Swiftech MCW6500-T TEC (Peltier) Assisted Water Block

hardnrg - 2007-06-06 17:11:58 in CPU Cooling
Category: CPU Cooling
Reviewed by: hardnrg   
Reviewed on: June 14, 2007
Swiftech
Swiftech
Price: $159.95 (optional $145.00 for Meanwell S-320-12 PSU)

Introduction:

Peltiers. What images does that conjure up for you? For me, I think of iced over waterblocks and sub-zero cooling. Nowadays, the term "peltier" is used when talking about the cooling device itself. This isn't strictly true, as the name is a Thermo-Electric Cooler (TEC) and Jean Charles Athanase Peltier is the person who discovered the scientific principal of the temperature differential created when applying power to a semiconductor. Essentially, all you need to know is that a TEC is usually a square flat device, where one side gets very cold and one side gets very hot. If you remove the heat from the hot side, the cold side continues to be cold. For processor cooling, if the cold side is placed to absorb heat produced by the CPU or GPU and the heat produced on the hot side is taken away by something with a high thermal capacity like water, then sub-ambient temperatures can be achieved for the processor.

Swiftech has been around for a long time now, producing cooling solutions since 1994 for high-end industrial computer systems. Many of you will remember the MCX air heatsinks with the unique helicoil heat dissipation system and later mass-producing the world-renowned Storm G4 water block, designed by Cathar. Swiftech produces a full complement of air and water-cooling products and combines its water-cooling products to offer complete water-cooling systems as kits. These kits are high performance cooling solutions that give much better results than the type of water-cooling kits generally frowned upon.

Most people think of Dangerden or Thermaltake when venturing into water-cooling, so it is interesting to see what new products Swiftech has to offer to bring itself further into the limelight. In this review, I will be taking a look at the TEC-assisted water-block, the MCW6500-T and comparing it to plain water-cooling. TEC-cooling is what I would consider to be the cheapest form of sub-ambient cooling and one step above the most efficient "regular" cooling method of using water. This product is a complete water-block, TEC, and coldplate all in-one and designed with TEC cooling in mind. So I will be comparing the MCW6500-T to one of the highest performing water-blocks ever made, the Swiftech Storm G4. Is it worth the effort? Does it increase stability and allow higher overclocks? Let's find out!


Closer Look:

The block itself comes in a simple, no-frills cardboard box with a black and white outlined technical drawing of the block on one side and a black and white plain text sticker on the front of the box.


This says to me that no cost has been wasted on needless, glossy, colourful packaging and really screams "understated". So, on the front, we have a straight-to-the-point product sticker basically saying, this is the water-block, it has a 226W TEC (Peltier Module) built in and it has been designed primarily for LGA775 Intel Core 2 processors.


The inside of the box has polystyrene inserts to keep the contents safe from damage during transit or handling. One main insert at the bottom that the blocks sits in and one on the top. The block sits low inside the box. so the pre-installed barbs are nowhere near the top of the box. This means that it is pretty much impossible that the barbs would be damaged whilst the product is inside the box.

As well as the block, there are two neoprene gaskets for the LGA775 socket, mounting screws, nuts, washers and springs. Also included is some Arctic Silver Ceramique thermal paste (without the usual sticker on the side). You also get a black and white installation guide that looks to be laser printed on two sheets of US Letter size paper, stapled at the top left corner. Continuing trend of simplicity and no-nonsense here then.


So straight away, you can see that this isn't a normal water-block - it has gaskets and a short power cable protruding from the side! A cable for 226W, that's some serious power. The hex screws that hold the top part of the waterblock down have an American/imperial size hex hole and being in metric-dominated Europe, it is somewhat difficult to get imperial size tools. I couldn't find any imperial hex drivers (Allen keys) to fit these screws and could not open up the block for you to see the inside. The picture above shows the inner diamond-shaped grid inside the block, similar to the Apogee water-block design. There are no impingement holes to create jets and the water flows from one side to the other through the diamond-shaped grid.

As I don't have an Intel Core 2 processor, I requested that Swiftech send the necessary accessories to use the block on AMD's Socket 939. One of these additional accessories is the 939 version of the neoprene gasket. The centre of it pops out to give you the actual socket gasket and a neoprene pad for the underside of the motherboard, directly under the centre of the CPU socket.


You might think that the top of the block is made of Delrin, the advanced plastic made by DuPont, which is used on other waterblocks. This is in fact an aluminium top housing, plated with nickel and zinc cobalt, and dyed black. The short power lead is neatly sleeved with nylon braiding and terminated with the common 4-pin Molex connector. This allows you to power the MCW6500-T with your PC's own power supply, although Swiftech's recommendation of an Enermax Galaxy 1kW unit and having 18 amps to spare on a 12V rail certainly raised my eyebrows.

