Processor Cooling Report for the practical course Chemieingenieurwesen I WS06/07 Z¨ urich, January 16, 2007
Students:
Francisco Jos´e Guerra Mill´an
[email protected] Andrea Michel
[email protected]
Assistants:
Max Wohlwend HCI D182.5 Neil Osterwalder HCI E105
Abstract Overheating is nowadays a huge challenge for the computer industry. Without a cooling system, working on a computer would be barely impossible, due to the system crashing down continuously. The aim of this exc periment was to study the heat transfer phenomenon by cooling an Intel c
Celeron D 2.8Ghz processor by different means. This processor emitted 17,48 Js and 31,95 Js and had an efficiency of 76,15% and 70,69% at normal c and full load respectively.1 Two fan coolers (an Intel original and a c Super-Silent), a ThermalTake heat exchanger, two radiators (black and white) and a water cooling system (operated at different flows) were used. As for the water system, as expected, increasing the water flow results on higher convection coefficients, meaning that the cooling capacity also increases. It was also possible to see the linear dependency between the flow and the coefficient. Water is by far the best cooling system. After anac lyzing different cooling devices, the Intel original fan cooler resulted the c best one, followed by a Super-Silent fan cooler. Surprisingly, a black rac diator had a better “cooling effect” than a ThermalTake heat exchanger with an implemented fan. 1 Values
obtained with the block of copper.
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
Contents 1 Introduction and Theory
3
2 Experimental Procedure
5
3 Results and Discussion 3.1 Block of copper . . . . 3.2 Water Cooling System 3.3 Cooling Devices . . . . 3.4 Comparison . . . . . . 4 Conclusions and Outlook
F. Guerra, A.Michel
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
6 6 6 9 10 11
2
ETH Z¨ urich
1
Chemieingenieurwesen I, WS06/07
Introduction and Theory
Overheating is nowadays one of the most important problems. Not only as a global matter, but also in our daily lives. As an example, computer processors can reach extremely high temperatures and represent a physical danger for the final user. To avoid fire and other damages, computers are programmed to crash down when the processor reaches a certain temperature, causing considerable loss of information. Therefore, CPU (Central Processing Unit) cooling devices are essential for computer manufacturers to offer an optimal performance. The aim of this experiment was to compare different CPU cooling devices by measuring the heat emitted by the processor at normal and full CPU load. To compare the different heat exchangers, the overall heat transfer coefficient was calculated for each device. By using different coolers it is also possible to observe the three different heat transfer mechanisms, known as conduction, convection and radiation. Conduction[9] is described as the flow of thermal energy through a solid from a higher- to a lower-temperature region. Heat conduction occurs by atomic or molecular interactions. Steady-state conduction is said to exist when the temperature at all locations in a substance is constant with time, as in the case of heat flow through a uniform wall. It can be expressed with the following equation: ∆T ∆Q = −kA (1.1) ∆t ∆x where: Q = heat [J] t = time [s] W k = conductivity constant mK A = transversal surface area m2 T = temperature [K] x = thickness of the body through which the heat is passing m2
Convection[7] is heat energy transfer between a solid phase and a fluid phase when there is a temperature difference between the fluid and the solid. Convection is a combination of conduction and the transfer of thermal energy by circulation or movement of the hot particles to cooler areas in a material medium. This movement occurs from, to or within a fluid. Convection occurs in two forms: natural and forced convection. It can be described as follows: q = h (Ts − T∞ )
(1.2)
where: q = heat flux rate Js h = convection coefficient mW 2K Ts = surface temperature [K] T∞ = ambient temperature [K]
F. Guerra, A.Michel
3
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
Radiation[4] is transfer of heat through electromagnetic radiation in the heat spectrum. All objects radiate heat, unless they are at absolute zero. No medium is necessary for radiation to occur; radiation works even in a perfect vacuum. A prime example of this is the energy of the Sun, which travels through the vacuum of space before warming the earth. The energy radiated is given by the Stefan-Boltzmann Law: Pr = eσA T 4 − Tc4 (1.3) where: Pr = net radiated power Js e = emissivity [] σ = Stefan-Boltzmann constant = 5.67 ∗ 10−8 mW 2K4 Tc = temperature of surroundings [K] A general temperature profile is shown in Figure 1.1. The temperature profile is normally linear. The heat flow occurs from the highest to the lowest temperature zone. Therefore, to calculate the heat transfer we basically only need a temperature difference.
