The ultimate “How-To” – Bill Adams
This article will look at bench testing methods and equipment useful in ascertaining the performance capability of waterblocks used as part of a CPU cooling system.
There are several different approaches to such testing; a unit’s in-place performance can be measured under a single set of (sometimes unquantified) conditions, or it can be “bench tested” under a range of conditions such that the unit’s performance capability can be predicted in any likely installation (assuming of course that the other components’ performance capability data are available).
A previous article (HERE) described the relative merits of bench vs. system testing. Here we will consider a somewhat complex bench testing equipment setup and a testing methodology designed to characterize a waterblock’s capabilities.
It is hoped that the test method outlined here might be a ‘talking point’ for some future collaboration between interested testing parties. After a general agreement on test methods, the real issue of cross-calibration could be addressed to enable the comparison of test results from different sources. And farther yet down the road, standardized test procedures could be drafted and validated with conventional round-robin testing.
Readers are advised that this is an article about testing – using waterblocks for sure – but about testing; the equipment and the procedures. Many technical terms are used without definition (in the interest of brevity); when an unknown phrase is encountered, please use Google as an understanding of much of what follows will likely depend on an accurate understanding of that particular word or phrase.
An understanding of how a waterblock functions is helpful in visualizing the temperature reactions of the various elements in the heat dissipation path from the CPU to the coolant.
Waterblock (wb) is the generic name of the small heat exchanger placed on the CPU die face to reduce its temperature. The wb consists of a baseplate (bp) – which may be flat or finned for increased area, whose face is pressed in contact with the die and which removes the die’s heat by conduction; and a water box through which the coolant flows in contact with the backside of the bp. The heat is transferred from the backside of the bp (and fins if any) to the coolant by convection.
Note that at equilibrium all the heat from the CPU (less the secondary heat path losses) will be transferred to the wb – at any flow rate. The heat in must equal the heat out or it is not at equilibrium; and the temperatures will continue to rise (or fall) until equilibrium is established. This is true of all wbs:
It is quite incorrect to say that a good wb transfers ‘more’ heat than a poor wb, it is the temperature gradient across the wb that is different.
The wb’s bp temperature (gradient) will be materially affected by changes in the coolant flow rate.
A higher coolant flow rate (implying higher velocity) will improve the heat convection effectiveness within the wb.
This will enable equilibrium to be reached at a lower temperature differential between the coolant and the bp, which in turn will reduce the temperature of the bp, and hence the temperature of the die itself.
Differences in wb efficacy will manifest themselves in this way as well, the more efficient wb (with better convection rates) will have a decreased temperature differential between the coolant and the bp. Differences in conduction ‘resistance’ through the bp will also differentiate between wbs’ performance.
Note that the coolant temperature rise is a consequence of the heat input, divided by the flow rate. The coolant temperature rise is therefore the same for all wbs at the same heat load and flow rate. (This is a greatly and excessively simplified description of the interaction between Fluid Mechanics and Thermodynamics in a 2 inch box. Many more words are needed to adequately describe the functions and relationships within a wb – but this article is not about wb design, just their testing.)
In between the wb’s bp and the die’s face is a thermal joint, which typically contains some form of Thermal Interface Material (TIM) to improve the heat transfer from one surface to another.
This joint will have the effect of a series resistance to the “flow” of heat and will always result in a temperature offset of some magnitude. As no wb can be tested without also including a TIM joint, the characterization, and more importantly CONTROL, of this variable becomes essential. A method of so doing, and the one utilized here, was described in this article (HERE).
To recapitulate then; under consideration is the CPU die face (the source of the heat), the thermal joint (filled with a TIM), the waterblock’s baseplate (between the TIM joint and the coolant), and the coolant.
{mospagebreak}
The term “C/W” has become common in the watercooling community to describe the performance of waterblocks, the units for which are °C/Watts.
The concept is simple enough, the temperature (rise) divided by the power, but as with many simplifications – much is glossed over. “C”, the temperature rise, will be defined as the CPU-to-air or the CPU-to-coolant temperature differential depending on whether the WCing system or the wb proper is being evaluated.