 


The lead itself is tethered to the 775 mounting plate with a cable tie, to prevent damage to the fragile solder joints near the edge of the TEC. The closer I looked at this water-block, the more I liked its industrial look, with the bolted down top housing and mounting plate. The thing just looks serious, as if  it means business! The top housing has 1/4" BSPP threads which are pre-fitted with 1/2" chrome-plated barbs. Taking a look round to the side of the block, you can see that the block is surrounded by neoprene all the way down to the bottom.

 


Down at the bottom, of course, we have the copper cold plate which has been lapped to 0.0003" flatness. This is incredibly flat and I doubt it's possible to achieve this level of flatness by hand. Now, some people obsess over the "mirror finish" of a heatsink or water-block's base when in fact, it is the flatness of the base that is the most important. You can see how flat the base is, by observing the reflection of objects, looking for distortions - the extreme case being something like a fairground funny mirror.

 

 


No funny business here then! Just serious flatness. The base has ever-so-slightly visible "grains" from the lapping process. They are quite hard to see and take a photograph of, but you can make them out on the last photo. You could polish this base and it would look more like a glass mirror, but it wouldn't perform any better and may well perform worse with polishing compound tainting the surface.

Okay, so back up to the top and as I mentioned, Swiftech sent me the 939 accessories, one of which being the neoprene gasket/pad, the other is the 939 mounting plate.

 


The 775 mounting plate is bolted down to the top housing with four hex screws. Luckily, these stand proud of the surface and I could start the untightening, using some snipe-nose pliers. The mounting plate is also glued to the neoprene surrounding the block, so you'll have to cut around it, right next to the underside of the plate to free it from the neoprene. You know how I pointed out that the power cable is tethered to the mounting plate to protect the delicate wires? Well, the 939 mounting plate doesn't have a tether hole! Actually, the 939 mounting plate looks like it might be a near-final prototype version, as it does not have the same chrome plating or rounded edges of the 775 plate. Because it is stated that the wires are so delicate, I felt it was necessary to make my own tether point and drilled a hole through the 939 plate. This was no small task! It took me about one and a half hours and I ended up breaking about eight titanium coated drill bits! There is no way this plate is going to bend or warp under pressure, a magnet doesn't stick to it, so it isn't steel. Whatever it is, it's harder than anything I've ever cut or drilled!

I used Bison Kit Contact Adhesive to attach the new 939 plate to the neoprene.


So with a little mounting plate swap, here is the TEC-block set up for mounting to the 939 socket.

Closer Look:

The TEC in the MCW6500-T has to be fed some serious power. You could run it off your PC's power supply, but you'd have to not be using 226 Watts on a 12 Volt rail. It's probably a better idea to use a dedicated TEC power supply. My main rig has about 80 Watts to spare, nowhere near enough to power a 226W TEC. So for me, the recommended Meanwell 320 power supply was a necessity, not an option.




The Meanwell S-320-12 is a variable voltage power supply that can supply 25 Amps at 12 Volts. This is over and above what the 226W TEC needs, so there will be no shortage of power. The unit itself is similar to an optical drive in size, a bit longer, a bit taller and a bit narrower in comparison. At the back you have Earth, Neutral and Live mains AC input terminals, three terminals for the positive DC output and three terminals for the ground DC output. To the left of the terminals you see the voltage control potentiometer. This lets you dial the voltage up or down 10%, up to 13.6 Volts for maximum TEC cooling operation, and also down below to 10.8 Volts if you wanted to pull on the reins at all, for example, if the TEC was freezing your CPU so much, that it began whimpering and begging for some hot chocolate.

Round the sides there are vents at the bottom that extend underneath the chassis of the unit. You may have noticed the fan hole at the top. This is a thermally controlled exhaust fan that stops the PSU from overheating much like a regular computer PSU.


As the unit is narrower than a standard 5.25" bay device, the PSU needs adapter brackets to be fitted in order to be mounted in a bay. These brackets are supplied as part of the Meanwell 320 accessory kit.

 


Four screws are supplied to attach the brackets to the PSU chassis. So you end up with the PSU like this:


The brackets have threads tapped on the outer sides to allow screws to be fitted to secure the PSU/bracket in the 5.25" bay. However, these threads turned out to be #4-40 UNC American threads! The standard drive bay screws for optical drives, fan controllers and any 5.25" bay device I can think of, use the metric M3 screws. This might not be a problem if you have a case that uses screws to secure a device AND you live in America. I live in England and I have a case that uses M3 "cheese head" screws that allow devices to slide in and lock into the 5.25" bay. There was no option but to modify the brackets, which I will show in due course.

 


So anyway, you end up with the PSU/bracket combo ready to fit in the bay. Also supplied in the Meanwell 320 accessory kit is a control relay and an IEC C14 mains connector. This connector is often called a Europlug or kettle-plug, but it is exactly the same as the ones used in ATX and many other form-factor PSUs.

