|
LinuxHPC.org/Cluster Builder 1.3
Computer Cooling |
|||
|
By LinuxHPC.org and Cluster Resources
|
|||
|
Many components in a computer system unit produce large amount of heat during operation, including, but not limited to CPU, chipset, graphics card, and hard drives. This heat must be dissipated in order to keep these components within their safe operating temperatures, and both manufacturing methods and additional parts are used to keep the heat at a safe level. Overheated parts have a shorter life and may give sporadic problems resulting in system freezes or crashes. This is done mainly using heat sinks (to increase the surface area which dissipates heat), fans (to speed up the exchange of air heated by the computer parts for cooler ambient air) and in some cases (usually CPUs) using "softcoolers". Causes of heat build up The amount of heat generated by an integrated circuit (e.g. a CPU or GPU), the prime cause of heat build up in modern computers, is a function of the efficiency of its design, the technology used in its construction and the frequency and voltage at which it operates. In operation the temperature levels of a computer's parts will rise until the temperature gradient between the computer parts and their surroundings is such that the heat flow matches the input and the temperature of the computer part reaches equilibrium. For reliable operation, the equilibrium temperature must be sufficiently low for the structure of the computer's circuits to survive. Additionally, the normal operation of cooling methods can be hindered by other causes, such as:
Damage prevention It is common practice to include thermal sensors in the design of certain computer parts, e.g. CPUs and GPUs, along with internal logic that shuts down the computer if reasonable bounds are exceeded. It is unwise to rely on this, however, as it is not universally implemented, and even if implemented is intended as a damage limitation feature and may not prevent temperatures from reaching dangerous levels such that repeated incidents will cause premature failure of the integrated circuit. The design of an integrated circuit may also incorporate features to shut down parts of the circuit when it is idling, or to scale back the clock speed under low workloads or high temperatures, with the goal of reducing both power use and heat generation. System cooling Air cooling While any method used to move air around or to computer enclosures would count as air cooling, fans are by far the most commonly used implement for accomplishing that task. the term 'computer fan' usually refers to fans attached to computer enclosures, but may also be intended to signify any other computer fan, such as a CPU fan, GPU fan, a chipset fan, PSU fan, HDD fan, or PCI slot fans. Common fan sizes include 60, 80, 92 and 120 mm. Air cooling in high density computing Data centers typically contain many racks of flat 1U servers. Air is drawn in at the front of the rack and exhausted at the rear. Because datacenters typically contain such large amounts of computers and other power-consuming devices, they risk overheating of the various components if no additional measures are taken. Thus, extensive HVAC systems are used. Often a raised floor is used so the area under the floor may be used as a large plenum for cooled air and power cabling. Liquid submersion cooling An uncommon practice is to submerse the computer's components in a thermally conductive liquid. Personal computers that are cooled in this manner do not generally require any fans or pumps, and may be cooled exclusively by passive heat exchange between the computer's parts, the cooling fluid and the ambient air. Extreme density computers such as the Cray-2 may use additional radiators in order to facilitate heat exchange. The liquid used must have sufficiently low electrical conductivity in order for it not to interfere with the normal operation of the computer's components. If the liquid is somewhat electrically conductive, it may be necessary to insulate certain parts of components susceptible to electromagnetic interference, such as the CPU[1]. For these reasons, it is preferred that the liquid be dielectric. Liquids commonly used in this manner include various liquids invented and manufactured for this purpose by 3M, such as Fluorinert. Various oils, including but not limited to cooking, motor and silicone oils have all been successfully used for cooling personal computers. Evaporation can pose a problem, and the liquid may require either to be regularly refilled or sealed inside the computer's enclosure. Waste heat reduction Where full-power, full-featured modern computers are not required, some companies opt to use less powerful computers or computers with fewer features. E.g. in an office setting, the IT department may opt for a thin client or a diskless workstation thus cutting out the heat of components such as hard drives and optical disks. These devices are also often powered with direct current from an external power supply brick (which still wastes heat, but not inside the computer). The components used can greatly affect the power consumption and hence waste heat. A VIA EPIA motherboard with CPU typically radiates approximately 25 watts of heat whereas a Pentium 4 motherboard typically radiates around 140 watts. While the former