47

A heatsink can be stuck on your CPU to cool it down. That heatsink feels cold when the system is not turned on. However when the CPU is turned on the heatsink is extremely hot. Isn't that contradictory to a certain extent? I expect a heatsink to always be cold and to have the air blowing system turned on for convection purposes.

Could someone explain what is wrong in the way I see things?

LandonZeKepitelOfGreytBritn
  • 631
  • 1
  • 5
  • 9
  • 63
    The heatsink is cold, relatively speaking... – jiggunjer Aug 22 '16 at 13:07
  • 14
    Touch the back plane of your fridge, its temperature will most probably surprise you. – Dmitry Grigoryev Aug 22 '16 at 14:34
  • 33
    where do you think the heat in the heatsink come from? – njzk2 Aug 22 '16 at 17:17
  • 3
    you're trying to cool down the CPU. why do you care if the heat sink is hot? is the CPU cooler than it used to be? – user428517 Aug 22 '16 at 23:06
  • 5
    The maximum rated temperature a CPU can operate at without errors (T_Jmax) is typically around 100 degrees C for modern CPUs, and they can get up to 90 C before they even start to thermally throttle. There's plenty of room for a heat-sink to feel uncomfortably hot while still being cooler than a CPU under load. If even the idle temps are very high, that's less normal. Software can read the CPU's built-in temp sensors, e.g. using the lm-sensors package on Linux. – Peter Cordes Aug 23 '16 at 07:18
  • 13
    @sgroves: He care's because he's interested in the physics behind it. If everybody wouldn't care about things, we'd still be trying to survive against bears etc. – phresnel Aug 23 '16 at 09:38
  • 1
    @phresnel I think that was an explanation attempt; to rephrase it: Sure, the heatsink is hot, but it's colder than the CPU would be without the heatsink, right? – user253751 Aug 25 '16 at 00:19
  • @immibis that is correct. CPUs are not completely efficient and a lot of the energy put into them is expended as heat, if you left only the top pad of the CPU exposed to the air you only have that tiny square for it to transfer its heat to the surrounding air, which would cause it to overheat and die. Instead, we stick a heatsink to it with as best conductivity (usually via thermal paste) so that the massive surface area of the heatsinks fins can move the heat into a larger volume of air (as it touches more). We then blow that hot air out with the fan so it is replaced with fresh, cool air. – James T Aug 25 '16 at 08:27
  • It's hot, but it isn't the SOURCE of the heat. Believe me, without that heatsink, if you touched the CPU (presuming it hasn't already burned itself out), you'd end up melting the skin on your finger. – Logan Aug 25 '16 at 13:17

9 Answers9

91

"I expect a heatsink to always be cold"

The purpose of the heat sink is to transfer heat. The rate of heat transfer depends in a complicated way on (a) the temperature difference between heat sink and air, (b) the exposed surface area, and (c) the air speed.

The heat sink is at its most effective if it is at the same temperature as the CPU. This is because this gives the maximum temperature difference between heat sink and air. This is why heat sinks are often made of copper: copper conducts heat well.

Let us suppose that the CPU is producing heat $Q$ (Watts) and that the heat sink transfer heat at a rate linear with the temperature difference:

$$ Q = K_\text{eff} (T_\text{CPU} - T_\text{air})$$

where $K_\text{eff}$ is the effective heat transfer coefficient. Solving for the CPU temperature:

$$ T_\text{CPU} = T_\text{air} + Q / K_\text{eff}$$

So, the CPU temperature is lowest when $K_\text{eff}$ is highest and $K_\text{eff}$ is highest when the heat sink, due it to having high thermal conductivity, is at nearly the same temperature as the CPU.