 


You might be wondering how this all connects. It's a lot easier just to show you how it works by showing the installation process and you can see then how the cables hook up and everything else.

Installation:

Right, so the TEC-block is ready for Socket 939, the Meanwell PSU has its brackets attached, so things are ready to rock and roll. What do you do now? Strip down your entire case to the bare chassis. What?! Yes, that's right, take everything out of your case in preparation for the hole you're going to cut. What hole? The one for the IEC C14 mains socket. Here's the back of my stripped down case chassis, you can see there are few places I could fit the mains connector.




The actual hole cutting is really easy. Swiftech suggest a 1 1/4" (32mm) holesaw. I measured the diagonal length of the housing of the connector to be 30mm and went for a 29mm holesaw instead going on previous experience of the holesaw producing a slightly bigger hole than its stated size. Here is the holesaw locked onto a 6mm arbor drill bit on my Ryobi drill.


Low RPM, high pressure, high torque and lubrication are needed for cutting thick or hard steel. My case is quite hard, so it took some steady cutting at around 240RPM to get through it. Most PC cases will be a lot easier to cut through so mine is sort of a "worst case" scenario. Ba-dum-tsssh!

I masked off an area above the 120mm exhaust and below the PSU. I'd eyed up this area before with the rig running and deemed it suitable for placing the connector there and fitting the relay next to it, running the cables neatly, etc.

 


If it's steel, remember to cut using drilling lubricant or some kinda oil. You could use WD-40 or something similar if you don't have actual cutting/drilling oil. Cutting dry into steel produces heat. Heat in turn makes the steel harder and more difficult to cut. So this can easily escalate to a point where the steel becomes too hot and hard to cut and you will just wear down your holesaw instead.

Place the IEC C14 connector in its mounting position and mark out the screw holes. Swiftech say 3mm, but I went up to 3.2mm. De-burr the edges with sandpaper or a half-moon needle file and you'll end up with something like this:

 


The mounting plate of the connector will cover up any damage to paintwork caused from cutting, drilling, sanding, or filing. Just keep in mind how big the mounting plate is and don't go crazy with the sanding / filing. At this point, you will want to clean out your case and make sure there are no iron filings (or aluminium filings?) in your case and de-grease the case to wipe up the cutting oil.

Okay, once you've done that, use the supplied screws and nuts to fasten the connector to the chassis. The relay module has an adhesive foam pad that lets you stick the relay to the inside of the chassis. Make sure the area is free of grease or dirt and stick it somewhere near the mains connector. Swiftech suggest a spacing of at least 1/4" (6.35mm) between the relay module and the mains connector. Here are the IEC C14 mains connector and relay module installed in the case:

 


How does that relay module work then? Glad you asked. Basically the relay module turns the Meanwell PSU on whenever your main computer PSU is on. How does it do that? Well, a 4-pin Molex is connected to one of your main PSU's molex lines and when your computer is turned on, the 12 Volts from the Molex activates the relay coil. This switches on the relay and passes the Live mains power to the Meanwell PSU. This makes it a lot easier to power the TEC when your PC is in operation. Carefully observe the mains wiring assignment for the socket terminals in the diagram below.


If your computer is turned on and the TEC was not powered up, the TEC acts as an insulator and the processor would become hot very quickly. Conversely, if the TEC is powered up, but the water pump is not powered (assuming it is a 12V pump powered from the main PSU), then the TEC itself will overheat and quickly deteriorate if its temperature exceeds 85°C.

Right, so the connector and relay are installed and we can just put the Meanwell PSU in the drive bay and hook it up, right? Wrong! I don't know if Swiftech or Meanwell produce the adapter brackets, but for some unknown reason, they are tapped with #4-40 UNC threads. I'm told they are very common in the US, but the standard for 5.25" drive bay screws is M3-50. Different diameter, different thread pitch, very annoying. So I had to buy a tap and die set and re-tap the threads to M3 to allow me to fit my "cheese head" screws. This has to be a manufacturing oversight. There is no reason to use #4-40 UNC threads here. The only screws I could find with those threads are the D-Sub jack posts - the things that are either side of VGA, serial, etc connectors.


It was an excuse to finally buy a tap and die set anyway, not sure how I've made it this far in life without one! So here is the Meanwell 320 PSU mounted in a drive bay.


The PSU is quite long, longer than most optical drives. The installation instructions suggest mounting it in the top drive bay. This would make sense I suppose, for cable routing and the exhaust fan would not be blocked by optical drives above it. But at the top spot, it's going to be blocked by the top case panel unless you mod a blowhole there. The PSU accessory kit comes with an 80mm fan guard and suggests that you mod a blowhole in the top of your case, if you mount the PSU in the top drive bay.

My case configuration makes it impossible to place the Meanwell in the top four, of six, 5.25" bays. This is because of my radiator air intake duct. No matter though, the PSU is fine down at the bottom 5.25" drive bay. It is still level with the main PSU, so it is roughly situated in the same place as the top 5.25" bay of a mid-tower case.