has considerably less computing power, both types are adequate and responsive for tasks such as word processing and spreadsheets. Opting for a LCD monitor rather than a CRT can also reduce power consumption and excess room heat. Conductive and radiative cooling Some laptop components, such as hard drives and optical drives, are commonly cooled by having them make contact with the computer's frame, increasing the surface area which can radiate and otherwise exchange heat. Spot cooling In addition to system cooling, various individual components usually have their own cooling systems in place. Components which are individually cooled include, but are not limited to, the CPU, GPU, hard disk and the Northbridge chip. Passive heat sink cooling This involves attaching a block of machined metal to the part that needs cooling. An adhesive may be used, or more commonly for a home-user PC/laptop/Mac CPU, a clamp is used to affix the heat sink tight over the chip, with a thermally conductive pad or gel spread in-between. This block usually has fins and ridges to increase its surface area. The heat conductivity of metal is much better than that of air, and its ability to radiate heat is better than that of the component part it is protecting (usually an integrated circuit/CPU). Until recently, fan cooled aluminium heat sinks were the norm for desktop computers. Today many heat sinks feature copper baseplates or are entirely made of copper, and mount fans of considerable size and power. Heat sinks tend to get less effective with time due to the build up of dust between their metal fins, which reduces the efficiency with which the heat sink transfers heat to the ambient air. Dust build up is commonly countered with canned air, which are used to blow away the dust along with any other unwanted excess material. Passive heat sinks are commonly found on older CPUs, parts that do not get very hot (such as the chipset), and low-power computers. Active heat sink cooling This uses the same principle as a passive heat sink cooler, with the only difference being that a fan is directed to blow over or through the heat sink. This results in more air being blown through the heat sink, increasing the rate at which the heat sink can exchange heat with the ambient air. Active heat sinks are the primary method of cooling a modern day processor or graphics card. Peltier cooling or TEC This involves a Peltier effect heat pump which comes in direct contact with the component. The Peltier pump forces heat transfer at a higher rate than a normal heat sink. One side of the Peltier will cool down and the other side will heat up. The big disadvantage is that a Peltier is not an actual cooler, it can only transport heat from one point to the other. A processor with a peltier on top requires far more cooling than a processor without a peltier due to the inefficiencies of the Peltier Effect. Watercooling While originally limited to mainframe computers, computer watercooling has become a practice largely associated with overclocking in the form of either manufactured "kits" or in the form of DIY setups assembled from individually gathered parts. Lately watercooling has seen increasing use in pre-assembled desktop computers, most notably Apple's Power Mac G5. Heat pipe A heat pipe is a hollow tube containing a heat transfer liquid. As the liquid evaporates, it carries heat to the cool end, where it condenses to the hot end (under capillary force). Heat pipes thus have a much higher effective thermal conductivity than solid materials. In computers, the heat sink on the CPU is attached to a larger radiator heat sink. Both heat sinks are hollow as is the attachment between them, creating one large heat pipe that transfers heat from the CPU to the radiator, which is then cooled using some conventional method. This method is expensive and usually used when space is tight (as in small form-factor PC's), or absolute quiet is needed (such as in computers used in audio production studios during live recording). Phase-change cooling The simplest approach, and somewhat similar to a heat pipe, is to boil a fluid in a vessel (evaporator) attached to the hot CPU die. This vapor is condensed in the tubes of an air cooled heat-exchanger. The condensed vapor drains by gravity back to the boiling vessel. This is known as a "thermo-syphon". A key limitation is that the condenser must be positioned above the boiling vessel. This system is totally passive and requires no pumps or compressors. A more extreme way to cool the processor. A phase-change cooler is a unit which usually sits underneath the PC, with a tube leading to the processor. Inside the unit is a compressor, the same type that cools a freezer. The compressor compresses a gas which is cooled (usually with fans and air) condensing it to a liquid. Then, the liquid is pumped up to the processor, which heats it, causing the liquid to evaporate, and absorb the heat from the processor. This evaporation can produce temperatures reaching around −30 degrees Celsius. The gas flows down to the compressor and the cycle begins over again. This way, the processor can be cooled to temperatures ranging from −15 to −30 degrees Celsius, depending on the load, type and speed of the processor. Liquid nitrogen By welding an open pipe onto a heat sink, and insulating