Home
Ask a Question
Secure Orders
Latest News
Reviews
Projects
Customer Gallery
Guides
FAQ
Live Testing

Radiical Ultra Extreme Water Cooling Kit and the Cooler Master Stacker Case Project 5

Radiical Ultra Extreme Water Cooling Kit and the Cooler Master Stacker Case Project

Build Log Date 19 January 2005

Test Bed

Abit Kv8 Motherboard
Amd 64 2800+ CPU
Corsair Ram 256 megs
Samsung 80 Gig hard drive
ATI Radeon 9600 Graphics card
Radiical Ultra Extreme Kit with Swiftech MCP 600 pump (details here)

Procedure and Results

We originally set out to use software temperature monitoring to establish a commonly used system of temperature measurement but were unable to get a happy medium in the accuracy of the results. We found MBM5 and uGuru results to be so inaccurate that we had to discard them as useless. I am sure this comes as no surprise to anyone who is involved in computers at any level. The problem is of course there are very few other options for temperature measurement of CPU's fitted with integrated heat spreaders.

The fitting of integrated heat spreaders to CPU's may have saved the CPU manufacturers large amounts of money in the replacement of damaged cores but it provides additional headaches for the developers of cooling solutions. We had considered removal of the IHS to get at the AMD64 core but decided this idea was impractical for future comparisons with other systems and CPU's where the IHS was still fitted. The best case scenario, it was decided,was to accept that by installing a probe into the core of the water block base we would be able to achieve the highest possible accuracy of measurement. The major shortcoming of this temperature measurement device would be that the probe would be cooled by water returning from the radiator on one side while being heated by the IHS on the other. Not ideal but far better than a software solution which was at best "out there" with the results it was reporting.

We settled on the use of PT100 temperature probes imbedded within the system in a constant monitoring state. The prototype water block measurement probe is shown here in Figure One.

Figure One

In the finalized test unit the probe was machined into the copper so that a complete base was presented to the CPU surface. That is to say that a hole was milled into the side of the copper base and the probe interference fitted to ensure good contact.

Standardization of the probes was achieved by measuring the temperature of water flowing through the system without a load and then with a load added. Each probe within the system; viz the water inlet temp probe; water outlet temp probe and CPU temp probe; were then monitored and the data from each probe collected and compared. To avoid test inaccuracy the probes were also individually sampled with a Delta ohm HD9215 hand held. An accuracy variation of 0.1 of a degree was felt to be acceptable.

Test fittings to ensure accuracy of the water block contact to CPU face were carried out. Figure One illustrates the fitting pattern or CPU footprint accuracy. Please note that the amount of TIM material used here is merely for illustrative purposes.

The test unit was run in a "loaded" and "unloaded" state for five days to establish the final pre testing mount for consistency of temperature recording and to permit TIM curing. Having completed the pre nuptials we then began temperature recording. There still remained the question of ambient temperature variations during the course of a day which had to be factored into the measurement data. To solve this problem we elected to record the temperatures without climate control so as to produce a graph like the one given here as Graph One for "baseline" of CPU at idle performance.

CPU Throttling

Most computer literate people are aware that CPU manufacturers prefer to present CPU temperature data in their own way. In fact all of the current top CPU manufacturers have developed methods of throttling back CPU functions so that the resting or idle temperature is presented at a much lower temperature. This throttling back gives the average consumer the impression of successful thermal management (the "feel good" syndrome). The problem with thermal management of this kind lies with the excessive load temperatures expressed by the same software temperature monitoring program when the CPU is given back its full capacity. To avoid thermal throttling during the duration of this testing we used Folding@Home to produce the necessary 100% CPU load we required.

Graph Two illustrates the temperatures for "fully loaded" totally passive performance.

Ambient Temperature and Hybrids

The one big problem with passive temperature control is the range of air ambient temperatures. Here in Australia we have large swings in the high ambient's in the summer months. We also have extended periods of high ambient temperatures in the 40's and above. It is highly unlikely, therefore, that a purely passive system would able to handle the variance of temperatures we experience. In order to reach low noise levels whilst maintaining performance it is only logical to look at hybridized systems.

Perhaps the most obvious solution for a hybridized "noise versus performance" based system is to thermostatically limit the radiator fans. Fans thermostatically tied to the system or ambient temperature would switch on in a staged fashion until maximum fan usage was reached. This is merely a refinement of the use of the fan controller.

Graph Three presents the data for a single fan used on the vertical radiator in this system. This data is intended to simulate a single thermostatically controlled fan which has been set to switch on (hour 9 in the graph) at a defined water temperature. This data (derived from Graph Two) represents the semi passive (hybrid) cooling system discussed above.

The temperature drop observed in Graph Three (hours 9-11) is then added to by the switching on of the remaining three fans spaced at one hour intervals. The results for the staged switch on of the fans is shown in Graph Four.

It is interesting to note that after fan one is switched on the addition of the remaining fans have increasingly less affect on the systems temperature as it approaches ambient. This is not an unexpected finding.

Stage Two Passive Overclocking

The question that has been posed by many of the overclockers that I meet is can you overclock passively. To answer this question I suppose you have to say yes and no. The real answer comes from the questions how much do you intend to overclock by? What voltage will you be using? In short by what value will you increase the heat load of your CPU?

We elected to try passive overclocking using default voltage on this AMD64 2800+ by overclocking it to 2250Mhz. This constitutes a 25% overclock on this CPU.

The fans were then turned off and the machine allowed to idle. The temp was monitored every hour until no further rises in temp were noted. The graph below indicates the temperatures reached.

A single fan was then switched on on the vertical radiator and the temperature recorded again after one hour.

The rapid return to close to ambient temperatures using a single fan reinforces the passive/hybrid possibilities of the Radiical Ultra Extreme water cooling kit.

The results seem to be very clear. Large multi radiator systems such as the Radiical Ultra Extreme allow a large margin of thermal management with plenty of reserve cooling capacity for when it is really needed. In the main passive cooling can be used for the greater part of the year with the need for some thermostatically controlled active cooling on hot or CPU intensive days or where more adventurous overclocking is being used for bench marking or the like.

Conclusion

When we first started out with this project the primary aim was to establish if it was possible to run an AMD64 CPU in a totally passive state using the Radiical Ultra Extreme Kit. Simply put we wanted to achieve a fanless state at ambient room temperatures. The answer to the question we have found is yes it is.

The test unit constructed on these pages is quite happy to run at idle with default voltage and default clocks all day long with a simple ten degrees temperature differential to ambient.

Simulating a fully stressed CPU state with Folding@Home, the test unit was then also quite able to carry out "full load" work accurately and without error at mid twenty degree ambient's with no significant temperature stress again totally passively.

In the next stage we have successfully overclocked this CPU by a margin of 25% passively. The overclocked CPU was able to successfully run Folding@Home without error.

Comments

Overall the most significant finding is that for the majority of time it is possible to run the Radiical Ultra Extreme Kit passively. It is also very clear that if the system fans were set up to respond thermostatically only a single fan would be required to control aggressive changes in the CPU temperature. Further fan usage could be switched on when preset water temperatures were reached either at very high ambient's or during very CPU intensive periods.

CM Stacker Makeover