 


There is a transparent plastic cover that clips in place to cover the terminals. You just pull it off either at the top edge or bottom edge and it lets you use the screw terminals to hook up the mains power cables. Make sure you got the wiring correct at the connector end and then match up the crimped connectors to the screw terminals. If the white, black, green colours are not familar to you (in the UK it's brown = live, blue = neutral, green/yellow = earth) then the cables are even labeled to avoid any confusion. This is good, because confusion and mains electricity aren't a good mix.

Installation:

Okay, case modding and mains wiring done and dusted, so what's next? Now you have to prepare the motherboard to protect it against condensation. When a surface is colder than the surrounding air (below dewpoint), the moisture in the air condenses on the surface. A common example of this, is when a can or glass of ice cold beverage quickly becomes covered in tiny droplets of water. As you know, water and electronics are seldom a good combination. You need to get some conformal coating, which is basically a spray on electrically insulating layer to prevent electrical short circuits from surface moisture. I used Electrolube DCA SCC3 Conformal Coating. Make sure you read the instructions, you will almost certainly be told not to shake the can before spraying, contrary to the instinctive action. This stops air bubbles in the protective layer being applied to the motherboard.


Spray the back of the motherboard, concentrating on the area directly under the socket, in the near vicinity and then vertically down, so that if droplets form and roll down the motherboard, they won't cause a short-circuit. I had to use a filtered paint mask to spray this stuff as it is quite harmful to inhale.


You need to let the conformal coating dry before you continue. The stuff I used says it needs to be thermally cured in a drying station or something like that. I guess that's a unit that is heated and extracts the fumes away. In my case this consisted of my bathroom with the extractor fan turned on, and a minute or so with the hairdryer set to low heat. The stuff is highly flammable so don't go crazy trying to thermally cure the coating super fast. Just warm it gently to get it going and then leave it to cure. When it's done, you can attach the neoprene pad to the area directly under the centre of the socket.

Pretty easy right? Next you are going to do the business side of the motherboard, masking off all the important sockets, header pins, edge connectors, jumpers, etc. I didn't mask off the chipset fan, so it got some coating on it. It probably would be a good idea to remove the chipset fan altogether and mask off the chipset itself. I used multiple layers of masking and parcel tape on the CPU socket to ensure it was sealed from the conformal coating.


So, same deal with the no-shake, spray, gentle heat, and cure. As I run "naked" processors with no heatspreader, I thought it would be a good idea to mask off the CPU die and conformal coat the surrounding CPU PCB.

 


This is where the cold plate area is going to be so it seemed to make sense in this situation.

Now, the neoprene gasket that you get seemed to be fairly small, with a very narrow, flimsy "wall" along the top edge. I decided to make a surrounding neoprene gasket to provide some sealing around the gap underneath the 939 retention bracket.

 


So above you see the supplied gasket, the extra gasket I made, and the two together. Below you see the supplied gasket fitted to the motherboard and then the extra gasket around it.

 


The supplied gasket has a peel-back adhesive layer on the bottom. I used Bison Kit Contact Adhesive to glue my extra neoprene gasket to the motherboard.


When I was happy with that seal, I moved onto adding the Luberex dielectric grease.


This stuff seals electrical contacts from moisture, but still allows electrical current to pass through it. You put a generous helping in the socket itself and then place the CPU down in the socket and let the grease squeeze out the sides.

 


With the CPU in place you can now add the Arctic Silver Ceramique thermal compound. This is used instead of Arctic Silver 5, because Ceramique works better at sub-zero temperatures and the TEC-block is capable of producing temperatures below freezing.


Then add even more grease around the CPU to seal all the air spaces.


At this point I have to pause for a moment. After I completed the subsequent steps of installation and fired up the computer, it didn't POST. It didn't even beep to tell me something was wrong. Only the four diagnostic LEDs on my DFI NF4 Ultra-D told me that the CPU wasn't being recognised. After some testing with a Thermalright SLK-948u heatsink, I reached the conclusion that somehow traces of conformal coating had seeped into the socket, maybe from underneath or around the edges. This was preventing the CPU pins making contact inside the socket.

Nightmare situation? Sort of. I had to buy some industrial solvent cleaner to clean and dissolve the conformal coating inside the socket. I couldn't get hold of the recommended Electrolube DRG SCC3 Conformal Coating Remover Gel, so I picked up some Servisol Coldklene 110 from work instead. It was even more toxic than the conformal coating! But it worked and after leaving it to soak in the socket overnight, finally my CPU was recognised by the motherboard.


So, that sucked, haha. Anyway, with the CPU socket functional again, I was all set to finishing up the installation.

Installation:

The previous extra gasket I made was a bit too thick and didn't really let the retention bracket sit on it properly. So I made a new one with cut-outs for the protruding "feet" of the retention bracket.