the pipe, it is possible to cool the processor either with liquid nitrogen, which has a temperature below −196° C, or dry ice. However, after the nitrogen evaporates, it has to be refilled. In the realm of personal computers, this method of cooling is only used for extreme overclocking trial runs and record-setting attempts. Softcooling Softcooling uses software to take advantage of CPU power saving technology to minimize energy use. This is done using halt instructions to turn off or put in standby state CPU subparts that aren't being used or by underclocking the CPU. Cooling and overclocking Extra cooling is usually required by people who run parts of their computer (such as the CPU and graphics card) at higher voltages or frequencies than manufacturer specifications call for, called overclocking. Increasing performance by this modification of settings results in a greater amount of heat generated. The installation of higher performance, non-stock cooling may also be considered modding. Many overclockers simply buy more efficient, and oftentimes, more expensive fan and heat sink combinations, while others resort to more exotic ways of computer cooling, such as liquid cooling, Peltier effect heatpumps, heat pipe or phase change cooling. There are also some related practices that have a positive impact in reducing system temperatures: Heat sink lapping All text used in this article is
available under the GNU Free Documentation License. It uses material from
the Wikipedia article "Computer cooling".This is the smoothing and polishing of the contact (bottom) part of a heat sink to increase its heat transfer efficiency. The desired result is a contact area which has a more even surface, as a less even contact surface creates a larger amount of insulating air between the heat sink and the computer part it is attached to. Polishing the surface using a combination of fine sandpaper and abrasive polishing liquids can produce a mirror-like shine, an indicator of a very smooth metal surface. However, it should be noted that even a curved surface can become extremely reflective, yet not particularly flat, as is the case with curved mirrors; thus heat sink quality is based on overall flatness, more than optical properties. Lapping a high quality heat sink can damage it, as although the heat sink may become shiny, it is likely that more material will removed from the edges, making the heat sink less effective overall. It is also worth noting that as the surface gets smoother, there is less and less ability for TIMs (such as Arctic Silver, or a standard thermal pad) to fill microscopic gaps from thermal expansion, thereby reducing over-all effectiveness of such materials and in some cases inhibiting thermal transfer. Use of exotic thermal conductive compounds Some overclockers use specialty thermal compounds whose manufacturers claim to have a much higher efficiency than stock thermal pads. Heat sinks clean of any grease or other thermal transfer compounds have a very thin layer of these products applied, and then are placed normally over the CPU. Many of these compounds have a high proportion of silver as their main ingredient. The resulting difference in the temperature of the CPU is measurable, and the heat transfer does appear to be much superior to stock compounds (several degrees celsius). However, some people experience negligible gains and have called to question the advantages of these exotic compounds, calling the style of application more important than the compound itself. Also note that there may be a 'setting period' and negligible gains may improve over time because the compund has reached its optimum thermal conductivity. Use of rounded cables Most PCs use flat ribbon cables to connect storage drives (IDE or SCSI). These large flat cables greatly impede airflow by causing drag and turbulence. Overclockers and modders often replace these with rounded cables, with the conductive wires bunched together tightly to reduce surface area. Theoretically, the parallel strands of conductors in a ribbon cable serve to reduce crosstalk (signal carrying conductors inducing signals in nearby conductors), but there is no empirical evidence of rounding cables reducing performance, this maybe because the length of the cable is short enough so that the effect of crosstalk is negligible. Problems usually arise when the cable is not magnetically protected and the length is considerable, a more frequent occurrence with older network cables. These computer cables can then be cable tied to the chassis or other cables to maximise airflow. This is less of a problem with new computers that use Serial ATA which has a much thinner cable. Airflow optimization Fewer fans strategically placed will improve the airflow internally within the PC and thus lower the overall case temperature in relation to ambient conditions. The use of larger fans also improves efficiency and lowers the amount of waste heat along with the amount of noise generated while in operation. For a rectangular PC (ATX) case, a fan in the front with a fan in the rear and one in the top has been found to be the optimum configuration. [citation needed] Fans in the side panels tend to add turbulence to the internal airflow precluding excess heat evacuating from the case. |
||
| © 2010 Adaptive Computing | |||