"Cooling - Methods and Madness" Scott Morrison - 4/14/00
Every day, computer hobbyists strive to squeeze every possible drop of potential out of their systems. These enthusiasts will eventually turn to overclocking, the mad art of running their components beyond their rated speed. In this eternal quest, the overclocker's enemy is, as you all well know, heat. Every computer component uses electricity and producing no other work (kinetic movement, light, etc) all the energy consumed must be radiated as heat.
For years, systems engineers sought ways to remove this heat from the system in nearly every possible way, from the simplest to the most extreme. This is a brief article touching on the various methods of cooling your components. This may even be viewed as a history of cooling, seeing as each method builds on the benefits of the last. Every lesson learned here was first learned by the professionals, and then taken advantage of or learned anew by us, the hardcore hobbyists. The power-users and system abusers.
Convection and Radiation:
These are far and away the simplest methods of component cooling. No fans, no active components aiding the cooling whatsoever. Just vents. (And not even necessarily that.) Most enthusiasts and hobbyists these days would look at a system cooled by this method alone and say that it has no cooling whatsoever. Not technically true.
For instance, Nortel, maker of PBXs uses convective cooling in its low end 11C PBX. This system must dissipate about 500 watts of heat per cabinet, fully loaded, and the system must meet Nortel's standard for stability and reliability. The boards within the system are mounted vertically; the cabinets are ventilated top and bottom. The heating of the air around the components causes the air to rise like in a chimney, drawing the heat out of the system. The natural convective currents of air are very effective here.
Heatsinks naturally improve the convective or radiative heat transfer of an object by increasing the surface area, and possibly by managing the airflow to improve convection. Another important consideration may be the material of the enclosing case. If the case is made of plastic, or any thermally insulative material, it will contain the heat as the components heat the air within the case. A case of aluminum or any thermally conductive material will shed heat to the surrounding air as the interior heats.
Fans:
You should all be very familiar with fans for cooling. Cooling fans have been a part of PCs since the first. Rather than explain the history or operation of fans, or the thermodynamics of conduction, I'll instead point out a few facts on improving your cooling.
Turbulent air cools better. Say, for sake of argument, you have a simple tube with a fan in the middle. The fan pulls air from one side of the tube, and blows into the other. If you have a hot component on the exhaust side of the fan, it will be more efficiently cooled than on the intake side.
This is because the air on the exhaust side of the fan is more turbulent. For lack of a better explanation, the loops and whorls of turbulent air moving across the surface pick up more heat. The effective surface area of the object is increased. (Actually, it was explained to me by saying the effective surface area of the air is increased.) The total volume of airflow remains the same, but turbulent air just cools better.
The corollary of this: Use directed flow from intake fans to cool specific areas or your hottest components, use exhaust fans to extract the heated air from the case. Most people have already learned this lesson.
Go with the flow. Obviously, your cooling system will work better if your airflow doesn't have to fight the convection of the heated air in your system. Intake near the bottom, exhaust near the top. Try to maintain a uniformity of flow, a consistency of direction. Front to back, or back to front.
Don't let your fans fight. Sibling rivalry it ain't, but you still don't want your fans duking it out. Fans pulling against each other or blowing into each other leave dead zones where there may be a reasonable amount of turbulence, but there's very little actual airflow. This creates hotspots.
This includes cooling loops. I've seen people adding blowholes in pairs on their case side covers; one intake, one exhaust, each immediately next to the other. While this is assuredly very effective at cooling the area between the blowholes, it may create boundary conditions that actually impede the airflow through the case around that area.
Just like when supermarkets and department stores use those big vertical flow "air doors" to keep the cold side cold and the hot side hot. Try running both those blowholes as intakes, maybe even at half voltage. You may see your cards stay just as cool, and your overall case temperature drop.
Heatsinks:
I saved the heatsinks for after fans since most of us use them together, either as heatsink-fan combos, or with a fan directing flow over, around or through a heatsink. You all know that heatsinks do their jobs in two ways: they provide additional mass for the heat to propagate into, and they provide additional surface area by which to radiate or conduct heat.
Which is more important? It depends on your needs, but I don't think any of us are too concerned about moderating temperature change rates. We just want to keep our friggin silicon cool, right? So we want surface area, and the more the better.
Turbulence can help, but only to a point. In our heatsinks, turbulence still improves the effectiveness of the heat transfer. We want turbulence in the airflow to help the air carry away more heat. But we don't want so much turbulence that we reduce the total airflow through the 'sink.
HP for instance, had a cast and machined heatsink, which featured tall, thin vanes, in the usual long straight rows. However, the vanes weren't completely straight. Each one was twisted slightly out of line, into a little 15-degree zigzag formation. This increased the amount of turbulence in the airflow (as well as increasing the 'sink's surface area a small amount), without reducing the airflow.