This allowed the retention bracket to fit a lot better and gave a firm seal all the way around underneath the edge of the bracket.


There was only one supplied 939 socket gasket, which didn't really seem to come up high enough to make a seal with the TEC-block's own neoprene. There actually appeared to be a gap of 1-2mm in some places between the MCW6500-T's neoprene and the gasket. So I made a further inner gasket. You can see that the CPU sits below this gasket now so the neoprene will definitely be sealed all around.

 


Rather than cover the CPU all around with a thick, even layer of grease, I sealed the surrounding air gaps around the socket, covered the surface-mounted capacitor networks on the CPU PCB, and tapered the grease off towards the CPU die so that the grease wouldn't end up between the CPU die and the MCW6500-T.


So, putting the inner gasket back, the retention bracket, and bolting the TEC-block with the supplied screws and springs through to the backplate, I finally mounted the MCW6500-T to a working CPU socket.

 


Now, the lead for the MCW6500-T is terminated with a 4-pin Molex connector, to allow you to use it with a high power ATX PSU. As the lead is only short and I needed to make a short extension, I decided to use Tamiya and Kyosho connectors instead of 4-pin Molex. These connectors are widely used on radio-controlled car battery packs, motorbikes, cars, etc. They can handle a lot of current and have a locking clip which is a lot less annoying than awkward 4-pin Molex connectors.


So, I made the short extension cable to connect to the Meanwell PSU. I then re-terminated the MCW6500-T with a Kyosho connector.


Now it's just a case of hooking up the two locking connectors.


The wires on the lead from the TEC do get quite warm, verging on calling it hot. You don't want these to rest up against the tubing for your water, else it may melt a hole through it, or deform it causing a kink, or other nasty surprises. Actually, the whole Meanwell PSU gets pretty warm, like about as hot as a mug of coffee that's been sitting there a few minutes - not too hot that you can't touch it, but very warm. When the Meanwell PSU's fan spins up, the air coming out is also very warm. It's a 60mm high performance fan, so when it spins up, you know it's spinning. It wasn't annoying to me because it was mostly drowned out by the other fans in my system, but it could easily be annoying if you have a near-silent computer.

Below, you can now see why I can't position the Meanwell in the upper bays. The massive black thing below the blue radiator is a fibreglass air duct that makes the radiator take air in from the side, outside of the case, rather than the warm interior case air.


Hooking everything back up and firing up the rig, going straight to the BIOS thermal monitoring section brought a smile to my face! Only 6°C. Finally, the hardware installation was over.


These initial checks to make sure the installation had been a success certainly bode well for the testing that was to commence.

Specifications:

Swiftech MCW6500-T






Performance curves:

 


Thermoelectric module characteristics:

U max I max DT max Q max L x W x H (mm) R (Ω)

15.2 VDC 24 A > 67°C 226.1 W 50 x 50 x 3.10 0.48
Measured power draw at 12 VDC: 18A


Housing:

CNC machined aluminum L2.5"xW2.5"xH.4"


Base plate:


Cold Plate:


Assembly:


Retention mechanism & CPU compatibility


Power supply requirements



Notes:

The only difference between the -775T and -939T models is which mounting plate it uses. The actual TEC-block is identical. The Meanwell S-320-12 is rated for a voltage adjustment of ± 10% which means the maximum should be 13.6V, not 13.8V. The highest voltage I measured with a multimeter was 13.56V, which tallies with the 10% and 13.6V specification on Meanwell's own product page.

Testing:

This review is going to compare the MCW6500-T with one of the highest performing water-blocks, the Storm G4. The test setup will include two GPU waterblocks in the same cooling loop. It is claimed by Swiftech that this should still be possible and that the GPU temperatures will still be well below stock air cooling. The graphics cards are volt-modded and overclocked and will be left at their fully overvolted and overclocked state for consistency.

The temperatures of Motherboard Monitor 5 (MBM5) with the custom DFI NF4 settings, has been found to give reasonably accurate readings, within a few degrees of a temperature probe and the newer, and supposedly more accurate, software monitoring program: CoreTemp. So I will use MBM5 with a time interval of 2 seconds.

For all tests, the Swiftech MCW6500-T is cooling the CPU and two graphics card are also in the same water-cooling loop. Only the load on the CPU and GPUs are varied, to assess the temperature changes. I devised three temperature scenarios:

Idle - the average temperatures over 30 mins of idling at the desktop,
Load (CPU only) - the average temperatures over 30 mins of Stress Prime 2004 Orthos Edition (a processor-intensive stress-test application based on Prime95),
Load (CPU + GPU) - the peak temperatures over the full run of 3DMark2006 (a 3D benchmarking application which stresses multicore CPUs as well as multiple graphics cards).