Figure 1.1: General temperature profile for heat transfer. To calculate the heat emited by the processor, equation (1.4) was used. For simplification purposes, it is assumed that the specific heat capacity is temperature independent. Z T Q=m cp dT (1.4) T0
where: m = mass of the solid [kg] h i J cp = specific heat capacity kgK The efficiency of a determinate heat exchanger is calculated with the equation below: Ptot − q η= · 100 (1.5) Ptot where: η = efficiency [%] Ptot = overall computer power Js
F. Guerra, A.Michel
4
ETH Z¨ urich
2
Chemieingenieurwesen I, WS06/07
Experimental Procedure
c c To analyze the heat transfer phenomenon, a CPU with an Intel Celeron D 2.8Ghz processor was given. It was possible to fit different cooler systems over the processor. Four thermometers to measure the processor and ambient c temperature were also provided. With a computer program called LabView , the temperatures were graphed on-screen live and the load for the processor could be switched between normal (almost no CPU activity) and full (100% CPU activity). With another single-purpose application, it was possible to see the CPU and Main Board temperatures. This were measured internally and due to the lack of information, it was unsure where exactly they were taken. As an advice from Mr. Wohlwend those measurement were seen as doubtful and were not used for the calculation of the results. A power meter was also used to measure the power consumption of the computer. All the data gathered by c the computer was exported to an Excel file for further use. Due to the fast overheating of the processor, almost immediately after turning on the computer a cooler had to be placed over the processor to avoid a sudden crash-down. All measurement (except where indicated) were taken at normal and full load. As a first step, the heat emitted was measured with a block of copper. This step had to be preformed quite fast, because the copper can get extremely c hot very soon. After that, different cooling systems were analyzed; an Intel c c
Original fan cooler, a Super-Silent fan cooler, a ThermalTake heat exchanger (without and with an implemented fan), a black and a white radiator and a water cooling system (operated at different flows). After placing a determinate cooler and finishing with its measurements, the next one could be placed without turning off the computer. It’s important to mention, that this had to be done quickly, to avoid a crash-down, due to the processor temperature going over 110 ◦ C. If this happened, the computer had to be started again.
Figure 2.1: Schematically draw of the experiment layout.
Figure 2.1 shows a schematically representation of the different coolers used c c in this experiment. A stays for the Intel and Super-Silent fan coolers, B c
represents the ThermalTake without ventilator, C shows the black radiator and D pictures the water cooling system. In the case of both fan coolers the temperatures were measured in the ribs and in the metallic case of the computer. c For the case of the ThermalTake and the radiators, they were measured near the processor and in the coolest extremes of the device. For the water cooling
F. Guerra, A.Michel
5
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
system temperatures were taken before and after the water had gone through the plate over the processor.
3
Results and Discussion
c With the help of Microsoft Excel and the equations (1.2), (1.4) and (1.5) the folowing results were obtained. The area of the heat transfer surface was A = 0.0009m2 .
3.1
Block of copper
J [10]), the mass of the block of copper With the specific heat (cp = 0.385 gK (m = 347.36g) and equation (1.4), it was possible to calculate the heat transfer rate. Equation (1.5) was used to calculated the efficiency. Table 3.1 shows the obtained results for normal and full load. The efficiency obtained is surprisingly high, considering that no more than 30% of the power gets lost in form of heat, while an engine looses approximately 70% to 80% of the energy provided [8].
Table 3.1: Results for η based on the heat transfered to a block of copper. CPU mode normal load full load
Power [W ] 73.3 109
Jq s
17.48 31.95
η [%] 76.15 70.69
It is clear to see, that the amount of heat transfered increases at full load. This is, due to the higher power consumption of the computer while working at full load. Figure 3.1 shows the temperature profiles at normal and full load. It shows the linear dependency between the temperature and the time. Therefore, adding a trend-line allows us to calculate the heat per unit of time. The values of R2 , close to 1 indicate that our results are reliable.
3.2
Water Cooling System
The implemented water cooling system operated at the flow rates shown in Tables 3.2 and 3.3. Using equation (1.2) the convection coefficient h and the parameter hA were calculated. It is important to mention, that the obtained values are calculated with the temperature difference of the water, but does not consider the temperature of the plate where the real heat transfer takes place. The results are shown in Tables 3.2 and 3.3. At normal load (Table 3.2) h and hA clearly increase proportional to the water flow. As expected, with a larger flow, the processor gets better cooled. Nevertheless, at some point, this “cooling capacity” should reach a top value, because the water can only take a determinate amount of heat. At full load (Table 3.3) the trend for the values of h and hA is the same than at normal load (Table 3.2). For this case, one could also expect a top value for F. Guerra, A.Michel
6
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
Figure 3.1: Temperature profiles for the block of copper at normal and full load.