John1024
  • 3,025
  • 17
  • 14
  • So the purpose of metal of the heatsink itself is not to cool down the CPU but rather to be able to transfer it. While the purpose of the built-in fan is to cool down the CPU. Did I understand correctly what you meant? – LandonZeKepitelOfGreytBritn Aug 21 '16 at 22:39
  • @trilolil Yes, that's it. Without the air flow, there would be no place for the heat to go. – John1024 Aug 21 '16 at 22:54
  • There are some heatsinks that incorporate thermoelectric 'heat pumps', which can get the chip colder than the heatsink fins. – Whit3rd Aug 21 '16 at 23:17
  • 2
    "The heat sink is at its most effective if it is at the same temperature as the CPU" Nope. That is patently wrong. The job of the heatsink is NOT to cool itself. Its job is to cool the CPU. It can only cool the CPU when there is a temperature differential between it and the CPU. There will always be a temperature difference. However, the better the interface with the CPU, the smaller the "REQUIRED" temperature differential to maintain the heat flow. – Aron Aug 22 '16 at 04:03
  • 63
    @trilolil The purpose of the heatsink is to touch lots of air, because when a hot thing touches a cold thing the cold thing gets hotter and the hot thing gets colder. The purpose of the fan is to blow the hot air out of the way after the heatsink heats it up, and replace it with cold air, because if the heatsink was only touching hot air it wouldn't be able to cool down very well. – user253751 Aug 22 '16 at 04:09
  • 8
    @Aron The usual model for an ideal heat sink is that, while convective heat transfer is finite, the heat sink itself has infinite conductivity. This is part of the concept of "fin efficiency," as discussed here or here for example. If the heat sink has infinite conductivity, then it is at the same temperature as the CPU. Any reduction in the thermal conductivity reduces the overall heat transfer and makes the heat sink less effective. – John1024 Aug 22 '16 at 04:23
  • @John1024 That is the usual model. As with all models it's very useful for understanding key parts of a system. However, the OP wants to know how a hot heatsink can cool a CPU. The answer is your model is wrong, and the heatsink is a tiny fraction of a degree cooler than the CPU. – Aron Aug 22 '16 at 04:27
  • 7
    @trilolil You can contrast it with what happens if you have a CPU without a heatsink - the CPU doesn't touch very much air because it's small, so it doesn't heat up the air very much, so it doesn't cool down very much. Also because it's smaller it heats up and cools down quickly, you don't want that because you want to have time to turn it off before it melts. – user253751 Aug 22 '16 at 05:56
  • 1
    @trilolil all cooling is about transferring the heat somewhere else. Heatsink takes the heat from CPU and transfers it to surrounding air. A fan blows away hot air and takes in new, hopefully cooler air from outside (heating it slightly in the process, because the fan motor is also a heat source). A refrigerator pumps heat from its inside to outside, again, producing a net heat in the process - there is no process that can reduce heat in a closed system, it can be only transferred to somewhere else. – Peteris Aug 22 '16 at 07:32
  • 7
    @trilolil Note that for very old CPUs (and nowadays, things like the Northbridge), the heatsink was enough, no fan needed. There's a certain amount of heat the heatsink can get rid of on its own, simply through convection (air gets hot from contact with the heatsink, rises and carries the heat away, while "sucking in" cold air). Fans are not necessary (and neither is strictly speaking the heatsink), they just help with the cooling. You can still build 100% fanless computer today, it just requires much bigger heatsinks :) – Luaan Aug 22 '16 at 08:11
  • 2
    (Additionally) note that humans do not sense temperature, but rather the flow direction of thermal energy. If the heatsink feels hot, it implies that it is capable of transferring energy towards you, which good heat conductors, connected to a heat source, will do even regardless of their own temperature. – Holger Aug 22 '16 at 10:03
  • 7
    @trilolil "the purpose of metal of the heatsink itself is not to cool down the CPU but rather to be able to transfer it [...] the purpose of the built-in fan is to cool down the CPU" The heatsink cools the CPU by transferring heat out of it. The air cools the heatsink by transferring heat out of the heatsink. The fan blows the the hot air away and replaces it with cold air. The new batch of cold air transfers more heat out of the heatsink. As long as heat is leaving the heatsink, more heat can flow out of the CPU into the heatsink. Heat leaves the heatsink fastest when it's hottest. – anaximander Aug 22 '16 at 10:53
  • @immibis Most CPUs from the last decade+ have thermal protect features that will prevent self destruction if they lose cooling. Intel had a basic panic off switch implementation as far back as the p3, with the p4 able to throttle itself down massively to continue running (at a much lower clock speed) without a heatsink. AMD added similar protection a few years later (IIRC with socket 754) after a series of infamous test videos by Tom's Hardware showing Intel P3/4 systems surviving heatsink removal while Athlon XP ones burned up and died. – Dan Is Fiddling By Firelight Aug 22 '16 at 14:54
  • @DanNeely I wouldn't rely on those protecting the CPU when it's running without a heatsink. If your fan stops spinning / gets clogged with dust then sure, because the greater thermal mass of the heatsink makes it heat up more slowly. – user253751 Aug 22 '16 at 23:18
  • "The purpose of the heat sink is to transfer heat" maybe I should open a new question but this just made me think, heat is only ever transferred so what really is the need of a heat sink in addition to a fan (since a fan transfers heat too)? – Celeritas Aug 23 '16 at 03:21
  • 1
    @Celeritas Heat sinks have fins so that heat is more rapidly transferred to the air. In other words, the shape and materials of the heat sink are chosen to maximize $K_\text{eff}$. – John1024 Aug 23 '16 at 05:51
  • 2
    @DanNeely: Just a note on those melting Athlon tests. They were a bit deceptive because AMD did have thermal protection as part of the motherboard BIOS. But the poll rate was low enough that with the CPU under load and completely removing the heatsink there wasn't enough time for the BIOS to power-off before destruction. – Zan Lynx Aug 23 '16 at 23:45
  • 1
    @Luaan, there's still plenty of CPUs that don't need fans or even heat sinks. You just won't find them in rack-mount servers and high-performance workstations. Look for them in cell phones. Look for them in your car. Look for them in your washing machine and your microwave oven and your blu-ray player, and in all kinds of other electronic devices. – Solomon Slow Aug 24 '16 at 17:37
  • @Luaan +1 to this. I built a NAS using an AMD A8-7600 processor with a fanless heatsink. It's just a 2-pound copper and aluminum behemoth (with heat pipes) :) – Doktor J Aug 24 '16 at 20:43
41