The temperatures being recorded are those of the CPU, the PWM IC (voltage regulator) and the two GPUs. Each temperature test scenario will be run for the test machine with the Storm G4 and then with the MCW6500-T @ 12 Volts, and then with the MCW6500-T @ 13.6V.


Testing Setup:


Results: Opteron 170 @ stock

Ambient = 26°C

This set of results compares the temperatures of each component for the system at idle on both the CPU and GPUs, against the temperatures achieved when the CPU is put under 100% load (the GPUs remain at idle)


CPU Temps (CPU Idle vs CPU Load)

PWM IC Voltage Regulator Temps (CPU Idle vs CPU Load)

GPU1 Temps (CPU Idle vs CPU Load)

GPU2 Temps (CPU Idle vs CPU Load)

Ok, so the next set of results shows what happens when the entire system is put under load, with the CPU and both GPUs put under very heavy load in a multi-processor (SMP) & multi-graphics-card (SLI) enabled application.


CPU Temps, PWM IC Voltage Regulator Temps (CPU Load & GPU Load)

GPU1 Temps, GPU2 Temps (CPU Load & GPU Load)


So, at idle, the Storm G4 manages a CPU temperature of merely +2°C above ambient, while the MCW6500-T gets the CPU -23°C below ambient! A massive difference in favour of the TEC-block. Strangely, increasing the Meanwell PSU voltage to 13.6V increased the CPU temperature by 1°C. Maybe the extra heat produced by the TEC with 13.6V applied just increases the water temperature.

The difference is not quite so dramatic when the CPU is put under load. With the Storm G4 the CPU temperature rises to +6°C over ambient and the MCW6500-T keeps the CPU at -8°C under ambient.

Similar results can be observed when under CPU and GPU load. The Storm G4 manages a +8°C rise over ambient and the MCW6500-T still manages to keep it below room temperature with a -6°C below ambient.

Also, it's interesting to see that the PWM IC voltage regulator actually runs colder when the CPU is so cold. It is right next to the socket, so the motherboard will be colder itself around this area.

Not very surprisingly, the added heat from the hot side of the TEC pumping round the water-loop has increased the GPU temperatures. The temperature increase ranges from +8 to +12°C for idle and load scenarios. This added heat did not produce any instability or visual artifacts. This came as quite a surprise as the graphics cards are heavily overclocked and overvolted right up to the point of instability.

Testing:

Okay, so now I wanted to take a look at the overclocked temperatures. The highest stable overclock (24+ hours of Stress Prime 2004 Orthos Edition) on my Opteron 170 is 2817 MHz, but I run it at 2799 MHz just to be extra sure it doesn't crash - 2 MHz less HTT speed is peace of mind for me.

Anyway, I tried all sorts of voltages and settings and I could not get the CPU stable under load at the same overclocked speed (2.8 GHz) as water. I think this graph on the Swiftech site can explain what happened in this part of the testing phase.


Along the x-axis, you have increasing heat output ranging from no heat at the left, to extreme heat output for high voltage and high Thermal Design Power (TDP) processors, at the right. The y-axis shows the temperature achieved in relation to ambient temperature. So if the CPU is producing no heat at all (pretend it's a magic CPU that produces no heat at idle), then the water temperature of the Swiftech Apogee block (yellow line) is the same as ambient. Makes sense right? No heat from CPU means it's just the same as room temperature.

Okay, so the blue and red lines are the temperatures achieved with the MCW6500-T. With the magical, no-heat-output CPU, the temperature is -50°C to -60°C below ambient.

As the heat output of the processor increases, the CPU temperature compared to ambient, doesn't increase all that much with the Apogee block. The Apogee and Storm G4 perform quite similarly. The MCW6500-T however, has a steeper change in CPU temperature as the CPU heat output increases. So for high TDP processors, like the Intel Core 2 Extreme and AMD Opteron Dual Core at high voltage, there is a point where the temperature with the MCW6500-T is higher than that of the Apogee. The TEC has a limit to the amount of heat output it can handle and once that has been reached, the TEC-assisted water block becomes less effective than a regular water block. This is indicated by the background shading. So for really hot processors, you are probably just better off with water alone. I'm not talking "oh my processor is 58°C on air" hot, I'm talking "I'm surprised my computer isn't on fire" hot. I cannot run my processor overclocked on air, it runs hot at stock on air!

I was unable to run the tests fully without either the OS or an application crashing. I did manage to run the CPU only test and SP2004 Orthos Edition and MBM5 continued to run, despite other applications crashing.

The test configuration only changed by overclocking the CPU.


Testing Setup:


Results: Opteron 170 @ 2.8 GHz

Ambient = 26°C

This set of results compares the temperatures of each component for the system at idle on both the CPU and GPUs, against the temperatures achieved when the CPU is put under 100% load (the GPUs remain at idle)


CPU Temps (CPU Idle vs CPU Load)

PWM IC Voltage Regulator Temps (CPU Idle vs CPU Load)

GPU1 Temps (CPU Idle vs CPU Load)

GPU2 Temps (CPU Idle vs CPU Load)

The gap closes somewhat for CPU temperatures when overclocked (with the necessary added CPU voltage). The rise above ambient is +5°C for the Storm G4 and -15°C below ambient for the MCW6500-T.

