This article will attempt to present some ideas along with a general engineering analysis of a CPU (or other component such as a graphics chip or chipset) cooling system. The object is to lay out the physics behind such a system in terms that most should easily understand. There will be a series of articles dealing with basic cooling principles and physics, as well as specific design guidelines for successful water and air-cooling.
To give a little background on myself, my degree is a Master’s of Engineering with emphasis on structures and heat transfer. I worked during my graduate years designing and building cooling systems for satellites that use extreme cooling methods such as liquid nitrogen and colder methods.
My current occupation is as a process engineer for a large contract manufacturer in the printed circuit assembly business. We basically build the boards that other companies (i.e. HP, Dell, Sun, etc.) put their name on and sell to the public. It is my (along with many other engineers) duty to make sure that all the necessary processes are in place to be able to properly build the boards.
If my education, current job, and hobbies are all rolled into one, what to you get? An overclocking freak that has to try everything at least once!
There is a lot of general confusion about the different modes of heat transfer and how they relate to system cooling. To fully understand how cooling works, it is imperative to understand the basic modes of heat transfer.
Conduction is the transfer of energy (heat) from a more energetic particle to that of a less energetic particle by direct interaction. This is what happens when someone sticks his or her finger on a hot iron (or for that matter an overclocked AMD processor, which may actually be hotter than the average iron!)
The particles in the iron have more energy than the particles in the finger, and thus there is a net energy transfer from one to the other (ouch!).
Convection is the transfer of energy from a solid surface to that of a moving fluid. In order for heat to be transferred by convection, it must first conduct from the hotter material (either the fluid or the solid) by molecule to molecule interaction, and then the moving fluid displaces the molecules closest to the solid with other molecules as they move along.
Radiation occurs when there are two surfaces at different temperatures that emit electromagnetic waves between each other. This is basically how the sun warms the earth and everything on it. This mode of heat transfer is virtually negligible when talking about cooling your processor and motherboard; unless of course you are in a vacuum (space) or you computer is in a black case sitting out in the sun.
It is absolutely vital to understand the differences between the modes of heat transfer and when they occur in a computer system. I cannot stress this enough! There is an absolutely amazing amount of myth and misinformation circulating the internet overclocking world that could be easily put to rest if everyone could grasp a basic knowledge of the nature of heat transfer.
As power is applied to a CPU, it manifests itself in the form of heat and that heat (energy) must be moved away from the CPU or the CPU will eventually rise to an intolerable temperature and possibly burn out. To compound this problem, we as overclockers will often apply even more heat (energy, in the form of additional voltage) to the CPU in an effort to get maximum performance.
To remove the heat, we use either a heatsink/fan combo, waterblock, or a peltier combined with either of the other two aforementioned methods. Both the waterblock and the heatsink/fan systems act in basically the same way. A solid material conducts the heat away from the CPU and a fluid, either water or air, convects the heat away from the waterblock or heatsink and dumps it outside of the system.
In a peltier cooled system, the cold side of the thermal electric cooler cools the CPU by conduction but in order to do so uses additional energy. This energy must be removed from the hot side of the peltier in the same way as direct heatsink or waterblock cooling.
Remembering the definitions of conduction and convection, let’s follow the path of the heat and its transfer methods from the CPU to the outside of the case. I’ll only use the terms ‘heatsink’ and ‘air’ for this particular discussion, but ‘waterblock’ and ‘water’ could both be substituted instead; the physics are nearly identical.
As heat leaves the CPU, it is conducted directly into the heatsink. In order to get good conduction and thus effective heat transfer, the heatsink and the CPU must be in good contact with each other. Whenever there are multiple layers through which heat must conduct, there is what is called a contact thermal resistance. This is defined as such that where the two solids meet (i.e. the CPU slug and the heatsink) there will be a thermal contact resistance that results in an immediate temperature drop across the joint.
There is no such thing as a perfectly smooth surface and thus the two mating surfaces will not fit together flawlessly. The rougher the surfaces and the more imperfect the fit, the higher temperature drop across the interface. This is why it is important to use a quality thermal interface material between the CPU and the heatsink.
