How hot is too hot?
Publication date: 24 March 2010Originally published 2009 in Atomic: Maximum Power Computing
Last modified 03-Dec-2011.
For many years now, computer components have been able to tolerate startlingly high operating temperatures.
You won't be able to keep your fingers in 50°C water (a common output temperature for home hot-water systems) for much more than a second without being scalded. Water above about 88°C will scald you instantly. But it's possible for a PC with lousy ventilation to get so hot that the outside of the case is too hot to touch, and still work!
For a while.
The general rule they teach you in high-school science, that heat makes reactions go faster, applies to many components in PCs. But you generally don't want reactions to happen in your computer components at all. You want all of the elements to remain compounded with all of the other elements in their original configuration, and you want all of the metal bits to stay the shape they originally were.
Heat works against this, sweating plasticiser out of wire insulation, increasing the resistance of all sorts of components (which begets a little more heat), and encouraging metal atoms to "electromigrate" down the minuscule conductive traces in the chips. Electromigration in computer chips is actually measured by artificially heating the chip, so that the atoms will get blown down the wire fast enough for lifespan numbers to be calculated in a reasonable amount of time.
(At least quality motherboards today all have "solid state" polymer electrolytic capacitors for voltage smoothing. They work in basically the same way as normal electrolytics, but replace the wet electrolyte with a conductive plastic, which will never evaporate. They also give more capacitance per unit volume, which is why "solid state" caps are smaller than the wet electrolytics that preceded them.)
Pretty much the whole population of the world seems to, at some point, have posted on a forum saying "My CPU temperature is X degrees, is that too high?"
My standard answer to this is always "if your computer isn't hanging, then it's not too hot", and then I usually ramble on about how temperature numbers are unreliable in the first place, since the CPU's thermal diode is not part of the actual processor core, and (much more importantly) the number you get depends on how the analogue diode value is converted to digital, and what temperature your computer then assigns to that digital value.
It's also possible to make a computer too cold, with for instance a phase-change refrigerator cooling system, so dew condenses out of the air onto components. If the cooler keeps running when the PC itself is shut down, you can easily end up...
...with ice on components.
This may even make it possible for thermal cycling - expansion when components are powered up, contraction when they're powered down - to do damage, which is usually pretty much out of the question for computer hardware. (That doesn't stop people from saying that you should leave your computer on all the time so that components don't expand and contract and thus kill it. But I'm unaware of any actual evidence that this actually happens, to everyday computers.)
If you're a (relatively) normal person with a (relatively) normal air-cooled PC, it's generally a good idea to keep its components as close to the ambient air temperature as possible. This is especially the case if you want to run your CPU and/or graphics card faster than stock. Which is why PC enthusiasts often have computers that sound as if NASA's got a hypersonic aerodynamic test rig hiding in there somewhere.
Just keeping a howling gale of air moving through a computer case at all times is a perfectly good solution to PC heat problems, but the noise is likely to be obnoxious, and excess cooling can even waste a significant amount of electricity. A better idea is to focus your cooling on the areas that really need it.
The CPU and graphics card are the obvious candidates, but stock coolers for those components are pretty good these days, and there are a million and one upgrade options. Even motherboard core-logic chips (the old "northbridge" and "southbridge", and the single-chip modern versions) now usually have a hefty stock cooler, sometimes adorned with questionably-effective heat-pipes leading to a remote radiator.
What you won't find, though, are many ways to improve your hard-drive cooling.
There are two reasons for this.
One: Hard drives don't actually make a lot of heat.
The laws of thermodynamics tell us that all of the energy that goes into an electronic device ends up as heat somewhere. Since hard drives aren't the kind of device that, say, emits light, almost all of their input energy ends up as heat in their casings.
Fortunately, though, 3.5-inch hard drives generally only consume single-digit watts of power. Some current mainstream drives creep over 10 watts when they're working hard, but it's still nothing that obviously calls for cooling fins and mandatory forced-air.
(The reason why mainstream hard drives have never exceeded the 7200-RPM rotational speed, while all of the other specs have zoomed upwards, is that full-sized 10,000-RPM-and-faster drives produce enough extra heat that they do need forced-air cooling. A Western Digital VelociRaptor is a 2.5-inch drive bolted into a 3.5-inch heat-sink, and only spins at 10,000 RPM rather than the 15,000 of high-end SCSI drives, but it still won't be happy if you put it in an airless corner of your PC.)
Two: Hard-drive cooling is difficult to do if it isn't incorporated into the case design.
3.5-inch drive mounts in PC cases are usually a cheek-by-jowl block, with no room to install after-market fans or heat-sinks. When there are only one or two drives you can usually install them with empty bays on either side, to allow maximum access to whatever air happens to be flowing by. PC enthusiasts have a tendency to fill more and more bays as time goes by, though, which can leave the drives in the middle with no breathing space at all.
You can mount a 3.5-inch drive in a 5.25-inch bay by using extender rails for the sides or by attaching it to a purpose-built converter with built-in fans (often stupidly small ones with a miserable lifespan...). But if you want the more straightforward option of fans blowing into the 3.5-inch bays, or sucking out of them, or both, then you're either going to have to get a case that has mounts for such fans as stock, or rig some gimcrack arrangement involving cutting holes in a side panel. Or, Lazy PC Enthusiast Cooling Option Number One, just leave the side off entirely.
There are lots of cases that come with decent cooling for at least some of the drive bays. Lian Li have been making cases with a 3.5-inch cage right behind the front intake fans for ten years now. It's still very easy to find cases that don't have decent drive cooling, though.
There's an obvious reason why heat is bad for hard drives. It's actually completely wrong, but it's still pretty obvious, so here's an explanation of it, so people don't send me letters about it.
Hard drives have bearings on their spindle, see, and the bearings are ballraces made from metal, right, and metal expands when it's hotter.
When you heat a metal thing with a hole in it - like, in the classic physics-class puzzle, a washer - all of the molecules move apart, so all of the dimensions get bigger, including the diameter of the hole. (Otherwise old-time cartwrights and barrel-makers couldn't have gotten the iron tyres and hoops onto their products by heating 'em up.)
If the heat's being conducted outward from a shaft into a ballrace ("ball") bearing, though, the inner ring of the ballrace may expand more than the cooler outer ring, causing excess wear of the balls. This problem may also occur if the outside diameter of the bearing is restricted because it's mounted in some less-expandable material. But in the case of a hard drive you're generally looking at stainless-steel bearings installed in a cast-aluminium chassis, and aluminium has a coefficient of thermal expansion about twice that of steel. Even badly-cooled hard drives don't really get hot enough for any of these situations to be a major factor, but I had to mention them. Or else, letters.
This does sound kind of plausible, though, since hard drives spin fast and have very fine tolerances; who's to say that a teeny skerrick of extra wear on a spindle bearing couldn't mess the whole delicate balance up?
Except most, if not all, modern hard drives don't have ball bearings in them any more. They now have fluid bearings, cunning contraptions that use a layer of practically immortal lubricant, kept in place by the bearing's own rotation, instead of balls or rollers. The fluid bearings are quieter, smoother, more durable and even cheaper than the old ballraces were.
(The reason why hard drives didn't always have fluid bearings is that it took a while to come up with fluid bearings that'd work in a drive. They not only need to be very small, but also need to be able to operate for years on end without ever needing more fluid. Early fluid bearings were neither small nor sealed; actually, the fluid-film thrust bearing invented in the early 20th century was renowned for its largeness, on account of how it was used as the thrust block that applied the force from the propeller of a ship to the vessel's hull.)
There are plenty of other things in a drive that wear out faster when hot, though. More chips subject to slow chemical changes, of course, and the fantastically delicate head assemblies, whose conductor diameters are probably small enough for the abovementioned electromigration to be a significant factor.
And then there are the bearings on the head assembly's pivot shaft, a component that still has a lot in common with antiques like...
...this one.
Head-assembly bearings oscillate back and forth, instead of spinning constantly. So they can't be fluid bearings. It's barely conceivable that those good old ballraces don't like heat, and will develop a deadly little bit of slop. (An individual data track on a current hard drive can easily be only one six-thousandth of a millimetre in width!)
So, after all this, I'm not certain exactly why it's a bad idea for hard drives to run hot. It might be because of some kind of mechanical wear, but it's definitely not the obvious spindle bearings that're the problem. I think it's much more likely to be the solid-state components that'll die young if they're baking hot all the time. I wouldn't bet my life on any particular explanation, though.
And the hard-drive companies are resigned to having their products jammed into breezeless corners, and do their best to make drives that will soldier on for years, even if they're so toasty that if you suddenly removed one from the computer you'd have to juggle it like a hot potato. The spec-sheet operating temperature range usually tops out at 60ºC, but drives that shut themselves down and hang the computer at 61ºC probably wouldn't be very big sellers, so you can run 'em at egg-cooking temperatures if you want.
My own personal single data point is that the PC I wrote this column on has one of those cheek-by-jowl drive bays. I've only filled four of the six slots (...with a couple more drives adapted into 5.25-inch bays, naturally), but that still made the hard drives piping hot if they didn't get the chance to spin down for a while.
So I removed the side of the case, and hung a fan off one end of the drive cage, blowing across the gaps between the drives.
(Yes, the one on the bottom is sitting on top of the video card.
And yes, those are my drive letters, written on the drives.
It's stupid, but it works.)
Down between the drives, the casings still make it to 18 or so degrees above ambient, but they're a long way from "too hot to touch". And, for three straight years, they've been the least troublesome hard drives I've ever had. I still cloned and upgraded the original boot drive on its second birthday, but I bet I didn't actually need to.
A half-naked computer is not, of course, a very elegant way to keep drives cool. It's much better to get this stuff right in the original design.
So when you buy or build your next PC, make sure everything gets some air flow, not just the CPU and cards.
