SUMMARY: The Swiftech waterblock is the best system I have seen for water cooled peltiers; pricey, but it works well.
Swiftech was kind enough to send a sample of their MCW2000 water cooled peltier system. It is extremely well built (pic below) and comes complete with the necessary foam to keep moisture out. The block uses compression fittings so that you can use copper tubing; for this test, I used plastic hose that fit over these fittings; no problems with water leaks.
Keeping moisture out of the socket is key – pictured below is the foam insert in the middle of the socket. Silicone is laid into the bottom of the socket and also around all four sides.
Shown below is the flip chip cold plate – it is a machined copper block that screws into the aluminum water block. The foam is in three layers and cut exactly to fit the slotket.
This shows the flip chip foam layers on the slotket. The white areas are the protective paper over the foam adhesive – once you set it in place, they are not designed to come apart.
Condensation is a real concern with any below-ambient setup. As a test of the Switftech unit’s ability to exclude moisture, I ran the peltier on the water block without any CPU in the slotket. This gets maximum cooling at the peltier; if there is going to be a moisture problem, running it like this should show it up.
I ran it for an hour, taking measurements on various parts of the system and found the following temps:
- Ambient temp 30 C, humidity 65%
- Back Plate: 23.2 C
- Block Front: 35.1 C
- Slotket Pins: 20.0 C
- Radiator temp inlet: 31.3 C
- Radiator temp outlet: 29.2 C
I then disassembled the unit and checked for water inside – NADA; not a hint of it. Now this is about as extreme a test as you can get, so I would have to say that the insulation package that comes with the Swiftech unit does the job very well.
One thing I would be concerned about is the temp of the slotket pins – more than 10 C below ambient, and moisture in the slot could be a problem. Additional insulation at the joint between slot and water block would not hurt.
For the “live” test, I used a CeleMine 533A on my ABIT BX6-2. I wanted to see if I could break 1 GHz with this setup. I used the Swiftech with the Celeron sized cold plate as I did not want to rip off the foam sandwich to use the flip chip arrangement.
Assembling the unit is a snap; Put it all together and squeeze the back plate onto the water block so that you can feel the CPU contact the cold plate. Screw it down (six screws), and you’re set to go. For this setup, I had to remove my power supply from the case as the nipples on the water block would not fit under the power supply.
I wound up with quite a jury-rig; the power supply rested on a box, the fan on top of the case and the water pump container on top of another case. I used a separate AT power supply for the peltier to give maximum cooling; I use PC Power and Cooling Silent 275s, so running the peltier off this unit would not be a problem. Not pretty, but OK as a test bed.
The water block is fairly large but did fit in the slot brackets. For long term use, more secure mounting is needed.
Live Test Results
No question as to the cooling power of the 72 watt peltier used here; as shown below, temps at 993 MHz, 1.75 volts, were quite cool as you can see from the pic below. Using a CPU idle program is not recommended with a peltier – increases condensation risk. Running Prime95 raised temps into the 10 C range. In normal use, temps stayed around 2 C. The temp at the radiator intake was 31.6 C and 29.2 at the outlet; the block was slightly warm to the touch.
I was able to hit 1030 MHz but not at all stable – just enough to tantalize. I can’t say that 993 MHz is totally stable as I have not run exhaustive tests on it, but this chip has never seen 993 MHz before; all in all, super performance – almost twice spec speed! I’m sure with tweaking, a stable 993 is attainable.
Shown below is CPU ID at 993 MHz – note that in this version of CPUID, the CeleMine is not recognized.
After I ran these tests and typed the article up, I shut down and disassembled the MCW2000 to check for condensation. NADA. But I again emphasize that proper assembly and attention to details are essential to avoiding problems.
Assembling and running the Swiftech MCW2000 was extremely easy. This is about as “off-the-shelf” as you can get with what must be regarded as extreme cooling. The workmanship is absolutely first class and, assuming careful assembly of the components, condensation even at more extreme temps can be effectively controlled.
Both the radiator and the water block are set up for compression fittings*. I prefer to use plastic tubing for my setups as I move them around a lot, and undoing copper tubing would be a hassle. You can either use 1/2″ plastic and mount it on the threaded nipple (not preferred) or fit a 3″ 3/8″ copper tube on the fitting and mount tubing onto it (preferred). Copper is more or less permanent.
In addition, I found the fittings on the waterblock made for a tight fit in a mid-size tower; with copper tubing, this may not be a problem.