During the CPU Only load test, using the Storm G4 results in a +14°C rise over ambient, but the surprise here is that the MCW6500-T can't handle the heat output of this processor at this voltage and the temperature shoots up to +23°C over ambient!

The PWM IC temperature difference is negligible. The GPU temperatures are 8 to 10°C higher using the MCW6500-T than with the Storm G4 whether the CPU is at idle or load.

The main result at this point of testing was that the Storm G4 produces a higher stable overclock than the MCW6500-T, when using a high thermal output processor that requires high voltage.

Testing:

The Opteron 170 testing left me a bit unsure what to do next. After some thought, I thought it would be a good idea to test a lower Thermal Design Power processor, a 3500+ Venice. It is single core rather than the 170's two cores, and 512K cache rather than the 1MB per core of the 170.


Here you can see the vast difference in die size:

 


Firing up the rig showed the 3500+ to settle at 2°C.


I was running over schedule in the review, partly due to the need to buy tools and extra materials and chemical sprays, and partly due to the socket / conformal coating issue. I didn't have time to test the 3500+ Venice with the Storm G4 as well as the MCW6500-T, so I decided to use the TEC-block and just run the CPU tests. If the temperatures turn out to be good, then the next stage of testing would be to get some new personal bests with one of the world's fastest Venice CPUs. I will use the same idle and CPU-only Stress Prime 2004 testing scenarios to gather temperature data (standard version of SP2004 rather than the dual core Orthos Edition).


Testing Setup:

The temperatures recorded here will only be to show the temperatures of the main system components in idle and CPU-only load scenarios.


Results:

Ambient = 26°C


All Temps (CPU Idle vs CPU Load)


My eyes lit up as my 3500+ Venice dipped belowing freezing point. The lowest I have got any processor is about 9°C on air in the winter and I was wearing a jacket and hat! A glorious -27°C below ambient. Under load, the CPU temperature rises to -16°C below ambient. The other major system components stayed pretty much the same, regardless of CPU load.

Testing:

With the CPU temperatures under serious control from the MCW6500-T it was time to go for some sick overclocks.

First, I took the 3500+ up to it's maximum stable overclock on air ,achieved using a Thermalright XP-120. This speed was 2.92 GHz - I love this Venice! The plan was to record the temperatures at this overclock and overvoltage, and then attempt to break my previous records.


Testing Setup:


Results:

Ambient = 26°C


All Temps (CPU Idle vs CPU Load)

Even at nearly 3 GHz and 1.600v CPU voltage, the MCW6500-T kept the CPU at only +3°C over ambient at load. The type of CPU that the MCW6500-T was designed for was becoming clear and the strange graph from before made more sense. If the CPU isn't a heat producing monster, then the temperatures achieved with the MCW6500-T will be better than air or water cooling. Lower temperatures mean more stability. But just how much stability can you achieve? On air with the Thermalright XP-120, the idle temperature hovered around 35°C while the load temperature was around 43°C. So for total system stability, this was the end of the line for air cooling.


Extreme Overclocking:

Some quick testing using Memtest86+ showed the 3500+ passing the easier tests at up to 3.2 GHz, much higher than ever before!

 


SuperPi 1M test calculates the first million digits of the mathematical value Pi. It's a reasonably hard thing for a processor to do and it has to be at least near to fully stable, in order to be able to calculate that many digits without any errors. The previous best on air cooling was 3001 MHz.


With the MCW6500-T the maximum overclock for this popular test went up 117 MHz to 3118 MHz. This is approaching crazy speeds for a Venice core CPU.

I really didn't have any time to do much else, so I just went for the highest "suicide" screenshot. This is aptly named as the aim of this overclock is to push the CPU as far as it will go until your system crashes. Sometimes the OS can get corrupted and you might have to reformat your computer in extreme cases.


3133 MHz! A whole 183 MHz faster than my previous record and 933 MHz above stock speed!

I felt the magic of overclocking again and the only thing that made me sad, was that I didn't have an Intel Core 2 Duo to test the MCW6500-T out on. I think the 65nm process of the current Intel Core 2 processors producing less heat and requiring less voltage, would gain stability and speed over plain air- or water-cooling.