The power, “W”, is vastly more difficult to quantify given the imperfect understanding of the actual CPU heat being generated, and the ignoring by the WCing community of the effects of secondary heat path cooling. As a result, the actual power input by a CPU to the wb is not known. People are guessing, and using exaggerated power values (with inaccurate temperatures) to arrive at fantasy “C/W”s.
There are people who will take umbrage at their ‘facts’ being called “guessing”. Should I say guesstimating instead? Fine, please refer the writer to any actual quantification of CPU ‘settings’ where the heat into the coolant has been accurately measured.
The writer has searched high and low but never a wisp of solid data encountered. The so-called Watt Calculators are mere smoke and mirrors, one referring to something else and to something else yet again. Intel and AMD described values are to define the required capacity of the attached sinks, with all the ‘fudge factors’ appropriate to such.
An informative schematic of the secondary heat path losses (as they relate to the “W” quantification difficulty) can be seen HERE. Worth noting is that the secondary losses will be different between motherboards due to their specific layout, and dependant on the effects of in-case air circulation as well.
While “C/W” is often described as a thermal resistance, such usage is imprecise. As with any term, the meaning is a matter of convention, compounded in this case by dissimilarity between the thermodynamic and electrical models. A lucid explanation is of this dichotomy is described HERE.
When considering a wb’s “C/W” (the CPU-to-coolant temperature differential divided by the power), there are two constituent elements; the TIM joint, and the wb itself. Each of these behaves differently as the coolant flow or applied power is varied. The die, bp, and coolant inlet temperatures can be measured by means of temperature sensors mounted within the heat die, bp, and coolant inlet stream as shown schematically below:
Where “C” is defined as the temperature differential between the die and the (inlet) coolant temperature, “T” is herein defined as the temperature differential between the wb bp and the (inlet) coolant temperature. Similarly to the “C/W”, the T/W can be obtained by dividing the temperature differential between the die and the (inlet) coolant temperature by the power. The TIM joint’s C/W then is the difference between the wb’s overall “C/W” and its T/W. These relationships are shown to the left in the above figure.
A ‘typical’ plot of the die and bp temperature differential obtained with the above described sensors is shown below, at four different coolant flow rates, with measurements taken at four different heat loads.
The difference between the die and the bp temperatures shown above is the temperature offset due to the thermal impedance of the TIM joint. Unfortunately this temperature measurement is not so precise as it might seem due to the temperatures not being those actually of the substrates’ faces, and more significantly the somewhat imprecise location of the bp’s sensor’s hole over the die area.
Yet while not absolutely precise, or directly comparable with the values from other wb’s bp sensors, the data is still quite informative. A rough check of the data’s validity is to observe the trendlines’ intersection at 0,0.
Below is a plot of “C/W”, T/W, and the TIM joint C/W:
As can be easily seen, the WCer’s hack “C/W” term consists of the wb T/W plus the TIM joint C/W. So why is a distinction being made between the two? Because they vary in response to changes in different variables. The T/W is significantly affected by the coolant flow rate as the above graph shows, while the TIM joint C/W is slightly influenced by the (nominal) temperature but hugely by the thermal grease application and applied compression – which will be a constant for a given installation, as shown.
Additionally, and of crucial importance, the relative magnitude of the wb’s T/W and the TIM joints C/W must be appreciated. While the above graph suggests the wb’s T/W as being about ½ that of the TIM joint’s C/W, such is overstated due to the sensor placement described earlier. A reasonable approximation is that the wb’s T/W and the TIM joint’s C/W are nominally equal in magnitude (This is for a ‘good’ TIM well applied; obviously a ‘poor’ TIM joint would change this nominal ratio.)
Is the T/W and TIM joint C/W difficult to measure ? No, given the equipment – only holes are needed, IF the bp is of sufficient thickness. Below are some assorted wbs having drilled bps, note the Cooltech wb in the lower right with the RTD sensor hole entering through the top cover.
{mospagebreak}
Understanding a bit about how wbs function and how their performance is characterized, it can be deduced that the variables of interest are:
- Applied power – as this is the source of the heat,
- Coolant – as this is the means by which the heat is removed, and
- Temperatures associated with this process.
In the world a wb designer might wish for; of interest would be:
- Power consumed/heat generated by the CPU,
- Identification and quantification of all secondary path CPU heat losses,
- CPU’s die face temperature profile,
- bp face temperature profile at the TIM joint,
- bp’s back side temperature profile in contact with the coolant,
- Temperature profile of the coolant stream, and even
- Mass flow rate profile of the coolant stream within the wb.
Well, none of these are to be had – at least not with the equipment reasonably available (translate = within my budget).
What can be measured (and of greater relevance to wb users) is:
- Power applied to a heat die simulator,
- Heat die temperature in close proximity to it’s face,
- ‘Internal’ bp temperature,
- ‘Average’ coolant wb inlet and outlet temperatures,
- Pressure differential across the wb, and
- Flow rate.
The accuracy of these measurements is very much a function of the equipment and calibration budget; and as with most things, you get what you pay for – no free lunch in the testing world.
The ‘value’ of any measurement is inexorably tied to it’s accuracy, which is expressed by means of a measurement uncertainty value related to the equipment’s absolute capability (defined normally by the manufacturer) and the equipment’s calibration ‘status’. If ‘measurements’ are made with devices not capable of the measurement uncertainty determination, then while a measured value may be ‘known’, nothing is known about the ‘goodness’ of that value.
From a bench testing perspective, such measurements are pointless. As is oft said in the computer world, GIGO (Garbage In = Garbage Out). This article is based on the assumption that calibrated test equipment is used for that purpose and range for which it was designed. This is not about Digital Docs, buckets, and weekend projects.
Measurement uncertainty arises from a number of sources; but, while of great significance to this and all testing activities, is not the subject of this article. Indeed, the complexity of the numerous aspects involved make a one paragraph, one page, or even one article summary more prone to error than the effort justifies – certainly by this writer.
Interested readers should review the NIST site where an accurate description of the US definitions and calculation methods are reviewed. It is hoped that a more qualified individual will step forward and write an article reviewing the aspects of resolution, accuracy, precision, and uncertainty as applicable to watercooling measurements.
The equipment list was described in the article previously cited HERE, but several aspects of the actual setup warrant additional explanation. It is the independent control of both the applied heat and the coolant that enable the wb’s loading to be manipulated, and it is the instrumentation of the heat die and the wb that permits the characterization of the wb’s capabilities.
The wb heating circuit consists of a regulated DC power supply (having a remote sensing circuit) providing power to a nichrome wire cartridge heater in the copper slug of the heat die simulator (the construction details of which can be seen HERE, though now somewhat modified).
The voltage and current are measured to mV and mA respectively. The heat die is extensively insulated to substantially reduce secondary path heat losses and an internal temperature RTD sensor within the insulation is monitored to verify that equilibrium is attained and maintained at each step through the test cycle. The applied power can be held to within an indicated ±0.01W of the set point.
While the applied power to the heater is easily and accurately measured, quantifying the actual heat conveyed by the wb is considerably more demanding. The calculation of Q, the heat added, is straightforward if the flow rate is known; but the required temperature resolution is quite high given the relatively low delta-Ts involved, in particular those below 0.15°C.
With a temperature indicator having only 0.01° resolution, the up-ticks and down-ticks preclude ‘high’ accuracy. Nonetheless, with judicious filtering, good data can be averaged to yield an indication of the losses to reasonably characterize the actual heat conveyed by the wb at the different applied power levels.
For this heat die, as presently insulated, the secondary heat losses are: 1.2%, 1.6%, 2.0% and 2.5% at 40W, 70W, 100W and 130W respectively. (Not considered in this secondary loss characterization are the effects of coolant friction or pressure loss within the wb.) All graphs are presently made using actual (corrected) power values, some of the older graphs have not been so adjusted, as the correction applies equally to all test results.
The image below shows the heat die in it’s phenolic casing. To the left can be seen the RTD which is mechanically restrained with a cut-away compression fitting; set with ASII into a hole in a shoulder on the side of the riser, which is the heat die face proper.
All wbs are mounted using studs having the 4 motherboard mounting hole spacing, with an adaptor plate if necessary. It should be noted that the heat die is off-center with respect to the mounting holes – this is as AMD describes it. Upon receipt, a wb is measured and the centerline of the CPU’s location is marked and aligned with the heat die during installation.
{mospagebreak}
The cooling circuit consists of a closed loop having a substantial reservoir with an in-line pump, and a liquid/liquid flat plate heat exchanger. Coolant temperature control is maintained with a recirculating chiller whose output is independently pumped through the heat exchanger – which is also placed within the chiller’s bath. The coolant temperature at the wb inlet can be held to within an indicated ±0.02°C of the set point.
Below is an image of the wb’s connection and instrumentation, prior to the lines being insulated.
The coolant entering and leaving the wb passes through two ½” ss swivels (hidden under the white plastic sleeve in the center) which facilitate rotating the assembly to initially purge it of air, and subsequently to drain the coolant back into the lines prior to disassembly. Each cross below the swivels has a pressure tap and an insulated RTD mounting in its branches, while the run is connected to the wb’s inlet and outlet. A 3/8″ cross with copper tubing connections is used to minimize the internal volume difference at the instrument connections.
On the side of the wb can be seen a 0.065″ diameter RTD probe inserted at the lower level of the bp into a just slightly larger hole which extends to the center of the die area. The end of the hole is slightly tapered and ASII is used at the tip. This sensor placement is not possible with all wbs, and the placement variability between wbs precludes the direct comparison of the temperatures; but the bp temperature information so obtained is quite useful nonetheless.
All temperature measurements are taken with class “A” 4-wire RTDs and a calibrated thermometer having 0.01°C resolution; when the display flickers between two values the higher is recorded, between three values the center one is taken. Temperatures are recorded only under steady state conditions; which is defined as no change after ½ hour (with prior experience indicating that no change is to be expected after an additional hour). The ambient temperature is also recorded at each measurement point or interval, and controlled to ±1°C.
The entire wb assembly is suspended from springs with an adjustable counterweight to permit its balancing slightly below the level of the heat die face to facilitate the wb’s installation on the heat die. After mounting, the wb/heat die assembly is raised slightly to be clear of the bench top. This arrangement isolates the TIM joint from any loads induced by the hose connections or instrumentation.
The weight of the total heat die assembly is determined and subtracted from the load due to the mobo mounting springs’ compression in the determination of their required total compression.
Below is an image of the flow control valving (after the wb and ahead of the reservoir), shown again without insulation:
The flow rate control is effected by means of ½” and ¾” needle and ball valves in parallel, each used to regulate the flow in their appropriate range. The flow rate is measured with a magnetic flow meter, as this type provides unobstructed flow.
Like most types of flow meters, they have about a 10-fold range of effectiveness; with the ¼” tube, ±0.5% accuracy can be had in the range of 0.3 to 4.0 gpm, but the pressure drop at the higher rates is substantial due to the tube diameter. But all other types of flow meters have much higher flow resistance, and generally much lower accuracy as well. Note that all such flow meters have requirements for unimpeded flow both upstream and downstream of the sensor.
Differential and line pressures are recorded at each measurement point or interval. In order to accurately define the pressure drop across the wb, it is first necessary to determine the loss due to the instrument fittings and wb connections; the appropriate correction may then be applied to the gross measurement across the wb. An example of such a correction chart is shown below, the units shown being those in which the measurements are initially recorded:
{mospagebreak}
The cross-platform direct comparability of test results is a highly desirable goal; both for the convenience of users, as well as to preclude ‘conversion’ errors and faulty assumptions. Standard test conditions should reasonably reflect the use conditions to the extent practicable, and must be clearly stated with the test results.
The ‘standard coolant temperature’ question was addressed a year ago in an article on “Radiator Heat Dissipation Testing” (now in need of some revision) and, for the purpose of radiator testing, 10.00°C above the ambient temperature was selected as the coolant temperature at the radiator inlet.
This was predicated on a somewhat ‘worst case’ scenario and in retrospect was too high; probably something around 5°C would have been more representative, in view of the current trend towards larger more effective radiators. Note that a temperature offset was defined, as that is how radiators work in dissipating the heat to the air.
For wb testing, 25.0°C has been selected as the ‘standard inlet coolant temperature’, based loosely on what a dedicated WCer could achieve in a cool room. This is not a crucial issue, as within several degrees all the temperatures are simply offset by the same amount and the thermal resistances virtually unchanged (well within the range of measurement error, ~±0.05°C with this equipment setup). And as will be shown in a graph in an article on wb testing results, the offset is little changed over the range of 10 to 35°C.
An aside regarding ‘standard temperatures’: Since systems utilizing air as the ultimate heat sink will always be related to the ambient air temperature, several times I have solicited comments on ‘typical’ ambient air temperatures. Responses have come from Norway to Australia and all climes in between; and I believe that one can only select a nominal value and run with it. Every house, indeed room, is different; hence the direct comparison of (presumed!) CPU temps is always a bit pointless.
The applied power testing points are not crucial as a curve is normally plotted from the data, but the assumption as to the heat actually applied to the wb is obviously central to the issue of quantified wb performance.
That the data is directly comparable to other data sets generated with the same test bench (given good procedures, etc.) is clear, but for it to be compared to the results from other test bench setups requires a means of cross-calibration. Despite several attempts, little interest in such has been shown by other testers; hopefully this will change.
The reason that the various heat dies need to be cross-calibrated is due to the differing efficiencies of the various heating mechanisms, and more significantly, the (probable) substantially different secondary heat path losses. All of the applied power is not conveyed as heat to the wb, and a truly accurate (heat balance) calculation of the heat conveyed by the coolant really necessitates greater resolution than the 0.01°C (I have) available.
Far easier to make a simple cheap ‘calibration’ heat die and send it around to the interested testers so they may establish (using a rigidly defined procedure) the ‘offset’ applicable to their particular heat die. Note that this ‘offset’ is most probably a curve rather than a constant value.
The (heater input) power levels selected are 40, 70, 100, and 130 Watts. After lengthy discussions with individuals more knowledgeable than I about CPU heat generation, it would appear that these Wattages are about 70% of say a “Radiate” value in terms of equivalent CPU/heat die temperatures (To be quite clear, 70 Watts of regulated DC power applied to this heat die is nominally equal to 100 Watts per Radiate on someone else’s computer. Hmmm… not a whole lot of precision in that statement.)
The flow rate’s effect is also normally described with a curve, so again the actual flow rate measurement points are not critical, but the range should span that of the wb’s potential use with sufficient values to reasonably extrapolate the curve and be able to discern anomalous readings. The flow rates selected were 0.3, 0.5, 1.0, 1.5, 2.0 and 3.0 gpm (or the maximum attainable with this pump and setup if below 3.0 gpm).
It is (reluctantly by the grey-haired author) accepted that the SI measurement units are the norm and, except for the way the pressure transducers’ digital indicators are programmed (in psi), all of the instruments indicate such values. The units to be generally used will be lpm, mH2O, °C, and Watts; except for some older graphs with other units which have not been re-plotted.
The ‘standard test conditions‘ utilized for the assessment of installation and mounting characteristics, and for a ‘rough’ comparison of wbs ( and also for troubleshooting the test bench equipment and setup) are:
- 70 Watts applied power,
- Coolant flow rate of 3.8lpm (1.0gpm),
- Coolant temperature of 25.0°C at the wb inlet,
- ~10kgf (22lbf) applied across the TIM joint (90% of the AMD maximum), and
- Ambient temperature controlled to 26°, ±1°C.
This common reference point enables prior setups to be referenced and immediately identifies possible problems before a great deal of testing time has been expended. Hopefully other wb testers will adopt these same ‘standard conditions’ to enable the comparison of test results from different sources.
{mospagebreak}
The following is an outline of the ‘Waterblock Test Procedure‘:
- Inspect and measure the wb
- Lap the baseplate as required
- Install in fixture and purge of air
- Determine the pressure drop at the defined flow rate test points
- Set test equipment to the ‘standard test conditions‘ and take first measurement after steady state conditions are attained
- Repeat 5 times, and establish a preliminary mean “C/W”
- Continue testing per the ‘standard test conditions‘ (for a minimum of 10 iterations in total), and when a “C/W” is observed close to the mean – then
- Reset the test equipment to the lowest flow rate and, after steady state conditions are attained, take measurements at incrementally higher flow rates to establish a complete wb “C/W” flowrate response curve
The wb to be tested is cleaned (boiling works well – but not with plastic), measured and the bp lapped. One must establish if it is the wb’s (design) capability that is being measured, or the wb as shipped; lap, or not, on that basis. In virtually all cases, lapping the bp will improve the wb’s performance (except for the Swiftech wbs).
The wb is connected to the coolant lines and, at the maximum flow rate, rotated several times through 180° to purge it of air, after which the counter-balance weight is adjusted so the wb bp is suspended somewhat below below the top of the heat die.
The thermal grease is applied in a thin layer and the wb is carefully mounted on the heat die using a dial caliper to verify the installed height of the (previously calibrated) springs. The counter-balance is then adjusted to elevate the assembly slightly off of the bench top.
The initial performance tests are to determine the pressure drop across the wb in relation to the coolant flow rate. This is done with the coolant at 25.0°C at flow rates of 68, 114, 227, 341, 454, and 681 lph (0.3, 0.5, 1.0, 1.5, 2.0, and 3.0 gpm).
Initial “C/W” readings are made with a 70W applied load at a flow rate of 3.79 lpm (1.0 gpm) – the ‘standard test conditions‘; once the system has come to equilibrium, ~4 hours typically, 5 hours to take a reading.
This single measurement sequence is repeated at least 6 times to get an understanding of the range of C/W values to be expected, and for a total of not less than 10 mountings recorded by the end of testing. This is the most important part of the wb testing sequence, as it is imperative that a (statistically) valid “C/W” be determined.
Given then that the TIM joint constitutes approximately half of the total thermal impedance between the die and the coolant, and the potential and quite real variability of this element, it becomes hugely important to control this variation, as it quite easily can obscure the difference between wbs, whose measurement is the purpose of the testing.
In the the article previously cited HERE, a large number of measures are described to limit and control this TIM joint variation, but nonetheless such will occur. As part of the testing process then, the TIM joint variation must be assessed, and this is done by making repeated installations using exactly the same procedure. An example of such for several wbs is shown below (some data removed for clarity):
As previously described, 10 installs of a wb are made to understand the range and to somewhat validate a mean nominal “C/W” (as a single point value) under ‘standard conditions’. The heat load can be very tightly controlled if desired (although not strictly necessary), but the die temperatures have a range of between ±0.2 to ±0.4°C, despite the most strenuous efforts at controlling the wb’s mounting.
Even so, the above groupings do indicate that there are genuine differences between wbs. The range has a significant impact on the presumed accuracy of “C/W” curves since the curve itself will be offset depending on the ‘quality’ of the TIM joint as assembled in that specific instance. It is reasonable to conclude that any “C/W” without a defined range, mean, and described standard deviation is somewhat speculative.
The reason for the ‘speculative’ label should be clear: If the best value for one wb is compared to the worst of another’s, then a conclusion could be drawn quite the opposite from what an adequate description of more comprehensive testing would justify. At the very least, test results should be ‘qualified’ with some (brief) characterization of their statistical validity. (/rant)
After a half dozen mountings, when a moderately ‘low’ (good) reading is encountered, the flow rate is set to the minimum. After an appropriate temperature stabilization period (~4 hours), data collection is started to generate a set of C/W curves. Note that the minimum number of mountings should not be less than 10 in any case.
Hysteresis (defined in NCSL RP-12 as the maximum difference between the upscale and downscale readings on the same artifact during a full range traverse in each direction) is a factor, unless many many hours are allowed between readings. For this reason, the data points for curves should be generated always in the same direction.
A shorter stabilization period has been found running from hot to cold than vice versa (1.5 vs. 2.5 hours between readings). For this reason, after the initial 5 hr stabilization at the low flow rate, readings are taken from low flow to high on 2 hour intervals.
A minimum of 10 installs is made with a wb at 70W and 3.79 lpm, and two C/W curves on different installs are normally run at 70W. Anomalous readings are not difficult to identify once a baseline has been identified. As is apparent, it is a very time-consuming test procedure.
A modification to the die heater assembly is in process that hopefully may reduce the waiting period needed to achieve steady state conditions. A ‘guard heater’ (similar to those used in conventional calorimeters) is being built using a PID controller to pre-heat and maintain the temperature of the insulated heat die assembly’s exterior. It is hoped that, at the least, the initial 5 hr wait may be substantially reduced and possibly the period between readings as well.
But wait, there is yet another adjustment/correction that must be applied to the “C/W” curve, IF one desires to compare it to that of another wb. If there is a difference between the “C/W” curve value at the ‘standard conditions’, and the mean value for such determined from the 10 installs, then the “C/W” curve itself must be offset by the appropriate amount.
This adjustment is necessary so that all wbs are being compared on an equal (and statistically validated) basis. Yes, such is a chore; but without this step, the ‘test’ results may amount to no more than a reflection of the tester’s bias.
{mospagebreak}
The following advice is offered to those interested in testing, interested in evaluating test data, interested in improving their test equipment, or even setting up a test bench from scratch:
- Tip # 1 – Google for info, it’s all there
- Tip # 2 – eBay for equipment; be patient and it will eventually appear (read descriptions very carefully); also LabX
- Tip # 3 – When evaluating testing results, evaluate FIRST the equipment used; crappy equipment = crappy results, always
- Tip # 4 – If a tester will not describe their equipment or procedures, something is being hidden
- Tip # 5 – All measurements have a measure of ‘uncertainty’ (aka ‘accuracy’), if undefined see #3
- Tip # 6 – Flow meters: Magnetic best, paddle wheel (Hall effect) satisfactory, all others too imprecise (tried ’em all)
- Tip # 7 – Lab chillers: Recirculating best, just about impossible to control coolant temperature without one (I would not even try)
- Tip # 8 – Pressure gauges: The needed resolution requires a digital display
- Tip # 9 – Digital thermometers: Consider carefully the intended goal; an accuracy in tenths requires resolution in hundredths!
- Tip #10 – Be prepared to spend a LOT of time learning how to setup and operate the equipment
- Tip #11 – Be prepared to spend some more time learning how to understand and interpret the resultant data
- Tip #12 – Waterblock testing -> avoid it like the plague
The cause of the difficulty with wb testing is that very small changes are being measured on very small components (Who in their right mind would need to know the coolant temperature rise for a 40W load at a flow rate of 11 lpm ? = 0.05°C !) This necessitates rather good equipment, well considered procedures meticulously followed, and a willingness to repeat extremely time-consuming tests over and over to establish confidence in the results. While such testing starts off as ‘fun’, it quickly becomes work, and not long after simply frustrating.
While the equipment and methods described in this article may seem hopelessly complex and even impractical, one needs to consider the intended purpose and use for the data so developed. The results are engineering data, of use to systems and product designers. The independent control of the heating and cooling functions – while being able to measure the effect – is in fact the only way that a design prototype may be validated. And also the only way that a wb’s performance attributes can be quantified.
Some end users may have an interest in bench testing and data, but the far greater majority will view such and derive an imperfect understanding – at best. Engineering data yields numbers, not ‘winners’. In the case of wb test results, it is always the case that a wb that excels in one respect will be inferior in another.
It is the task of the WCing system designer to select the wb whose performance characteristics most closely match the WCing system characteristics (the overall interaction of the pump, radiator, hoses, and wb). And so we return to the complicated testing needed to obtain such distinctions. If one wants to do ‘engineering’, one must deal with the numbers.
Future articles will describe “Waterblock Bench Testing Results”, “Modded and DIY Waterblock Performance” (testing always in progress – and never complete), and perhaps even “Watercooling Systems Component Selection”. It is in the area of system design that WCers have the greatest potential for improvement, not to select pieces that have complementary capabilities.
Thanks must be expressed to Eric Braeden, Antoine Dechaume, Ralph Nelson, Derek Peck, Les Round, and Dave Smith for their insightful comments and constructive criticism; and of course, also to my family for the too many months of humoring the ‘mad scientist’ cursing his machines while endlessly rebuilding them.
be cool
Be the first to comment