Asetek VapoChill XE: Extreme Cooling System for Extreme People

Today we are going to offer you an unprecedented material devoted to super-extreme overclocking of the top Intel Pentium 4 processors, namely Pentium 4 Extreme Edition 3.20GHz (Gallatin), Intel Pentium 4 3.20C GHz and 3.4C GHz (Northwood) and Intel Pentium 4 3.20E GHz (Prescott). With the help of the outstanding phase-change cooling system from Asetek we will make these CPUs overcome the 4GHz barrier in no time!

by Kirill
03/20/2004 | 01:40 PM

As the technologies evolve and the frequencies grow, the central processor generates more heat. That’s a well-known axiom. Working in its normal operational mode, a modern top-end processor dissipates about 100W of heat, thus calling for additional cooling, and this number only grows as you are trying to overclock the CPU.

 

For example, the peak heat dissipation of the Pentium 4 at 4GHz frequency and 1.8V Vcore is about 170W. No air cooler or even the most efficient water-cooling system can provide enough cooling for a processor with a heat dissipation of 170W. What’s more, overclockers don’t need just “enough cooling”, but want to reduce the temperature to the minimum. The maximum frequency the central or graphics processor can work at is not only limited by its technical specifics. The lower the temperature is, the higher the stability ceiling gets. This tendency is strong at any temperatures, so no cooling can be considered absolutely sufficient. Even if we compare the effect of -40°C with that of -80°C, we will see some frequency gain. This fact encourages overclocker enthusiasts to invent new cooling methods for their computer systems.

For today, we have a choice of several methods of the kind (in the order of increasing efficiency):

But what is available for use as off-the-shelf products? Not much, actually: traditional air coolers and water-cooling systems, Peltier units and phase-change systems. Dry ice and liquid nitrogen don’t suit us since you can’t use them constantly (it is passive cooling, with the coolant evaporating). Cascade systems are very efficient, but bulky and difficult to manufacture. I doubt they will be produced in mass quantities for the use in computers in the near future. Water-chiller models are practically unavailable in the market and the efficiency of such an apparatus is questionable. So the most potent CPU cooling solution for today is the phase-change system.

The market of advanced cooling system is split between two companies from Denmark: Asetek (selling its products under the VapoChill trademark) and nVENTIV (earlier known as Chip-Con, the Prometeia trademark). Kryotech, the third player and the company who was the pioneer in the field, has left this market already. There are four systems available today. Asetek offers SE, PE and XE models, while nVENTIV promotes its cooling monster, the Mach II. This review is dedicated to the apex of Asetek’s model range, the VapoChill XE (eXtreme Edition) system, particularly with its version for Socket 478 processors.

What’s Wrong with Jean Peltier?

Striking out systems unaffordable for an average user, we’ve got only two methods to reach temperatures below the environmental (preferably, below zero), which are implemented as ready solutions. They are Peltier units and phase-change systems. Alas, there is not much of a choice here, too.

Today’s Peltier-based systems can only be used to cool the graphics processor (although there are processor solutions from Swiftech). Their problem lies in the very principle of their operation: for example, for taking off 70W of heat and providing a satisfyingly cool thermal environment of the component being cooled, the unit must have wattage of 150-200W depending on its efficiency. And you have to dissipate those 200W (plus 70W from the chip) somehow.

If we set ourselves to cool down a 150W chip (that is not even the upper limit), the wattage of the unit should be as high as 350-400W! Add the original 150W and you have a nice paradox: for cooling an electronic chip of 150W wattage to 0°C or thereabouts, you must assemble a system that really dissipates 500-550W! This is beyond the capabilities even of Swiftech’s cooling system. The cost of the contraption will exceed that of an off-the-shelf phase-change system, but you get poorer results. Moreover, when your computer system consumes 500W or something to cool itself down, you will have to pay more for electricity. Think of this as if you were always having your electric iron turned on with the ensuing consequences to your electricity bill. Jean Peltier was not a computer user, that’s for sure.

Package and Accessories

Asetek ships its VapoChill XE preinstalled into their original and distinguishable system case, specially designed to accommodate the compressor and the condenser of the refrigerating unit. Thus, the package is a very large two-color paper box. Fortunately, the manufacturer added a carry handle, otherwise it would be difficult to transport this machine.