Conclusion:

So the Opteron 170 testing phase was a bit of an eye-opener. I really thought the TEC-block would be better than the water-block, but evidently the heat output from the processor at that voltage is too much for the TEC to handle and you end up with worse temperatures than plain water-cooling. The Thermal Design Power (TDP) of the Opteron 170 at stock speed and voltage is 110W. The TDP of the 3500+ Venice is 67W. You might think "oh, well, a 3500+ Venice is really old, why did you even bother testing it?". Well, guess what? An Intel Core 2 Duo E6600 has a TDP of 65W, just 2 Watts less than the 3500+.  How would you feel if you could keep an overclocked E6600 at or below ambient temperatures? The CPU voltage certainly isn't going to be as high as the older 90nm CPUs, so you can bank on the temperatures being lower too.
The 90nm CPU fabrication process is getting on a bit now. 65nm has been around for quite some time, reaching its own maturity and 45nm will be widely available in a manner of months. The power consumption and thermal output of processors seems to becoming a slightly more important issue in CPU design and marketing, with the current Intel Core 2 processors making a step in the right direction. The move from 65nm to 45nm can only mean even less power consumption and heat output.

These are merely logical conclusions about processors that I could not test. The factual results from this review clearly show that the MCW6500-T is best suited to processors that do not produce an immense amount of heat in the first place. I think the ideal processor would be one of the "golden" processors that overclocks 50% or more with no increase of CPU voltage. As the increase of voltage largely determines the CPU temperature, one of these golden CPUs operates pretty much at the same temperature whether at stock speed or overclocked.
That isn't to say that the TEC-block cannot handle overvolted processors. The Venice core has an almost identical TDP to that of the Conroe core and the results from this review make me want to install the MCW6500-T on one of the Intel Core 2 65nm processors, or possibly hold out for the 45nm versions, due to be available later this year.
So the Swiftech MCW6500-T proved itself to provide lower temperatures and thus gain higher overclocks for a "normal" TDP processor. That's obviously good. How about the bad? What was annoying about it?

Well, the fact that it couldn't handle my Opteron 170 was a let down, but it was nowhere near as annoying as having to tear down and rebuild my entire rig to install the required Meanwell PSU. I'm fairly handy with tools these days, but I can see this as a limiting factor.  Mounting the IEC C14 socket on the case chassis doesn't seem 100% necessary. Maybe a PCI blanking plate with a suitable grommet could be used with an in-line IEC C14 socket. This would save the nightmare task of stripping a fully loaded case and make the installation process a lot easier. That said, you may have to strip most of your case anyway, just to get the motherboard out. For me, I have to remove at least the pump and both graphics cards, the CPU block and the reservoir to get the motherboard out, so there isn't much difference to a full tear-down anyway. The complexity of the installation process is pretty much the highest I can think of for any hardware product. The amount of work to prepare the motherboard and case are up there with installing a phase-change unit.
Getting conformal coating inside the socket even though I'd masked off the CPU socket with multiple layers of masking tape, was extremely annoying. I thought at one point I'd have to buy a replacement motherboard, as the socket would not recognise the CPU for about a day, even using industrial solvent cleaner. The level of risk during installation is quite high here and it should be made more clear that conformal coating can seep under the socket, so simply masking the top surface isn't good enough.
There really can be no excuse for the Meanwell PSU having "non-standard" threads for the drive-bay screws. The whole world uses M3 screws for mounting 5.25" devices, so why should these brackets use the "crazy" #4-40 UNC threads? Ridiculous.

All in all, what do we have here? A superior cooling device that can give you higher overclocks, but it can potentially also give you a few days of hell installing the whole kit and kaboodle. It doesn't claim to be easy to install. It doesn't claim to be able to cool extremely high heat output processors, like the Intel Core 2 Quad / Extreme or AMD Opteron Dual Core at high voltage. What it can do, is keep processors like the 3500+ Venice and E6600 around or below ambient and net you a couple of hundred more MHz, maybe more.
I wasn't sure whether or not to recommend this TEC-block from Swiftech. I will personally definitely use it on my next processor (either an E6600 or upcoming 45nm CPU), but I was hesitant to give a product a recommendation given the complexity and difficulty involved during installation. However, the fact that it is a more involved installation process both for the Swiftech MCW6500-T and the Meanwell S-320-12 PSU that you may well need, is clearly pointed out on the product pages and in the installation guides. The only downside that I didn't previously know about, was that the PSU brackets had those useless #4-40 UNC threads.

Do I recommend cutting the heatspreader off your socket 939 CPU to everyone? No. Or how about soldering on components to a graphics card to give it more voltage? No. Replacing chips and capacitors on a soundcard? No. Using this combined TEC and waterblock? No.
I don't recommend these things to just anybody and yet I have done and will do these things again and again myself. I will recommend these things to people that are competent and willing to work a little bit in the quest for higher performance.

I love high performance. Do you love high performance? If you have a processor like the Intel Core 2 E6600 or AMD 3500+ Venice, or are planning a system based around the upcoming 45nm processors, then you really should consider a TEC-assisted water-cooling system. If you have no experience with modding, then it would probably be a good idea to stick to less extreme cooling methods.


Pros:


Cons: