CPU Heatsinks, Fans, and Fan Noise Information

written by Robert Hancock

I put this page together to cover some issues relating to CPU heatsinks and fans, the level of cooling they produce, and the noise levels they generate. Hopefully this will help you become a bit more informed about CPU cooling in general.

Chip manufacturers, including those that make CPUs, are always trying to find ways of reducing the power consumption - and therefore heat production - of high-speed chips. Most often, this is done through using smaller sizes for the transistors and circuit pathways on the chips (this is the 0.18 micron, 0.13 micron, etc. that you often hear about). However, this doesn't completely compensate for the increasing clock speeds of CPUs, which means that the heat output of CPUs has been increasing. For example, the Pentium II and Pentium III-class CPUs tend to produce on the order of 20-30 watts of heat at maximum, whereas the Pentium 4 and AMD Athlon CPUs can produce as much as 60-70 watts of heat. Think of the heat produced by a 60 watt light bulb, concentrated in the area the size of a fingernail..

The heat the CPU produces has to be removed, otherwise it will heat up enough that it can't function properly, and will eventually melt down. (For example, an AMD Athlon CPU can burn out in as little as 1 second with no cooling!) Barring relatively exotic systems such as water-cooling, the way this is done is by strapping a heatsink made out of aluminum or copper onto the CPU, which absorbs the heat and dissipates it to the surrounding air. The heatsink has fins or pins molded into it which increase the surface area, and help to dissipate heat more effectively.

In order to get the maximum cooling, you would want the heatsink to be as big as possible, in order to get the largest surface area. There are, of course, some limits: obviously the cost increases, and it has to fit inside the case. However, there are also some others. For example, if the heatsink is too heavy, it runs the risk of being ripped off the CPU socket if the machine is jarred. There are some "hard-core overclocker type" heatsinks which, according to the manufacturer, are supposed to be removed before the system is moved because of this risk. Obviously this is unacceptable for a manufacturer like Dell, which has to ship their machines around. Major system manufacturers actually perform drop tests on their systems, to make sure the heatsink isn't likely to be ripped off through reasonable abuse. This means that the weight of the heatsink has to be limited.

I mentioned that heatsinks are usually made from either copper or aluminum. Traditionally aluminum has been used, but copper has become more common because it transfers heat more quickly. However, copper heatsinks cost more to make, and they're also heavier, which means it's easier to run into weight problems. Some heatsink manufacturers have taken the approach of making only part of the heatsink out of copper, to try to get some of the benefits without making the heatsink too heavy.

Of course, with current CPUs, a simple passive heatsink usually isn't enough, so you need some kind of fan to move air over the heatsink to dissipate heat more quickly. This is either done with a fan on the heatsink itself, or a separate fan which pulls air over the heatsink through a duct of some kind. Many Dell machines take the second approach, as this way the main chassis fan can also serve to cool the CPU. However, this setup really has to be custom-designed for a given motherboard and system case, so it's mostly used by the largest system manufacturers.

Generally, the more airflow the fan can produce, the better cooling you will get. The two basic ways of increasing the airflow of a fan are to increase the speed (RPM), and increase the size of the fan so that the blades move more air. Once again, there are some drawbacks for each:

• Both will increase the power draw of the fan. For example, some fans, like the infamous Delta "black label" which is known to many overclocker types, and which runs at over 7000 RPM, pull enough current that they can burn out the circuitry on the motherboard for the fan connector, so they should be connected directly to the power supply using a drive power connector. With more "sane" fans, this shouldn't be a concern.
• Increasing the fan RPM increases the noise level. It also makes the fan tend to produce a higher pitch sound, which may be more annoying.
• Increasing the fan size doesn't increase noise as much (or, to put it another way, you can get the same airflow at a lower RPM). However, the fan size should likely match the heatsink size - if it's too big, not all of the air will actually reach the heatsink unless you use some kind of duct, which will restrict airflow in itself.

The bearings used in the fan also have some effect. Sleeve bearings tend to produce less noise than ball-bearing fans, because there is no physical contact between the rotating parts (the fan shaft is suspended on a film of oil). However, as the oil in the sleeve bearing evaporates or leaks out, physical contact starts to occur, and at this point the fan usually starts making horrible grinding or vibrating noises, and eventually stops entirely. Ball bearings usually last considerably longer, so they are probably worth the slightly increased noise. (Note that there are two bearings in a typical fan. If a fan just says "ball bearing", it may have one ball and one sleeve bearing. If it has two ball bearings, it will probably specifically say "double ball bearing".)

As an experiment, I took 3 CPU fans and recorded the noise that they generate. Two of them, the Evercool heatsink currently shipping on Powerleap's Slot 1 CPU upgrades, and the Powerleap heatsink used when the PL-iP3/T first came out, were tested using a 12-volt power supply while not installed in the machine. Another, the Intel retail-boxed heatsink from the Celeron 1.2 GHz, was recorded while installed in the machine - because I was using it :-) - so these recordings have some more background noise.

I also used some audio-analysis software to determine the primary pitch (in hertz) of the noise they generate. The results, as well as some links to audio files, are shown in the table below.

Note that the audio files here have all been normalized to the same volume level - this is because I made little attempt to be scientific about microphone placement to allow accurate volume comparisons, therefore any differences that existed would have likely been misleading. Therefore, you should only judge the type of noise from these files, and not the loudness.

None of these fans are really all that noisy, but myself, I find the Intel heatsink's noise to be by far the least objectionable. As the pitch numbers show, it makes quite a bit lower-pitched noise, which blends in more easily with the other noise from the machine (hard drives, other fans, etc.) while the Powerleap and Evercool fans produce a higher-pitch noise which stands out above all the other sounds.

It's surely no coincidence that the Intel heatsink also dwarfs the other two in terms of size - in fact, it barely fits between the capacitors on the Powerleap adapter. This means that a slightly larger fan can be used, which helps somewhat, but the greater heat dissipation area also means the air velocity required is lower, so a less powerful fan is needed.

The Evercool heatsink, in particular, was apparently designed originally for 1U rack-mount servers, which can only be about 1 1/2 inches tall. This restricts the height of the heatsink quite severely, which probably explains why it uses a relatively high-RPM fan - so that it can move heat off the heatsink more quickly.

In conclusion, what everybody probably wants is a cooler that cools as efficiently as possible, is as quiet as possible, and isn't excessively heavy. However, these goals are all somewhat contradictory, so when you pick a cooler, you have to decide which of them are the most important for your needs.

Return to Main Processor Upgrade FAQ Page
Return to Robert Hancock's Home Page