The box conceals the case and a pack of accessories:

Closer Look

The VapoChill system, like any other phase-change cooling solution, consists of four basic parts (the fifth part is the control & monitoring unit called ChillControl):

1. Compressor. Asetek uses compressors from Danfoss, the world’s leading company in this field. VapoChill XE is equipped with a BD35F model. Such compressors are considered the most perfect for today, feeding from the 12V power line and varying the rotational speed in a wide range. BD35F is filled with the R134A gas (this gas is also used in VapoChill SE and XE as well as in nVENTIV Mach II), but for the XE model Asetek selects best specimens of this compressor model, tests them and fills them with a more efficient gas called R404A. This gas contributes a lot to the high overall efficiency of the XE version of the VapoChill.

In fact all three Asetek model differ from one another only by the compressor they use and the gas inside it.

* Power means here the maximum dissipation power of the cooled element at which its temperature is -5°C.

If you are meticulous about details, here is some piece of info about the gases: R134A is a coolant with a boiling point of -26.1°C (under atmospheric pressure). R404A starts to boil at ‑46.6°C under the same pressure, thus increasing the efficiency of the system. R404A is a mixture of three gasses: R125 (44%), R143A (52%) and R134A (4%). Both coolants use synthetic oil, so refilling them is not difficult.

2. Condenser. The system uses a copper 12x12mm condenser that resembles some models of copper heatsinks of water cooling solutions.

3. Capillary tube.

4. Evaporator. This is the most arguable solution from the team of Danish engineers. Its obvious advantage is its small size and easy installation, but its small volume somewhat reduces the overall system efficiency. Anyway, the evaporator of the VapoChill system is a well-thought construction – just take a look at its scheme as drawn by the manufacturer.

All versions of the VapoChill come with the same evaporator, but each system is supplied with a CPU kit that helps you to install the evaporator onto a certain type of processor socket. For today, there are Socket A and Socket 478 versions as well as the universal CPU kit for Sockets 478/754/939/940. The modularity of the construction allows upgrading it for another platform by simply changing the CPU kit (its price is similar to that of an ordinary CPU air cooler).

The lagging is implemented by means of traditional neoprene pipes and plates. The CPU kit also includes one or two 4W heater elements, which are glued to the processor and the mainboard (at the back side) to prevent condensing by generating some heat. These heaters reduce the efficiency of the system just a little, by a few degrees, while the stability grows significantly, so the manufacturer advises that you don’t turn them off.

The necessary cooling is provided by:

The two ordinary versions of the VapoChill come equipped with only one 120mm fan on the condenser, but the higher load on the condenser made it heat up more and called for a more serious cooling solution.

5. ChillControl rev.2.0. All electronic stuffing of the system is placed on one circuit board of the control and monitoring unit called ChillControl. The board carries:

All VapoChill XE units come with the ChillControl of the new 2.0 version, with a more convenient placement of the connectors.

In fact, the VapoChill XE is a VapoChill SE filled with a more efficient coolant and with two additional fans. On the outside, its standard version is distinguishable for its black case with a side window.

Case Design

The VapoChill case is easily recognizable with its characteristic barrel-like front panel and the overall massiveness. A closer inspection produces an ambiguous impression, but let us start from the very beginning.

The dimensions of the case nearly match those of a full-tower system case, but the volume of the computer part is closer to the midi-tower form-factor.

The front panel is made of plastic painted with powder dark-gray ink. The decoration panel covers the ChillControl indicator – it is made of obscure translucent plastic. There are elegant vent slits at the top and bottom of the panel. Three (!) 5.25” bays and one curtained 3.5” bay are placed in the center. The case seems to be larger visually because of its width: the bays are simply lost against the background. The two-color VapoChill logo is placed above the bays. The Power button looks exactly like the Reset one, you just have to learn that the Power is above Reset. By the way, you also use the Reset button to switch between the modes of the ChillControl indicator. The front panel is larger than the case itself, so there are special decorative plates at the side panels fastened to it. Overall, the panel looks original and refined, although the plastic of the buttons is of surprisingly low-quality.

You may like or dislike the exterior of VapoChill cases, but I’m sure you won’t be indifferent to it. I personally like it, but other people say the white variant looks better, while some are strongly against the “barrel”. Anyway, this is a matter of taste.

The chassis is covered with three metal plates: two side panels that open on the “computer” part of the system and one U-shaped panel for the compressor bay. The panels are made of thick (over 1.0mm) steel – the manufacturer didn’t try to save on metal. All panels are painted with powder black ink. The right panel has an acrylic window, glued from the inside. This is not the most elegant window I’ve seen, but it fits into the overall style. There are vent holes in the left side of the U-shaped panel, against the installed 80mm fan. Side panels are fastened with four thumbscrews while the U-shaped panel with four ordinary screws (the user is not supposed to take it off often).

The construction of the chassis differs greatly from ordinary system cases. It is made of stainless steel and is nearly polished. The metal is as thick as in the side panels and this explains the heavy weight of the device.

