cut the sheet so that it fit perfectly inside the projectbox, and then carefully cut out the holes for the CPU and the slotket, the waterblock and the coldplate. It is very important to make it a tight fit, or else you might run into condense problems. The Aerofles is very cheap (I bought the big sheet you can see in the pic below for about $8), but it's about the best you can get.
Fig. 15 Insulating the CPU.
Fig. 16 Detailed picture of the insulation.
Fig. 17 CPU enclosure is done.
Insulating around the slot and the back of the motherboard is also very important when you are dealing with so low temperatures. I cut the aeroflex to roughly fit around the capacitors and filled the holes with silicone to make everything airtight
Fig. 18 Insulating the Slot1 connector and backside of the motherboard..
Fig. 19 CPU enclosure mounted.
The last picture show you the whole combo mounted and ready to run. The white cord is for the thermometer mounted on the front of my case, and the other cords are for powering the two 90W TECs.
With a coolant at 25C, 5 degrees above the ambient temperature, I reached about -10C under max load on the CPU. This is where the variable powersupply comes in. These temps were without having tweaked anything at all, it was just "plug and play". After adjusting the voltage to the peltiers I ended up at -18C, 8 degrees lower than before.
The reason to this is that peltiers have an ideal curve: you can increase the voltage to a certain point and still get better cooling, but when you pass that sweet-spot, it will start producing more heat than chill, i.e. the efficiency decreases. What you want to do is to find the sweet-spot and stay there.
Fig. 20 The digital thermometer.
I'm sure you're all waiting to see some benchmarks, so just click your way to the next page.
I started out with a Celeron II 566 and bumped it up to 1GHz right away, as I knew it would handle that. The corevoltage was 1.75V. But I wanted more...much more, so I increased the corevoltage to 2.1V and started increasing the FSB, step by step all the way up to 133MHz. It ran flawlessly at 1100MHz, but I had to get a 100% overclock, right?
Well, at 1133MHz, the CPU ran for a few hours at full load, and then just stopped. It was dead. I suspect that this was a case of, ironically, electromigration, and because of that I don't have any screenshots with the CPU above 1GHz.
Fig. 21 SiSoft Sandra CPU benchmark for the Celeron 566.
Fig. 22 SiSoft Sandra CPU multimedia benchmark for the Celeron 566.
Fig. 23 SiSoft Sandra memory benchmark for the Celeron 566.
As a replacement for the 566, I got a Celeron 700ES (Engineering Sample) that was pretested to 1050MHz with a retail Intel cooler, so there was a potential for a 1.2 - 1.3GHz overclock here. Strangely enough, this CPU wouldn't overclock to anything higher than 1050, no matter what I did. Very frustrating indeed.
Fig. 24 WcpuID screenshot of the Celeron 700ES.
Fig. 25 SiSoft Sandra CPU benchmark for the Celeron 700ES.
Fig. 26 SiSoft Sandra CPU multimedia benchmark for the Celeron 700ES.
Despite a few mishaps, like a burnt-off CPU, I think this project was successful. I managed to make the cooler much easier to mount, easier to maintain, and much more silent. The latter is in particular importance to me, as I'm woking a lot on my computer. I hate noisy fans, and I find it hard to concentrate with fans roaring all the time.
I have also increased the coolers efficiency by making the insulation better, it gave me in fact a 3 degree improvement compared to the old cooler. With the new CPU enclosure, it's also much easier to replace the CPU, and it looks much more professional as well.
Please leave any comments or questions in the forums
For those of you that aren't tired of reading, I suggest you check out these links below.
The original FrankenCooler.
Article about electromigration and overclocking.
The article about my mahogany wooden case.
Lots of great info here.
Steves homepage. You can hardly get more extreme than this.
Great sources for peltier info!