No question that for getting the maximum out of any CPU, low temps are required. No question that peltiers can get you there, and water cooled peltiers are by far the preferred setup.
However, we have seen cases where CPUs have been destroyed by condensation from peltiers. In my book, before I would attempt a peltier rig, I would do research about the pros and cons and invest in a quality setup. Proper materials, attention to detail and knowing what you’re doing are going to keep you out of trouble.
*NOTE from Gabe at Swiftech: “Compression fittings also work very well with polyethylene tubbing. Actually, that’s what I sell in my kits, and that’s what I have in my system. The “deluxe” line of Anderson Barrows fittings which I use, have a built-in insert and a long ferrule which are approved with this type of tubing. I even used soft vinyl and it works equally well. The ferule prevents the tube from squashing while the insert is being shrunk during tightening.”
Thanks again to Gabe at Swiftech for allowing us to play with this unit.
SUMMARY: A beautiful product, but does price/performance justify a purchase?
Swiftech was kind enough to send me a sample of their water cooled peltier and fan. If there is one thing about Swiftech’s products, it is that they are really well made. In many conversation I have had with Gabe about water cooling, there is no question that the modus-operandi is “always go first class”.
Naturally you pay for quality; the question I always have is, does performance justify the price?
The Swiftech radiator/fan combo (pic below) is well built and designed for integration into a case. The units measures 6 1/4″ x 7 3/4″ (160mm x 195mm) total height and requires an opening of 4 3/4″ x 5″ (120mm x 128mm). The radiator itslef is 2″ (50mm) deep; with the fan inplace, 3″ (75mm).
As you can see, the tubing snakes its way through the core for maximum cooling. The fan is toally shrouded on the radiator, so no wasted air flow here.
I used my portable peltier water cooler test-bed; two stacked peltiers throwing off about 100 watts cooled with a copper bllock and a 90 gph pump. I compared the Swiftech radiator to BeCoolings single unit, using the Swiftech fan (Nidec Beta V TA 450) on the BeCooling unit:
Swiftech: -35.7 C
BeCooling: -34.9 C
The Swiftech unit outperforms the BeCooling unit, but not by much. I think this is due to the relatively low heat load that CPU cooling places on a water cooling setup. Once water is cooled to somewhere around room temp, a larger radiator quickly bumps up against diminishing returns.
Swiftech, however, has designed the system to handle multiple peltiers, and this is where a larger unit will be more effective. How much more effective requires more testing and more powerful peltiers to simulate load, a step I will be taking shortly.
UPDATE 6/19/00: After discussing the situation with Gabe, I reseated the 533A by using a slice from an index card; I made an “L” with the index card, marked one edge and then carefully measured all around the edges of the MCW2000 so that the back plate was equidistant from all four corners. I spent about 20 minutes doing this. Theoretically, this should ensure that the CPU is seated perfectly on the copper cold plate.
I then fired up the whole system again on my ABIT BX6-2 (pics below). The screen shot shows that I did in fact attain a stable 993 MHz (stable at least to run Prime95 for about an hour). While running Prime95, I cruised the net and had no problems.
The following temps were measured after an hour at 993 MHz:
- Ambient temp 24.0 C, humidity 70%
- Back Plate: 25.7 C
- Block Front: 28.4 C
- Radiator temp inlet: 29.1 C
- Radiator temp outlet: 26.4 C
The tricky part of the MCW2000 is ensuring that the CPU is seated flat on the cold plate; as many of you know, CuMines can be a pain because the CPU area is so small. With very careful alignment, the MCW2000 can be an effective peltier solution. As a product, it is first class construction and, with proper assembly techniques, will allow you to effectively use peltiers as a cooling solution.
Highly recommended, assuming you are knowledgeable about peltiers in general and possess the requisite skills to assemble it.
UPDATE: I had a nice chat with Gabe at Swiftech and it looks like results were impacted by misalignment between the copper cold plate and the flip chip. Because the bolts are so far from the CPU, it is very easy to be off enough to impact results. I will be re-testing soon, so stay tuned.
SUMMARY: The Swiftech MCW2000 with two peltiers allowed me to take my CeleMine 533A to a stable 935 MHz at 1.75 v; 993 MHz was a lot closer, but still a bridge too far – or is something else going on?
I tested the MCW2000 with one peltier (HERE) – I then peeled off the insulation and proceeded to build up the parts to a FCPGA water block.
As the next two pics show, the copper cold plate is designed for two peltiers. Also, note that only a small platform sticks up to contact the CPU.