The top accommodates the compressor and the condenser of the refrigerator unit. There are vent holes for 120mm fans at the front and rear ends. The evaporator is attached to the processor through a hole in the chassis.

The bottom of the chassis is an ATX case for the computer system components. 5.25” and 3.5” devices are fastened in their bays with ordinary screws, without any removable baskets. There are five landing places for hard disk drives. The ChillControl unit, the electronic brain of the VapoChill, nestles nearby – you access it from the left part of the case. At the bottom of the front panel there are spots for two 80mm fans or one 120mm blower. Besides their main function – blowing cold air into the system – these fans will help in cooling your hard disk drives.

The mounting tray for the mainboard is inserted into the groves of the chassis at the back and is screwed up at the front. To detach the tray, you have to remove the left panel of the case. Instead of the traditional screw fastening, the mainboard is attached to the tray with plastic clips that facilitate its installation.

The back panel looks somewhat weird. Yes, the PSU is installed vertically, like in mini-towers of those old days when the ATX form-factor had just appeared. Back then, it was difficult to cool and assemble the system. In this case, it’s not that bad. First, the case is wider than usual and there is enough space above the mainboard. Second, the processor is cooled by the evaporator and needs no additional air cooling at all. Third, a number of modern PSUs use two- or three-fan systems. The internal fan thus performs the role of a blower in the system, removing hot air from the North Bridge area, memory chips and the backside of the graphics card.

The chassis is intended for installation of ATX and Extended ATX mainboards (although I can hardly think of a server with such cooling). It is equipped with seven slots for expansion cards. All seven brackets come fastened with screws. There are three vent grids nearby, 50x50mm big. If you feel like doing this, you can install three small fans there, although the efficiency of this solution will be close to nothing.

The case stands on four rubber legs to prevent vibration. You actually feel this vibration much more than with an ordinary case, because of the working compressor.

The overall impression from the case is like it’s done for ages – the manufacturer was not stingy about metal. However, the ease of use doesn’t quite match the status of the VapoChill (although there are no serious issues about it). This case is analogous to entry-level models in its functionality, save for the removed mainboard plate. We might expect something more from a case for a high-end system.

Assembly Process

The user manual is a voluminous book with every thing covered in much detail. If you follow the manufacturer’s recommendations and read through the instructions thoroughly, you’ll get through the installation process very easily. As for me, it took me about an hour or more to assemble the system for the first time. Second and third times were much easier and faster (for example, if you are swapping mainboards and processors every other day). Processor replacement is performed in about the same time as it takes to do the same thing with an air cooler (or a little more as the screws have a long thread and the evaporator should be removed, too), so it’s quite possible to do it in two-three minutes with some skill.

So I took the manual and followed it scrupulously. The brief instructions are listed below (I omit various routine actions like attaching connectors). That’s for evaluation purposes only, as you must read the original user manual thoroughly before attempting an installation:

That’s all. You’re past the main problem. The rest of the system is assembled as usual, the PSU is connected to the ChillControl and to the mainboard through an extension cord and the system is ready to start up.

Exploitation and ChillControl

After you press the Power button and turn the compressor on, you should wait for 1-2 minutes for the computer to power up. This time is necessary for the compressor to create the required temperature on the evaporator (by default, the system is powered on when the CPU temperature is -5°C). During this interval, the processor hangs on in the status analogous to the locked Reset button. This is necessary for the processor not to overheat while the compressor spins up.

The ChillControl indicator wakes up with all its lights in the first two-three seconds, but then it shows an animated line as the compressor creates the operational temperature. Starting from +10°C, the indicator displays the precise temperature of the evaporator. At -5°C, the system starts up, while the temperature goes down as low as the compressor allows. If you push the Reset button anytime when the system is working, you change the indicator’s mode (to reset the system, you press and hold the button for 2 seconds). There are four indication modes, which you can browse through: the evaporator temperature, the data of the second thermal diode (optional), the compressor rotation speed, and the CPU “clock rate”. The last thing is set up manually from 0000 to 9999MHz using the ChillControl control panel.

To control the system parameters, you connect a cable with the COM interface to the ChillControl and boot up from the floppy disk enclosed with the system. The program on the floppy looks like a BIOS Setup and allows controlling the following parameters:

“Hold temp” is the evaporator temperature (as displayed on the ChillControl indicator) the system will be trying to maintain. By default, it is set to -30°C and it’s practically impossible to go below this temperature until you try to experiment with low room temperatures. The lowest evaporator temperature I could achieve at +5°C room temperature was -37°C.

“Start PC at” is the evaporator temperature at which the system starts up. By default, it is -5°C and there’s no need to change this parameter.

“Warning at” is the evaporator temperature beyond which a warning signal sounds. It is fixed at 0°C and you can’t change it. Well, you don’t really need to.

“Shut down at” is the critical temperature at which the system shuts down. You can set it from 0°C to +30°C.

“Fan1 Speed” and “Fan2 Speed” control the fan rotation speed from 30 to 100%.

“Pin Heater Load” is a tricky option as it allows controlling the heat of the heater elements from 0% to 100% (4W each). In theory, this option can help to somewhat reduce the temperature, but at the risk of getting some condensate. Asetek honestly warns you that you experiment with this setting at your own risk.

“CPU speed (MHz)”. You set up manually the CPU frequency from 0000 to 9999MHz as displayed on the ChillControl indicator.

“Temperature Output”. Toggle between degrees of Celsius and Fahrenheit.

“Default view”. This setting determines the data type which is displayed on the indicator when the system is powered on (see above).

Some parameters are available for monitoring. Frankly speaking, they are quite useless.

“Temp. 1 (CPU)”. The evaporator temperature.

“Temp. 2 (External)”. The temperature of the optional external thermal diode.

“Compressor Speed”. The rotation speed of the compressor.

“Compressor type”. The compressor type (it writes “BD-TYPE”).

“Fan Speed (Condenser)”. The rotation speed of the fan on the condenser, in percents.

“Fan Speed (External)”. The rotation speed of the additional fan, in percents.

“ChillControl Status”. The status of the ChillControl unit.

“Current Error” informs you about the current error if one occurs.

“Current Display View” is the type of info displayed on the indicator.

“Connection Status” is something quite mysterious.

“Firmware Version” that the ChillControl has.

For making the life of the user easier in our advanced era, there is a Windows version of this software.

I should confess that I encountered various problems with the system during my three months of using it and several times I had to assemble/disassemble it, which were mostly related to the powering up process. However, all problems were solved after I had read the user manual and corrected my own mistakes. So you shouldn’t hope for taking the VapoChill barehanded, without reading the manual at all.

One more problem concerns the power supply unit. The compressor is supplied by the 12V power from the system PSU. The average power consumption of the compressor is about 100W, or 8A on the 12V line. This is actually very high, and only the best PSUs can bear this load in addition to the load of the computer system itself. So only the Antec TrueControl 550 showed an impeccable performance in our tests. Even at the system startup, when the compressor consumes as much as 150W for a short while, the voltage of the +12V line was practically the same – another testimony to the high quality of this PSU. I tried to use a lower-wattage Sirtec 420W and it upheld the system for several days without any problems, but the +12V voltage would go down to 11.3-11.5W under a load. Just for fun, I tried to use a very-low-quality 250W Codegen PSU ($8 retail price) and only connected the compressor to it (without the system), but the voltage slipped down from 11.9V to 8.0V (!) in two seconds and the PSU broke off completely.

So it is a necessity for you to buy a really high-quality PSU to use with the VapoChill. Users say that, besides the units from Sirtec, 400W PSUs from Zalman can also supply the compressor reliably. You can solve this problem completely (never worrying about the PSU at all in the future) by using a 500W PSU like an Antec TrueControl 550 or PCPower&Cooling 510 DeLuxe.

Testbed Configuration

Other system components don’t affect the performance in this case.

Performance

In fact, we are not that much interested in the temperature numbers as in traditional coolers tests, although we will provide them too. The maximum stable clock rate of the processor is the main goal of our test session.

Intel Pentium 4 3.20C GHz

Rated frequency: 3200MHz
Clock rate achieved with air cooling: 3550MHz
Clock rate achieved with water cooling: 3660MHz
Water-chiller: 3780MHz
VapoChill XE: 4065MHz

The testing methods are the following: in every case we made sure the processor worked stable at the maximum possible frequency. The frequency at which you can boot up the OS to make a screenshot is in average about 100MHz higher than that.

Until the water-chiller, we left the Vcore intact (1.55V), as overclocking was only worse with a higher voltage.

Air cooling was provided by an Akasa 670-CuBL cooler with a copper sole. The average CPU temperature under a workload was about 60°C.

Water cooling was represented by a highly efficient home-made system. The CPU temperature under a workload and at the above-indicated frequency was about 32-35°C.

The temperature of the liquid inside water-chiller was +7°C, the CPU temperature being +16°C, and the Vcore = 1.75V.

When using the VapoChill, we set the Vcore to 1.75V. Further increase of the core voltage resulted in worse overclocking. The CPU temperature at 4065MHz was about -5°C. The maximum frequency at which I managed to make a WCPUiD screenshot was 4181MHz.