Often we overclockers even go to the extreme of lapping, or sanding down, either the CPU or the heatsink, or both, in order to get a more perfect fit. If there is a large gap or imperfect fit between the CPU and the heatsink, heat from the CPU will not be effectively transferred into the heatsink and the CPU will remain at a much higher temperature than the heatsink (bad!).
After the heat has been transferred into the heatsink, the material of the heatsink will then distribute the heat throughout itself. How well a heatsink does this is primarily a function of the thermal conductivity of the material.
Thermal conductivity is defined as the proportionality constant that when multiplied with the ratio of the temperature change to the change in distance from the zero plane, will result in the value of the energy flux (Fourier’s Law). This basically means that a higher thermal conductivity constant will result in a material moving heat along its geometry more effectively. For example, the handle of a wooden spoon stuck in boiling water will not get nearly as hot as the handle of a metallic spoon because metal has a much higher thermal conductivity than wood.
There are a few substances that are at the top of the thermal conductivity charts, namely:
- Diamond (2300 W/mK)
- Pyrolytic Graphite (1950 W/mK)
- Silver (429 W/mK),
- Pure Copper (401 W/mK), and
- Pure Aluminum (237 W/mK).
The first two are cost and geometrically prohibitive and are thus not candidates that will be looked at. Alloys of silver, aluminum and copper will always have a lower thermal conductivity than their pure counterparts, and often have thermal conductivities much lower. Thus it is very important to use pure metals! I.E. 6061 aluminum alloy will perform much worse than pure aluminum.
Aluminum is the most common heatsink material because of its cost, low density, availability, and machinability. Copper is beginning to become more popular but is much more difficult to work with and has a density approximately three and a half times that of aluminum. There has been talk recently of some silver heatsinks and/or waterblocks that may prove to be very interesting should they be able to be produced cost effectively.
Assuming that a system is constantly outputting heat, the material with the higher thermal conductivity will better move the heat away from the heat source. In the case of aluminum vs. copper, assuming identical geometries, copper will more effectively move the heat away from the point of contact with the CPU and into the extremities of the heat sink. This will give the heatsink a higher average temperature overall.
Aluminum will have a higher temperature difference between the point of contact and the heatsink extremities. This becomes and important factor because the heat must be removed from the heatsink along its entire surface through convection. Also recall that anytime there is an interface between two solids or materials, there is an associated drop in temperature across that interface. Thus if a heatsink is made of multiple pieces that are not properly bonded together, the heatsink will lose efficiency.
Now that the heat has been moved throughout the heatsink, it must be removed by forced convection. Air that is cooler than the heatsink is blown over the surface and the individual air molecules pick up energy from the heatsink and are (hopefully) ejected out of the case. There are only two factors that determine how much energy can be transferred from an individual air molecule that is in contact with the solid surface at a given temperature: time of contact and the lower of the two material’s thermal conductivities.
In this case, air has a much lower thermal conductivity than the metallic heatsink and is thus the limiting factor. HUGE BIGASS NOTE: There is no such physical phenomenon as to how well a material ‘gives up heat’. This is an internet-overclocking myth that has propagated for far too long and will now be laid to rest!
Aluminum does not ‘give up its heat’ better than copper! Let me repeat this once more; aluminum does NOT ‘give up its heat’ better than copper. It is true that, in general, aluminum will radiate heat better than copper but radiation is such a miniscule part of heat transfer in a computer system as to be deemed completely inapplicable.
The physical action of conduction/convection relies solely on the two material’s individual thermal conductivities, their proximity to each other, and their time in contact with each other. Thus, a pure copper heatsink will always outperform a heatsink of the exact same geometry of a pure aluminum heatsink assuming that both have the same contact with the heat source and the same rate of airflow over the surface.
So why don’t the current copper heatsinks far outperform (all tests I have seen show that copper heatsinks do outperform aluminum, just not by much) their aluminum counterparts? In my opinion it is because of a few things; namely poor design, multiple piece heatsinks (remember thermal contact resistance!), impure copper, and difficulty of producing a copper heatsink in the desired form.
Once the air has picked up the heat from the heatsink, it is simply ejected from the case and replaced with cooler ambient air. Thus our look at the journey of heat through a system comes to a close!
The next article will deal with specifics on laminar and turbulent flow and their importance to a cooling design. If there are any points that are unclear, or if there is anything else you would like to see in future articles, feel free to send me an email. Until then, happy overclocking!