Introduction

Chilled water presents many advantages to the PC enthusiast. Processor temperature in a water cooling loop is dictated by several factors: the TIM joint, the waterblock's efficiency, and coolant temperature.  By chilling the coolant, one can dramatically lower processor operating temperature. 

but is now sold on their website to anyone with ~ $300US.  Bill Adams of Swiftech was kind enough to loan us a MCW-Chill for (protracted) testing

The MCW-Chill and Setup

The MCW-Chill is a water chiller that has been produced largely from standard Swiftech waterblock parts.  At first glance it seems to be a couple of MCW5002 waterblocks cooling peltiers. The cold side of the peltiers are in turn cooling a small chamber where water is chilled.   And this is true, but other  Swiftech parts lurk inside the insulation.  A diagram should be helpful in demonstrating how the MCW-Chill is put together:

Instead of traditional cold plates, the MCW-Chill uses baseplates from MCW5000 waterblocks.  This dramatically increases the surface area exposed to the water on the cold side and is one of the keys to the MCW-Chill's success. 

Every overclocker should want chilled water; it's the logical step up from water cooling and can, if put together correctly, handle a CPU, GPU, and northbridge.   In the case of the MCW-Chill, there is definitely enough headroom for cooling all your system's hot bits.  But while dual 226W peltiers may chill water quite effectively, there is a downside: You now have to cool those peltiers! No free rides here guys; look forward to dissipating well over 500W total heat and using a lot of electricity to power the peltiers.  And even if that doesn't scare you away, there's still the issue of space.  And you'll need a lot of water cooling parts: 2-3 complete cooling loops in fact.  When I set up the MCW-Chill, I opted to cool each 226W peltier with a separate loop.  I think that I made the right choice as performance was quite good and going overboard on cooling is always fun.  This meant, however, that I required 2 large heatercores, 4 100CFM 120mm fans, two 300GPH pumps, and a lot of ½" hose for the two hotside loops.    In addition, a smaller pump, CPU waterblock, and 3/8" hose were needed for the chilled loop. My final cooling loops looked like this on paper:

And like this in practice:

(note the 2 bench PSUs above right and the overall urban sprawl).

When running the MCW-Chill and associated loops noticeably heated my basement room and was very noisy.  But it did get the water very cold…very rapidly.

Testing the MCW-Chill

To characterize the MCW-Chill's performance, I used the following test equipment: 

• Swissflow flowmeter.  This sensor is rated down to -20C and has 1% error.  I later compared readings from this unit with my GPI flowmeter and there was a difference of no more than 0.01GPM over the 0.5-1.5GPM range.
• CrystalFontz 633.  I used Dallas One Wire probes from this unit to measure motherboard backside and waterblock surface temperatures, and I also used one of the 633's fan headers for the Swissflow.  The 633 was controlled by a notebook computer rather than by the test system.
• Test system was a TBredB 1700+ JIUHB running on an Epox 8K3A motherboard.  The 8K3A had to be modified to no longer read from the CPU's diode so that the MAX6655 would work properly.  This was done by breaking pins S7 and U7 inside the ZIF socket (An old OCWorkbench guide to modifying an K7S5A motherboard was my guide).  The CPU was placed under load by using CPUBurn at high priority.

Testing Results