The peltiers lie side by side, separated a small distance by the bolts that secure the copper plate to the water block. I greased both sides, bolted the cold plate carefully to the water block and then proceeded to lay down silicone all around the cold plate. I did this to ensure that no moisture would make it to the peltiers.
There are three foam layers to build the insulation sandwich; each layer has adhesive on one side, so as you build it, use the screws to make sure the holes line up. As you can see from the pic above, the copper which contacts the CPU is small compared to the rest of the water block (US one cent piece). The cutouts are for the ASUS slotket.
Condensation is a real concern with any below-ambient setup. As a test of the Swiftech unit’s ability to exclude moisture, I ran the peltier on the water block without any CPU in the slotket. This gets maximum cooling at the peltier; if there is going to be a moisture problem, running it like this should show it up.
I ran it for an hour, taking measurements on various parts of the system and found the following temps:
- Ambient temp 22 C
- Back Plate: 20.8 C
- Block Front: 30.5 C
- Slotket Pins: 18.1 C
- Radiator temp inlet: 29.7 C
- Radiator temp outlet: 28.6 C
I then disassembled the unit and checked for water inside – once again no moisture. The insulation package came through.
For the “live” test (next day – hot and muggy), I then proceeded once again to load the whole thing into my case. I used a different setup – the Swiftech radiator but this time with a Beckett in-line pump (150 gph); this was a lot easier to move around and set up. Note that I also used two separate AT power supplies – one for each peltier. I once again had to remove the case power supply to get the water block to fit. Not pretty, but OK as a test bed.
Live Test Results – 935 MHz
I wanted to see what would be stable with this setup, so I started at 935 MHz. As you can see from the screenshot below, I ran Prime95 for about an hour with no problems. After one hour, the odds are pretty high that going further will not be a problem.
The following temps were measured at the end of the test:
- Ambient temp 26.7 C, humidity 75%
- Back Plate: 28.0 C
- Block Front: 34.7 C
- Radiator temp inlet: 35.4 C
- Radiator temp outlet: 32.6 C
Live Test Results – 993 MHz
Well, here again looks like 993 is a bridge too far. With one peltier, Prime95 ran for about 2-3 minutes before crashing; with the two peltier rig, I could go for about 20 minutes before crashing. This was at 1.85 volts. I did not go further; 20%+ over spec voltage is about as far as I wanted to go.
However, as I was running this test, something seemed not right. I looked at the temps and thought “They should be lower”. Strange. I checked and both power supplies were running (only one running gets you up to 20-25 C). I then backed the voltage down to 1.75 and did a Waterfall-on Waterfall-off reading and compared it to the single TEC in the earlier test:
Now this is interesting! Why should two TECs, running full blast, run warmer than one TEC? Well, there is one significant difference that might account for this – the size of the cold plate.
In the single TEC run, the cold plate was PPGA size; for the dual TECs, FCPGA size. It may be that the PPGA cold plate had more “reserve” than the FCPGA cold plate; that is, as the smaller contact area heats up, there is no surrounding metal to “store” cold and act as a buffer – hence, higher temp.
You ask “But what about results at 993 MHz? You said with the single TEC, you could only run for a couple of minutes while with the dual TEC (which you say is hotter) you ran for about 20 minutes?”
Yes, at 1.75 volts, NOT 1.85 volts!
Could be that the higher voltage with the single TEC would have resulted in a stable 993 MHz; unfortunately, too late as the foam is kind of wrecked.
These results are counter-intuitive; that’s what makes it so interesting. I can’t tell you if this is a fluke, but it does raise some design questions that I hope Gabe at Swiftech can answer.
As I am writing this up, I decided to try running only one peltier at 800 MHz. The results are quite interesting – with only one peltier and Waterfall off, temp stays around 25 C – no potential problem with condensation here. With Waterfall off, temps averaged about 15 C.
I mention this because I’m a believer in shifting to overdrive when you need it – with a dual peltier rig, it’s easy to keep the CPU at a “safe” temp by turning one peltier off and still get good cooling.
Well, as I started this, I thought two TECs are better than one; as I look at the results, I know two are better but design considerations are equally important. There’s a lot of art that goes into designing these systems, and I think this is a case where the artist needs to look more closely at each painting.
Many thanks to Gabe at Swiftech for allowing me to test the MCW2000 – absolutely first class construction.
PS: I ran the MCW2000 on the BX6-2 for about 5 hours straight. I just opened the block up and found the Sahara Desert – no moisture at all.