Intel Pentium 4 Extreme Edition 3.20GHz

Rated frequency: 3200MHz
Air cooling: 3650MHz
Water cooling: 3760MHz
VapoChill XE: 4150MHz
Testing methodology is the same.

Air cooling: full-copper coolers CoolerMaster Fujiyama and CoolerMaster Aero 4. The Vcore was standard.

Water cooling: a highly efficient home-made system. The temperature under a workload was in between +32 and +35°C at the above-mentioned frequency. The Vcore = 1.75V.

VapoChill XE allowed us to reach a stable frequency of 4150MHz with the Vcore = 1.8V. The processor was working at 4200MHz in our tests, but this mode was not absolutely stable. It passed the Hexus PiFast test at 4234MHz. The maximum frequency for making a screenshot was 4260MHz.

Intel Pentium 4 3.40? GHz

Rated frequency: 3400MHz
Air cooling: 3750MHz
Water cooling: 3810MHz
VapoChill XE: 4215MHz
The testing conditions are the same.

Air cooling: a copper cooler CoolerMaster Fujiyama. The Vcore = 1.70V.

Water cooling: a highly efficient home-made system. The temperature under a workload was in between +32 and +35°C at the above-mentioned frequency. The Vcore = 1.75V.

We stopped at 4215MHz with the VapoChill – that’s an impressive result.

Not bad, yeah? We’ll have to wait long to see off-the-shelf products reaching similar frequencies. Moreover, only phase-change systems allow modern processors to get beyond 4GHz. This makes them indispensable where extreme performance is required: in professional applications, in hi-end gaming stations and in systems of computer enthusiasts.

Benchmarks Results

We couldn’t miss the opportunity of benchmarking our processors in the extreme operational mode. So we compared CPUs on Northwood, Prescott and Gallatin cores at two frequencies, 3200MHz and 4200MHz.

For the test at the increased frequency, we set the CPU multiplier to x14: all tested processors were engineering samples with an unlocked multiplier. The system memory was working in the 5:4 mode (240MHz) with 2-5-2-2 timings and Vmem = 3.35V.

One of the reasons why we tested the processors at the extremely high frequencies is our intention to check out the potential of the Prescott core at the frequencies it is going to reach in the future. In fact, this processor only starts living a full life at 4GHz and higher. At such clock rates, the negative influence of the very long pipeline diminishes and the performance per megahertz starts to grow.

By the way, our comparative test is not quite correct with respect to the new core: neither Northwood nor Gallatin will ever reach 4.2GHz by any means save for extreme cooling, while we will surely see a Prescott-based Pentium 4 4.2GHz with a simple air cooler in the near future.

There is also not much sense in presenting the benchmark results for the Pentium 4 3.4C as it is always slightly faster than the 3.2C in the regular mode due to the higher clock rate, while at 4200MHz both Northwood-based processors show exactly the same performance, like might have been expected.

The results of processor tests at their rated frequencies brought no surprises: Pentium 4 Extreme Edition is always ahead, while the 3.2C model and the Prescott swap positions in different tests.

We noted the following tendency (also in tests that we carried out but didn’t publish here): the Prescott is faster in multimedia tasks and at archiving (and the optimization for the SSE3 extension will make the gap even wider!), it also wins in SPEC ViewPerf and runs neck and neck with Northwood in games, but loses in raw computational power (Hexus PiFast, ScienceMark 2.0). It is strange but SiSoft Sandra doesn’t show any definite growth of the memory speed (although the Prescott is better than the competitors in all modes), while the “honest” CacheBurst says it is quite hefty.

PCMark 2002 is sensitive to the amount of cache memory in the first hand, and the Prescott is just in between Northwood and Gallatin.

As for our experiment with extreme overclocking of all three 3.2GHz processors from Intel, I believe some comments are necessary.

Of course, Pentium 4 Extreme Edition is beyond competition: this processor shows impressive results across all the benchmarks. However, we are more interested in the tests where Prescott loses to Northwood. And we see the gap getting smaller as the frequency grows: twice in 3DMark 2001 and Enemy Territory! At the same time, the gap between Northwood and Gallatin remains the same, if not bigger.

This evidence is important for understanding the idea implied in Prescott. This core just can’t show its best here and now as it is intended for different frequencies and operational environments. We can draw a parallel with the first Pentium 4 and Pentium III. At the same clock frequency Pentium 4 would lose in many applications, but as its frequency grew higher it would reach a level of performance, unattainable by processors of the previous generation.

Conclusion

Highs:

Lows: