Radiator Heat Dissipation Testing 2076

SUMMARY: This article will describe a poor-man’s calorimeter; the equipment and the procedures used to generate comparative heat dissipation performance data for several different types and sizes of radiators of interest to the overclocking community. It should be noted that all of the data was actually recorded by (reasonably calibrated) instruments, NOT manufacturer’s “ratings” which too often have little relation to the item’s real-world performance.

Radiators are one of the principal components (with the pump and waterblock) of a CPU watercooling system, and may have a substantial effect on the system’s performance. May, rather than will, because watercooling systems are designed empirically as they consist of elements whose specific performance is often poorly defined (or understood, or applied, or all of the foregoing). The goal of cooling “back” to ambient temperature necessitates substantial overcapacity and the “design process” is often a pragmatic serial selection of components to identify, and then replace, the limiting element. (In my hot-rodding days we used to call this “part replacement troubleshooting”.)

Radiators do not function independently; rather their specific performance is dependant on each of the other system components’ capabilities and optimization. In particular, the selection and installation of fans is of primary importance, pumps are obviously crucial, and if the waterblock can’t extract the CPU’s heat then nothing ever has a chance to perform. But this article is not about fans or pumps; their outputs are considered simply as variables in the determination of a radiator’s heat dissipation capability.

No amount of testing ever changed a product for the better, or made it any worse; but properly designed and performed tests can generate hard data to aid in the selection of a component suitable for the service conditions. Radiators (forced air-cooled crossflow heat exchangers) are tested under arbitrarily defined conditions to provide a consistent means of determining their absolute, and relative, performance capability. Good tests will utilize conditions closely approximating the actual conditions of use to minimize extrapolation and provide higher confidence levels for proposed applications.

Obviously a radiator’s specific capacity will be determined by its frontal area, tube size and number and configuration, fin dimensions and spacing, construction materials and methods, etc.; while its actual performance will be a function of how these “design” choices interact with the service conditions. The principal variables affecting a radiator’s performance are liquid and air flow rates, and the temperature difference between them. A good description of the “Thermal Calculations for Heat Exchangers” can be found HERE.

These variables are not open-ended as there are practical limits to the flow rates and temperatures; at high flow rates the resistance becomes disproportionately high, and the temperatures are normally dictated by the application and environment. A decreasing rate of response to the change of a variable is indicated by an asymptotic curve (approaching a straight line / limiting value), the so-called “point of diminishing returns”. Performance data plotted as curves are useful for graphically indicating reasonable “limits” and some excellent examples for radiators can be seen HERE.

So, given some well designed tests generating reproducible performance data; how does one determine what a radiator’s “rating” might/could/should be? A Designer, Tester, Salesman, and User might all have different opinions, but the common ground should be that the radiator will “work” in the described conditions. Ah, the User too has an obligation; to know their (approximate) heat load, air flow rate and temperature, and the coolant flow rate and temperature in order to select an appropriate radiator. Sorry Sports, the canned phrase “They’re all good” won’t work here; numbers are needed.

From the excellent test data from Lytron (previously cited), the flow regimes to be investigated were fairly well understood and uniform data recording points were established. The testing at (exaggeratedly) high air and water flow rates was necessary only to rough out the “limits” of practicality.

The highest liquid temps tested (120F) are the maximum (and may even exceed those) appropriate for the aquarium/utility pumps used by almost all watercoolers, they were included to make the data somewhat more complete. Plastics have an upper service temp limit that is related to the applied stress, and this applies particularly to the vinyl tubing so prevalent. (The “cure” for vinyl tubing stress relaxation is called a hose clamp.)

The goal of CPU watercooling is to cool the baseplate of the waterblock as much as possible, and to do this a coolant temperature of only slightly above ambient is sought. As the efficiency of both the waterblock and radiator is greatest with the largest possible temperature difference, this would suggest lower coolant flow rates to maximize the heat absorption and rejection. But in fact, assuming reasonably effective cooling by the radiator, high flow rates will yield greater cooling due to the exposure of more cold coolant to the waterblock and hot coolant to the radiator. (Of course it may be that the waterblock is not effectively transferring the heat …, but that’s another article.)

In practice, both the air and coolant flow rates can be varied by the selection (or control) of the fan(s) and pump. These choices may affect the ambient (in case) air temperature, and will affect the coolant temperature. A reasonable quantification of the thermal load can developed (CPU + TEC + ?). A comprehensive solution would necessitate dealing with the coolant system’s hydraulic characteristics (waterblock, lines, fittings, etc., and an approximated radiator) to size the pump for a design maximum flow rate.

Given the total absence of waterblock performance data, one could assume that it is thermally “transparent” and passes 100% of the CPU’s (+ TEC’s) heat into the coolant, which heat must be then dissipated by the radiator. It should be appreciated that an attempt to return high temperature coolant to something close to the ambient temperature is a “sizeable” undertaking and becomes progressively more difficult as the coolant temperatures rise – recall the diminishing returns comment above.

It is difficult, if not impossible, to assess radiators without becoming very involved with pumps and their characteristics. The first task is to measure their output and this is easily done by timing the filling of a gallon jug at various heights (Revelation #1); then one checks the flow through the test loop without a radiator (Revelation #2); finally one checks the flow through the complete system (Revelation #3). These flow rates could be called low, very low, and vanishing; bring on the next pump. The plastic centrifugal pumps used by overclockers have relatively large impeller clearances and are intolerant of intake backpressure, and not much more capable on the discharge side.

The flow through the system will be the pump capacity at the total head resulting from the sum of the individual flow resistances of all the components; with units of similar cooling capacities, the lowest flow resistance radiator is to be preferred to maximize the coolant flow. One might presume that a oil cooling radiator (designed for the high head output of a positive displacement gear or sliding vane pump) and a water cooling radiator (utilizing a centrifugal pump) would have rather different head losses, in addition to other design, construction, and material differences.

The equipment included:

• Haake A82 recirculating heater/chiller as the heat source (1000W heat capacity w/0.01C control);
• Fluke 2190A Digital Thermometer and Y2001 Selector, plus numerous thermocouples;
• 3 pumps, Haake 25L/min, Little Giant 500gph, Flojet 4300-142 – 4gpm, 45 psi;
• 4 flow meters; Brooks, Fischer Porter, Lake Monitors, ElectoThermal – 0.05 to 5.0 gpm;
• Dwyer Capsuhelic #4220B, 20psi liquid differential pressure gauge;
• Fans; 2 Panaflo FBH-12G12L – 136CFM, 4 Nidec TA600DC – 960 cfm;
• Dwyer Magnehelic # 2002, 2in H2O air differential pressure gauge
• Hygrometer;
• Kunz 441S thermal anemometer;
• A bucket of fittings, yards of hose, and way too much ducting, pop rivets and 200 mph tape.

Just a tad beyond a Styrofoam cup and thermometer.

The testing procedure was fairly straightforward:

• Liquid utilized was distilled water with 8% Water Wetter added.
• Set water and air flow rates.
• Desired water temperature set at bath, then adjusted after 1/2 hr to achieve desired temperature at the radiator inlet under steady-state conditions.
• Temps and flow rates recorded every 5 minutes for 1/4 hr to demonstrate stability.
• Humidity, barometric pressure, and differential pressures recorded at end of run.

[ insert setup.jpg here ]

The following are descriptions of the radiators tested, which will subsequently be referred to as Rad A, Rad B, etc. (No doubt the mfgrs will inform me of any errors in the descriptions). Note that these descriptions are substantially incomplete with respect to the description of water flow area and heat dissipation area (including the fin surface area); both essentially radiator design parameters.

Round Continuous Tube:

• PN:
• Opening (Area): 5 x 5 in. = (25 sq.in.)
• Thickness (Volume): 3/4 in. = (18.8 cu.in.)
• Connections: 3/8 in. barbed hose connectors
• Tube: 4 parallel 3/8 in. copper tubes in series, on 1 1/4 in. centers
• Fins: swedged on corrugated aluminum fins, 16 per in.
• Outside Dimensions: 5 x 6 1/2 x 1 3/8 in. (w/connections)

• PN:
• Opening (Area): 3 1/8 x 3 5/16 in. = (10.4 sq.in.)
• Thickness (Volume): 1 in. = (10.4 cu.in.)
• Connections: 1/8 in. CTS straight tube
• Tube: 8 offset 1/8 in. parallel copper tubes in series, on 5/8 in. centers
• Fins: swedged on flat copper fins, 15 per in.
• Outside Dimensions: 5 1/2 (w/connections) x 3 9/16 x 1 1/4 in.

• PN:
• Opening (Area): 4 1/4 x 3 7/8 in. = (16.5 sq.in.)
• Thickness (Volume): 4 1/2 in = (74.1 cu.in.)
• Connections: 1/4 in. CTS straight tube
• Tube: 15 offset parallel 1/4 in. copper tubes in series, on 1 in. centers
• Fins: swedged on flat aluminum fins, 8 per in.
• Outside Dimensions: 6 3/4 x 4 3/4 x 5 in. (w/connections)

• PN:
• Opening (Area): 6 x 4 in. = (24 sq.in.)
• Thickness (Volume): 3 1/2 in. = (84 cu.in.)
• Connections: 1/4 in. CTS straight tube (modified)
• Tube: 16 offset 1/4 in. parallel copper tubes in series, on 1 in. centers
• Fins: swedged on flat aluminum fins, 10 per in.
• Outside Dimensions: 8 1/4 (w/connections) x 4 1/8 x 4 1/8 in.

Flat Parallel Tube:

Rad E – unknown
• PN: 03-5371656 SN: 43860
• Opening (Area): 5 1/2 x 5 1/2 in. = (30.3 sq.in.)
• Thickness (Volume): 1 1/8 in. thick = (32.2 cu.In.)
• Connections: 1/4 in. NPT
• Tube: 30 flat copper tubes in 15 parallel pairs with 5/16 in. gap
• Fins: Soldered folded copper fins, 16 per in.
• Outside Dimensions: 7 3/4 x 6 3/8 x 2 3/8 in. (w/connections)

• PN: Big Momma
• Opening (Area): 6 x 6 in. = (36 sq.in.)
• Thickness (Volume): 2 in. thick = (72 cu.in.)
• Connections: barbed 3/8 in. hose connections
• Tube: 13 full thickness corrugated brass “plates”
• Fins: Soldered slit folded copper fins, 12 per in.
• Outside Dimensions: 7 3/8 (w/connections) x 6 x 3 in. (w/connections)

• PN: 18115-4002 SN: ACO2343
• Opening (Area): 9 5/8 x 4 3/4 in. = (45.7 sq.in.)
• Thickness (Volume): 1 7/8 in. thick = (85.7 cu.In.)
• Connections: 1/2 in. tubing compression fittings
• Tube: 16 parallel single flat 1 7/8 in. wide aluminum tubes with 3/16 in. gap
• Fins: Brazed folded aluminum fins, 19 per in.
• Outside Dimensions: 14 1/2 (w/brackets) x 6 (w/connections) x 2 in.

• PN: Black Ice
• Opening (Area): 4 7/8 x 4 3/4 in. = (23.2 sq.in.)
• Thickness (Volume): 0.625 in. thick = (14.5 cu.in.)
• Connections: straight 1/4 in. CTS, 1/2 in, long
• Tube: 3 parallel flat 1/ 2 in. wide copper tubes, 4 series passes
• Fins: folded soldered copper fins, 20 per in.
• Outside Dimensions: 5 1/4 x 6 x 1 1/2 in. (w/connections)

Stacked plate (without fins):

Rad I – OCWC

• PN: Miracle Midget Wonder
• Opening (Area): 4 x 4 in. = (16 sq.in.)
• Thickness (Volume): 3/4 in. thick = (14.3 cu.in.)
• Connections: barbed 3/8 in. hose connections
• Tube: 19 diagonal channel stacked aluminum plates, 2 1/2 in. long
• Fins: none, connected to adjacent plates at diagonals
• Outside Dimensions: 6 3/8 (w/connections) x 4 x 3/4 in.

Rad J – OCWC

• PN: Double Miracle
• Opening (Area): 8 x 4 in. = (32 sq.in.)
• Thickness (Volume): 3/4 in. thick = (24 cu.in.)
• Connections: barbed 3/8 in. hose connections
• Tube: 38 diagonal channel stacked aluminum plates, 2 1/2 in. long
• Fins: none, connected to adjacent plates at diagonals
• Outside Dimensions: 10 3/8 (w/connections) x 4 x 3/4 in.

A note on copper tubing sizes (CTS): The table below describes the dimensions, but it seems clear that the 3/8 in. “tubing” size often attributed to watercooling radiators is, in fact, 1/4 in. CTS – whose OD nicely (if loosely) accepts 3/8 in. ID vinyl tubing.

It is worthwhile to spend a moment to appreciate the significant differences between these radiators, not just in terms of their physical size but also in the different designs, materials, and fabrication methods utilized. Both absolute and “normalized” (for size) performance differences are to be expected.

The assessment of a radiator’s capability, and its selection as a part of a cooling system, cannot be done without an understanding of it’s hydraulic characteristics. Since flow resistance is a function of velocity (among other things), it is a simple task to obtain the data to plot the liquid head loss (in psi) vs. the flow rate (in gpm). Likewise the air flow can be varied to develop the data for a plot of air pressure drop (in in. of H2O) vs. air flow (in cfm).

The Flow Resistance graphs below are a plot of:

• Water or Air Side Pressure Drop (psi or in.H2O), vs.
• Water or Air Flow (gpm or cfm)

where the (unadjusted for size) data is plotted for direct comparison of the different radiators.

The Radiator Cooling graphs below are a plot of the:

• Temperature decrease (^F), vs.
• Cubic feet per minute of air (cfm)

where for each set of tests the temperature difference between the coolant radiator inlet temperature and the ambient air temperature was set at 20^F or 40^F, the air crossflow varied from 5 to 300 cfm/sq.in., and each radiator was tested at coolant flow rates of 1/2, 1, and 2 gallons per minute (gpm).

Since the purpose of this exercise was to compare radiator performance, it was decided to plot the cooling capability per cubic inch of air flow area, thus enabling the comparison of the efficiency of differently sized units.

The Thermal Performance graphs below are a plot of the:

• Temperature decrease per cubic inch of air flow area (^F/cu.in.), vs.
• Cubic feet per minute of air per square inch of air flow area (cfm/sq.in.)

where for each set of tests the temperature difference between the coolant radiator inlet temperature and the ambient air temperature was set at 20^F or 40^F, the air crossflow varied from 5 to 300 cfm/sq.in., and each radiator was tested at coolant flow rates of 1/2, 1, and 2 gallons per minute (gpm).

This article is flirting with a thermal characterization known as Specific Dissipation (SD), which is defined as the heat transfer rate of a heat exchanger divided by the maximum temperature difference across the heat exchanger. As I have not the ability to display equations, nor most readers the inclination to view them, those interested in the “theory” are referred to an SAE paper, 2000-01-0579, on “The Effect of Changes in Ambient and Coolant Radiator Inlet Temperatures and Coolant Flowrate on Specific Dissipation” which can be found HERE.

The test results graphed above are directly applicable only to those radiators actually tested, but should serve as a useful guide to estimating the performance of others of similar construction and materials. The “performance values” plotted are on a “per cubic inch” or “per square inch” basis and cannot be applied to a different type of radiator than that tested, but can be readily extrapolated to characterize similar units of different sizes by calculating the volume or area to estimate a unit’s cooling capability.

The data suggests the following:

Joe, I have no idea if I want to include this “stuff”
I could write pages on radiator design, radiator selection, watercooling “system” design, etc.; but these topics are not the subject of this article. So ????

A cost effective system is one that could be said to be efficiently performing its intended function; its got to work, and its elements’ capabilities should be reasonably balanced. In reviewing smaller radiators designed for water cooling (but not “watercooling”) some trends become apparent:

• Air flow recommendations are limited to a pressure drop of 0.25 in.H2O or less.
• Air flow rates range between and cfm per sq in of air crossflow area.
• Liquid flow recommendations are limited to a pressure drop of 7 psi or less.
• Liquid flow rates range between 1/2 and 2 gpm.

The above are not limits, one can select a fan or pump to push more or less – but its output may decrease the efficiency of some other part of the system. Increased flow rates will always result in increased cooling – up to a “point” (diminishing returns again). The costs are greater expenditures, size, and noise. Note that if the waterblock, for example, is at the limit of its effectiveness (for whatever reason), an increase in flow may result in decreased cooling.

Radiators cannot tell if the air is being pushed or pulled; such depends only on the fan, presumably selected for those characteristics. Two pushing and pulling fans together do NOT provide twice the air flow, and fans should be set at least 1 in. away from the radiator. Two small radiators in series will double the hydraulic flow resistance and have a marked effect on (reduction of) pump output. Larger lines, and fittings with chamfered edges, (with a larger pump ?) may be cheaper than upsizing the whole system again later.

It must be observed that the manufacturers’ “ratings” for pumps and fans in no way reflect their installed performance. This is not really deceit as much as a reflection of the very large differences between the manufacturer’s test conditions and the user’s service conditions. The principal use for “ratings” is to compare relative performance between units, rather than to believe that the “rated” flow is that actually being achieved.

Bill Adams