Table 3.2: Results for the water cooling system at normal load. Flow mL
h W
hA W
s
m2 K
K
1.73 4.53 9.60 14.00 18.67
7.13 13.18 26.11 39.98 48.85
0.0064 0.0119 0.0235 0.0360 0.0440
the “cooling capacity” of the water system. The results show that this was not yet reached. Comparing Tables 3.2 and 3.3 it is possible to see, that the convection coefficient gets bigger at full processor load. An exception to this affirmation is found at the flow rate of 1.73 mL s . From the Tables 3.2 and 3.3 it is possible to determine, that a higher value of h or hA represents a better “cooling effect”. Figure 3.2 shows the convection coefficient as a function of the water flow. As said, h grows proportionally to the flow, but with the help of a trend-line it is also possible to confirm that it is linearly proportional. The values of R2 near to 1 give a certain reliability to the assumption. Nevertheless it is possible to see, that at normal load the trend could be reaching the mentioned top value. It is not really possible to determine if the variation of the fourth point of the curve is merely an experimental error or a reliable value. If the trends for normal and full load are compared and detailed analyzed, it is possible to conclude at first sight that the “cooling effect” is better at full load. Having said that a higher value of h represents a better cooling, the line with the highest values should be for the better “cooling effect”. This could also be interpreted on a different way.
F. Guerra, A.Michel
7
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
Table 3.3: Results for the water cooling system at full load. Flow mL
h W
s
m2 K
K
1.73 4.53 9.60 14.00 18.67
6.79 14.27 28.21 41.50 54.55
0.0061 0.0128 0.0254 0.0373 0.0491
hA W
Figure 3.2: Convection coefficient as a function of the water flow.
The results obtained are based on the temperature difference between the water going into the heat exchange surface and the water going out of that area. The real heat exchange between the processor and the implemented cooling system (at the metal plate) is not being taken on account. Obviously, at full load, the heat transfered from the processor to the convection plate will be higher. This causes the water to get warmer at full load and therefore the ∆T results bigger. When entered into equation (1.2) the resulting values get bigger.This does not strictly means that the the processor had a lower temperature. J With the specific heat (cp = 4.1813 gK [5]), the flow rate and equation (1.4), it was also possible to calculate the heat emitted by the processor per unit of time. Equation (1.5) was used to calculated the efficiency. Table 3.4 summarizes the results for the efficiency of the system based on the water. For each flow the heat emitted and the efficiency were calculated. Interestingly the results at 1.73 mL s seem to be quite different than all the other flows, but are very similar to the results obtained with the block of copper (Table 3.1). Therefore it is assumed, that the most confident results are obtained at lower flows. This deviation on the values can be because at a lower flow rate the system works with a more accurate ∆T . A mean efficiency was calculated
F. Guerra, A.Michel
8
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
Table 3.4: Results for the efficiency based on the heat transfered to the water. Flow mL s
1.73 4.53 9.60 14.00 18.67 Mean
normal load Power η Jq [W ] [%] s 75.00 17.73 76.35 74.00 25.09 66.10 72.00 26.82 62.75 73.00 25.54 65.01 72.00 27.87 61.289 73.20 24.61 66.30
Power [W ] 110.00 107.00 110.00 110.00 105.00 108.40
full load q J s
34.02 42.35 45.36 44.98 45.63 42.47
η [%] 69.07 60.42 58.76 59.11 56.55 60.78
at normal and full load. This values are lower than those obtained with the block of copper, but this gets a little closer to the statement that an engine looses approximately 70% to 80% of the energy provided [8]. For comparison purposes, all the heat coefficients h and hA are calculated based on the results obtained with the block of copper.
3.3
Cooling Devices
For this part, different cooling devices were compared. All of them work thanks to the convection phenomenon and therefore equation (1.2) was used for the calculations. It is important to mention, that in all cases, the heat transfer was strictly done with the ambient air. For the fan coolers, the ∆T was measured between the ribs of the device and a temperature taken at the metallic case of the computer. For the heat exchanger and the radiators the two temperatures were measured at the nearest point possible to the processor and the opposite c extreme of the device. The fan implemented for the ThermalTake was a 12V fan opperated at 4V . Tables 3.5 and 3.6 show the obtained results. Table 3.5: Results for the different cooling devices at normal load. Cooler c Intel c Super-Silent c ThermalTake c
ThermalTake w/fan White radiator Black radiator
h W
hA W
m2 K
K
3.24 1.49 0.48 0.64 0.47 0.68
0.0029 0.0013 0.0004 0.0006 0.0004 0.0006
Like previously mentioned, a high value of h or hA indicates a better cooling c capacity. As shown in Table 3.5 surprisingly, the original Intel cooler is the best one. Nevertheless it is clearly visible, that the two fan coolers are by far c the best. One can also notice, that the ThermalTake works, as expected, c
better with a fan. The ThermalTake used should theoretically be placed with
F. Guerra, A.Michel
9
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
