There are two other types, radial/centrifugal (blows air outwards) and cross-flow; those will be covered later (if ever…
).Fans
- At the simplest level, the choice of fan is a compromise between air movement, noise output, and physical size. Improving two of these requires sacrificing the other. You can get a fan that is average in all ways, you can get a small and quiet but weak fan, you can get a small and powerful but loud fan, or you can get a large, quiet, powerful fan.
- The dominant noise emission from most fans is a prominent tone caused by the impeller interacting with the air. The frequency of this tone is equal to the rotations per second (= RPM/60) multiplied by the number of blades of the impeller. (So with seven blades at 2700RPM, typical of many 80mm fans, the resultant tone is at 315Hz - roughly D#4 on the musical scale.) Obstructing the airflow in "front" of a fan (the side it sucks air from) will exacerbate this noise. Running two similar fans at nearly the same speed can result in an irritating fading in/out type effect - as the relative rotation angle of the two fan rotors varies, their noise output cycles between cancelling out and adding together. To avoid this, do not run two identical fans at the same speed unless there is no alternative (besides using a single fast fan of the same size - anything beats that - reason in the next bullet point).
- In the same fan, rotation speed and airflow are pretty much linearly correlated. Rotation speed and noise, however, are nowhere near it.
This makes high-speed fans undesirable - just one speed grade higher can make a big difference (those seemingly small steps are there for a reason).
But fast fans do have a benefit - they can move air at higher pressure (correlated to the square of rotation speed). - While bigger fans have better noise/airflow ratios under the same operating conditions, compromises are sometimes made to fit them into a small space - ATX PSUs being the classic example. The form factor was designed for an 80mm rear-mounted fan, and using a 120mm top-mounted fan (well, most cases mount the PSU upside-down in relation to its internal construction so it therefore faces down) requires so many sacrifices that it ultimately results in much worse performance. Some old Socket A coolers funneled an 80mm fan down onto a 60mm heatsink, which caused enough back-pressure that the same heatsink would be better-cooled by a 60mm fan with the same noise level (if you must use an oversize fan, just let it overhang the heatsink).
- The fan frame should have a pair of arrows marked on it. One indicates the rotation direction and the other indicates the airflow direction.
Counter-rotating fans (although rarely encountered in PCs, other than servers) have 3 arrows (or 4 if the stages have individual frames). - Metal (or plastic where RF shielding isn't required) grilles are used to prevent accidental contact with the fan blades.
However, they reduce airflow and can increase noise dramatically if mounted to the front of a fan (as noted in the second bullet). Some of them are very badly designed, with more metal than holes (or decorative types may have such large holes that they don't provide meaningful protection). - Fans used in PCs are (usually) powered from 12VDC (there are also 5V, 24V and 48V; the latter two can't be used in PCs) and can have two, three, or four wires. 2-wire fans just have power connections ("negative" usually being ground). 3-wire fans (as used in PCs*) have the power connections and a speed output (switches between open-circuit and grounded in sync with the rotation, or actually twice per rotation as most fan motors have 4 poles). 4-wire fans have the aforementioned speed sensor, plus the facility to control the fan speed using pulse-width modulation. Many older motherboards attempted to use PWM on 3-wire fans but this corrupts the tachometer signal, rendering it useless until the fan is returned to full speed. These are typically colour-coded as:
Ground (pin 1): usually black, occasionally blue
+VDC (pin 2): red (common) or yellow (Intel stock coolers)
Sensor (pin 3): yellow or white (common); blue (Delta); green (Intel stock coolers)
PWM (pin 4): blue (Intel stock coolers); yellow (Delta); brown (NMB); others?
The pin numbers referenced are the wiring order of the small 3-pin and 4-pin (PWM) connectors used to connect fans to motherboards.
4-pin headers are backwards compatible with 3-wire fans, but then using PWM control will do nothing; 4-wire fans should default to maximum speed if plugged into a 3-pin header, but this is not always guaranteed. 2-wire fans are generally fitted for direct connection to the PSU outputs.
One notable exception: stock coolers for the PPGA Celerons, which while only 2-wire, had the same connector as for 3-wire fans.
Fans can usually also be speed-controlled by varying the input voltage, which can make them run slower (or faster, if you dare) than design.
However, too low and the fan won't even start spinning, and too high will destroy the fan motor (or its control IC). Not all manufacturers recommend voltage reduction, but within the specified range you should be OK; however some high-end fans have a regulated speed and can't be voltage-controlled. - There are several types of bearings used to support the rotating impeller. The two most common are sleeve bearings, which consist of a steel shaft loosely turning inside a sintered bronze cylinder (filled with oil to reduce friction), and ball bearings, which work on the principle on "things roll easier than they slide", featuring multiple steel (or occasionally ceramic, either silicon nitride or zirconium dioxide) spheres rolling between inner and outer steel "races". A "cage", which may be made of steel or plastic, maintains even spacing of the balls; and grease is packed inside to further reduce friction.
Generally speaking, sleeve bearings are quieter (with adequate lubrication) but do not perform well with the shaft vertical (i.e. with the fan mounted horizontally) and the lubricant will dry up over time, which is accelerated by high heat. Ball bearings can operate in any orientation, and generally last longer, but are more easily damaged by shock (due to the minuscule contact points). Ball bearings have to be used in pairs for stability. Some fans have one ball bearing and a sleeve bearing, but I don't know what use that is, except to sell a fan on the basis of "ball bearings = more reliable" and then penny-pinch without technically lying. Preloading is required for ball bearings to run smoothly (if you disassemble a 2BB fan, you'll find a spring in addition to the bearings themselves), which can't be done with 1B+S as it would put pressure on the retainer, creating excessive friction and noise. Fortunately, if you're in doubt, there's a way to tell without having to look up the model number.
Simply hold the fan frame in one hand, and try to slide the impeller in and out. If you can slide it axially with very little effort, there's a sleeve bearing inside. If it springs back to its original position, it's dual ball bearing.
Many manufacturers have their own bearing system, which is often a modification of the sleeve bearing e.g. ARX's CeraDyna (a sleeve bearing with a ceramic center shaft (CeraDyna A) or entirely ceramic (CeraDyna C)), Delta's Superflo, ebm-papst's Sintec, Noctua's SSO, and Panaflo's (now part of NMB-MAT) Hydro Wave.
Small axial fans usually have square frames, available in the following sizes (listed are typical thicknesses though there are exceptions):
- 25mm (20mm between holes), 10mm thick
- 30mm (24mm between holes), 10mm thick
- 35mm (29mm between holes), 10mm thick
- 40mm (32mm between holes), 10~20mm thick
- 50mm (40mm between holes), 10~20mm thick
- 60mm (50mm between holes), 10~25mm thick
- 70mm (61.5mm between holes), 15~25mm thick
- 80mm (71.5mm between holes), 15~25mm thick
- 92mm (82.5mm between holes), 25mm thick (also 15mm thick in some SFX PSUs)
- 120mm (105mm between holes), 25 or 38mm thick (also 15mm thick in SFX-L PSUs; Phanteks PH‑F120T30 is 30mm thick)
- 140mm (125mm between holes), 25 or 38mm thick
I generally wouldn't recommend 38mm thick fans as they are usually designed to be powerful, not quiet. (NMB 4715VL has a 6400RPM model. And yes, there was a 12V version, so if you were a masochist (and had 36W to spare), you could use it in your PC when it was available.)
Many very small fans have relatively massive motor hubs, leaving little room for the blades and hampering noise/airflow performance - so it may be worth comparing several fans of the same size to see which performs best. That aside, however, physically similar fans usually have similar noise and airflow characteristics. In some cases, the choice between two somewhat different fans of the same size may come down to airflow vs. pressure.
Some manufacturers make fans with transparent plastic rather than the usual black; but the amorphous plastic (polycarbonate, I think) used in transparent fans is prone to resonance. Opaque fans are generally glass-fiber reinforced polybutylene terephthalate (PBT) and/or nylon, but premium models (e.g. Noctua, Phanteks) may use liquid-crystal polymer (LCP; it doesn't have much relation to liquid-crystal displays, and the finished plastic is very much solid).
Moving air also produces broadband noise due to general turbulence. With a fan operating in free air that noise will be drowned out by the blade-induced buzz/whine, but in restrictive environments (such as poorly ventilated PSUs) it can start to dominate.
Other key cooling parts
For basic information on heatsinks and thermal compounds, I refer you to this site. The fan article there, on the other hand, I considered inadequate. A critical thing worth mentioning is that, like PSU reviews, the majority of heatsink reviews are done wrong. It is often said that the grease/paste (I prefer "paste" over "grease" because, as necessary, it's sticky enough that it doesn't flow by itself) performs better than the solid pads. That advice is obsolete. While the historical graphite and filled-mesh pads were indeed effective thermal resistors, modern phase-change pads (the only type used for CPUs since ≈2001) melt at a defined temperature (which is chosen to be above realistic room temperatures, but within the safe operating range of the component), squeezing the excess material out of the way under the pressure of the heatsink mount, just as happens with paste. The thermal conductivity of the compound itself is not much different between the two. This test should tell (not my favourite reviewer by any means but at least the thermal testing was done with proper methodology - which is more than can be said for Hardware Secrets). (I do think Dan's attitude toward longevity was rather cavalier, though. Not upgrading until I have to (until mid-2014, I was still using a system I assembled from mostly circa-2005 hardware), I insist on material, component, and design choices to last over a decade under normal conditions. The thermal conductivity of the compound isn't the big deal to me.)
Water cooling
A system that has its uses but is usually way over-rated. Instead of mounting the heatsink directly to the heat source, it uses a circulating liquid (which can be anything that has suitable physical properties, but water is the most common given its near-zero cost and exceptional heat capacity) to transfer heat from the component to a heat exchanger (usually referred to incorrectly as a "radiator"). Compared to the sheer simplicity of a direct-mounted heatsink, a water cooling system has these components, all connected by tubing (usually made of silicone rubber or soft plastic):
- Block(s) that pick up heat from the source - there can be several of these in one system as long as the heat exchanger can handle the combined dissipation of all the sources, the pump can provide enough pressure to circulate the water at a sufficient rate, and they are sequenced (the order in which they receive water flow) to avoid heat-sensitive components (e.g. HDDs) being presented with the heated water from high-dissipation components (CPUs and GPUs).
- The heat exchanger, which does the same job in the same way as a normal heatsink, only with the heated water from the blocks instead of contacting the source directly. They are constructed with the metal internal pipes looped through the fins (to which they are bonded for thermal connection) many times, maximizing contact with them.
- A reservoir which holds reserve water - if you like, a "buffer" of it - to keep air bubbles out of circulation.
- The pump. In principle this is actually like fans used for air movement, only at lower speed and higher torque, and it has to be watertight. High quality pumps have the stator coils outside the water chamber, coupled to the rotor magnets through the plastic.
- The size of the heat exchanger is no longer restricted by the available space around the heat source, as it can be mounted remotely.
- A single system can cool multiple separate heat sources (given the considerations listed in the above section).
- Mechanical stresses on component mountings are greatly reduced compared to using large direct-mounted heatsinks.
- The pump used is another motorized device to produce noise and fail (and its failure is just as bad as a fan failure, if not worse).
- Low quality pumps often use DC brush motors. These must be avoided at all costs.
- The water loop adds extra thermal resistance to the system.
- The heat exchanger must be more powerful than the conventional heatsink that would otherwise be used or there is no performance gain.
- The reservoir must be correctly oriented to work properly.
- Sealing is absolutely critical as if the water leaks...well, if it's reasonably pure (plain water isn't actually a great conductor, though the more salt you add, the more conductive it gets, so deliberately using salt water is not advisable), doesn't get into the PSU (which, for that matter, I wouldn't advise water-cooling) and you're there when the leak happens to pull the power plug, your hardware might survive. But regardless of the circumstances, there's no guarantee.
External HDD enclosures
These deserve special mention. Way too many pre-built external drive units (including the official HDD-manufacturer-brand ones, sadly) have fanless plastic cases - which excel at trapping heat, leading to early failure of the drives inside, and must therefore be avoided at all costs. HDDs are usually rated for an absolute maximum of 55~60°C, and the cooler, the better. After all, you can't just swap a failed HDD out and go like you can with other components - even if you have a backup, you need to restore it to the replacement drive, and you still lose anything created since the last backup. Fanless aluminium cases (so long as they don't have anything that insulates the drive from the case) are not great, but a huge improvement. Using fans is the best option - if they are high quality.
