Originally Posted by
Plasmon
I feel uneasy about correcting a mod... especially a nice one, but what you've said about heat transfer isn't correct at all. I know this because I'm a materials engineer.
First of all there are 3 types of heat transfer, convection, conduction, and radiation. The transfer of heat from the chip to the heat sink is purely conduction, while the transfer from the heat sink outward is dominated by convection. What you have described about the heat transfer from the chip, through the thermal paste, and to the heat sink is not an accurate representation of the mechanisms of heat conduction.
Here's what happens. When two surfaces are pressed together, no matter how flat they appear to our eyes, they aren't anywhere near atomically flat, and likely have a roughness somewhere near 1 micron. Diamond paste that is often used to smooth materials to a mirror-like finish often has a grain size of 1 micron, so it can't polish to a degree finer than it's own grain size. This roughness in the surfaces of the two materials (the heat sink and the chip casing surface) means that the actual direct contact surface area is significantly smaller than the macro-scale area of the surface. This means there are many voids or air pocket. The problem is that an air gap is nowhere near as conductive as two surfaces in direct contact at the nano-scale. In heat transfer calculations we often model them like an electric circuit, using resistors to represent the resistance to heat transfer. At any surface we add a resistor to represent the "contact resistance" between the two materials, which takes into consideration the poor heat transfer ability due to the presence of the tiny air gaps.
Heat transfer by conduction strongly depends on the surface area, (it's directly proportional to surface area), so the presence of these low conductivity air gaps is detrimental to the heat transfer efficiency. The role of the thermal paste is only to fill the microscopic voids with a somewhat conductive material and the result is the reduction of thermal contact resistance. The paste material must be very good at wetting the two solid materials to ensure there are a minimal amount of gaps and so it has a low surface energy, low viscosity when under pressure or elevated temperature, and the higher thermal conductivity the better.
The ideal application of thermal paste keeps it as thin as possible while minimizing the amount of air gaps. You want to keep it thin because the conductivity is nowhere near as good as that of the two solid materials surrounding it, so it still causes thermal resistance itself. The spreading technique of choice depends on the viscoelastic properties of the thermal paste. If it is the type that spreads into a very thin layer upon the application of pressure and elevated temperature then the dot method would be superior. If it's not good at spreading on it's own, then manual spreading would be more beneficial. They spreading techniques are not equivalent and the method of choice has nothing to do with the nature of heat transfer.
What you've said about heat preferring to transfer through air isn't right at all. For example copper is approximately 16000 times better at heat conduction than air. (Yes that number is accurate.) Heat transfer by conduction occurs through phonons, which are quantum waves of atomic vibrations. The conductivity is therefore a function of the lattice structure at the atomic scale, and it is much more efficient in most solids compared to most liquids or gasses. Monocrystalline diamond is one of the materials with the highest thermal conductivity due to it's sp3 hybridized lattice structure. If diamond was super cheap and easy to shape it would be heavily used instead of copper or aluminum. Many people confuse thermal and electrical conductivity, but they are not the same mechanism. Metals are highly thermally conductive because the free electrons transfer heat as well as electricity but this correlation in conductivity between electricity and heat doesn't occur in most other materials. Diamond is terrible at conducting electricity without the presence of substitutional dopant atoms. The solids that are poor at heat transfer have microstructures or crystal structures that hinder the transfer of phonons. Plastics are generally bad because they have a low density and a mostly amorphous structure as well as poor inter-chain transfer mechanics. Glass is poor because it is amorphous. Some ceramic materials are good (single crystals of sapphire and diamond for example) and some are bad, but this strongly depends on the microstructure of the specific piece of material as well as the crystal structure. There are many ways that the structure can scatter phonons and render the energy transfer much less useful.
As a general rule for thermal conductivity you can assume this: solid>liquid>>gas. There are exceptions though.
Your experience with water cooling and tube size has nothing to do with heat transfer, it's about fluid mechanics. No matter what the diameter of your tube, the surface area between the waterblock and the chip casing is exactly the same, and the surface area between the water and the internal surface of the waterblock is also the same. Thicker tubing won't change much if your pump is using the same amount of power unless the tubing was so thin that it was causing an excessive pressure drop. Generally when you increase the diameter of a tube/pipe, the speed of the liquid decreases proportionally, so you have the same amount of liquid moving through the tube in both cases. Thin tube means faster moving liquid with a small profile, while a thick tube means slower moving liquid with a larger profile, and both have equal volume transferred per area per second.
/end science rant.