the processor in an upright position (which was not the case). Being placed horizontally, the convection effect within the ribs cannot take place as expected for an optimal performance. Using a little fan, helps the air flow through the ribs and provides the system with fresh air. As for the radiators, the heat transfer takes place through the whole surface. It is well known, that the white color absorbs more energy than the black. Therefore it is clearly visible, that the black radiator has a better cooling effect. Interestingly the cooling capacity of c the black radiator is more or less the same than the one from the ThermalTake with fan. Table 3.6: Results for the different cooling devices at full load. Cooler c Intel c Super-Silent c
ThermalTake w/fan Black radiator
h W
hA W
m2 K
K
6.41 4.62 1.70 2.22
0.0058 0.0042 0.0015 0.0020
Table 3.6 presents the results obtained for the different coolers at full CPU load. The ThermalTake without the implemented fan and the white radiator c were not analyzed. In the case of the ThermalTake , the ventilator was used to simulate the optimal upright position. The measurements taken without it were only to confirm the performance decreasing. As for the white radiator, advised by Mr. Wohlwend, we skipped those measurements. Apparently, the cooling effect is not strong enough, and reaching steady state would be rather difficult, due to the system crashing down just before getting stable. The results obtained at full load correspond to those obtained at normal load. Once again, c the Intel fan cooler is the best one, followed by the Super-Silent. This two are considerable superior than the others. At full load it is possible to see, against the expectations, that the black radiator works quite better than the c . ThermalTake
3.4
Comparison
To have a better overview of the results, all the cooling devices were compared side by side at normal and full load. The water cooling system is by far the best cooling method. Therefor only the results at the lowest flow were included in the cooler devices comparison chart. Figure 3.3 compares the “cooling capacity” of the different coolers analyzed and the water cooling system at a flow rate of 1.73 mL s , based on the convection coefficient. A higher value represents a higher “cooling capacity” and was obtained due to a larger ∆T . Like previously said, apart form the water, in c normal and full load, the Intel original fan cooler is the best one. Once again, the convection coefficient at full load is much bigger than at normal load. It is important to remember, that it is due to the larger temperature differences. While the processor gets much warmer at full load and the ambient air temperature remains more or less constant, the temperature gradient is bigger. A bigger gradient induces a greater heat transfer. This does not represents in any matter, F. Guerra, A.Michel
10
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
Figure 3.3: Comparison chart for the different cooler devices at normal and full load.
that the processor temperature is lower. It is worth to mention that unlike the other cooler devices, the convection coefficient for the water gets lower at full processor load. This is, as previously said, an exception that only appears at this flow rate.
4
Conclusions and Outlook
Cooling is not an easy task. Nevertheless it is a huge challenge for the future. In this experiment it was possible to study the heat transfer phenomenon by cooling a processor in different ways. Although different devices and a water cooling system were used, at the end, the medium that receives the heat is the air. Also for the water cooling system, the heat exchange was made between the water and the ambient air. This gives an idea of the limitations to improve cooling systems. Liquid nitrogen and acetone cooling systems are often mentioned, but when implemented for real-life applications, the expectations get lost. This techniques only work for short periods of time. As for future experimentations, c it would be interesting to study the ThermalTake placed at its recommended position, to see if its performance really improves considerably.
References [1] Answers Corporation. heat conduction. http://www.answers.com/topic/heat-conduction, 06/07.
December-January
[2] eFunda. Convection Theory in Heat Transfer. http://www.efunda.com/formulae/heat transfer/convection/overview conv.cfm, December-January 06/07. F. Guerra, A.Michel
11
ETH Z¨ urich
Chemieingenieurwesen I, WS06/07
[3] QUICK-OHM K¨ upper & Co. GmbH. QUICK-COOL. http://www.quick-cool.de/, December-January 06/07. [4] R. Nave. Stefan-Boltzmann Law. http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/stefan.html, December-January 06/07. [5] R.H. Perry and D.W. Green. Perry’s Chemical Engineers’ Handbook. McGraw-Hill, Inc., 7th edition, 1997. [6] Unknown. Fourier’s Law. http://wwwrses.anu.edu.au/∼uli/Teaching/Heat/FouriersLaw.html, December-January 06/07. [7] Wikipedia. Convection. http://en.wikipedia.org/wiki/Convection, December-January 06/07. [8] Wikipedia. Engine efficiency. http://en.wikipedia.org/wiki/Engine efficiency, December-January 06/07. [9] Wikipedia. Heat conduction. http://en.wikipedia.org/wiki/Heat conduction, December-January 06/07. [10] Wikipedia. Specific heat capacity. http://en.wikipedia.org/wiki/Specific heat capacity, 06/07.
December-January
[11] Wikipedia. Stefan-Boltzmann law. http://en.wikipedia.org/wiki/Stefan-Boltzmann law, 06/07.
December-January
[12] Wikipedia. W¨ armeleitung. http://de.wikipedia.org/wiki/W¨armeleitung, December-January 06/07. [13] M. Wohlwend and N. Osterwalder. K¨ uhlung eines Prozessors, WS 06/07.
F. Guerra, A.Michel
12