In general, a heatsink should feel hot, if it's doing its job right. If it feels hot, that means it's transferring energy to your hand, which means it's transferring energy away from the CPU.

This also holds for other cooling things. The cooling coils on a fridge or an old A/C should feel hot for the same reason: the heat you're taking away has to be dumped somewhere. If the cooling coils themselves felt cold, your fridge would be working in reverse; it'd be heating up your food.

knzhou
  • 101,976
  • 3
    "Conversely if it felt cool, it'd be heating up the CPU" - no. – John Dvorak Aug 22 '16 at 22:48
  • 1
    @JanDvorak Actually, yes. The only way to make it cool in the first place is to dump heat from the heatsink to the CPU, by an active mechanism. – knzhou Aug 22 '16 at 22:50
  • You're suggesting a heat pump from the air to the CPU. But since we have such powerful heat pumps handy, why not have them pump heat to air far away from the CPU? This would then cool down the CPU as well as any organic stuff on top of it. This suggestion of mine should even be less power hungry than your one. – John Dvorak Aug 22 '16 at 22:59
  • @JanDvorak Sure, but if you allow heat to be directly pumped far away, then we're talking about a different question. In this case a heatsink is completely irrelevant. – knzhou Aug 22 '16 at 23:05
  • 3
    The point is, if you want to dissipate heat by radiation, which is what a heatsink does, you want the heatsink to be as hot as possible, not cool. – knzhou Aug 22 '16 at 23:06
  • 6
    If it felt cool it might be heating up the air and cooling both your hand and the CPU. – user253751 Aug 22 '16 at 23:19
  • "Conversely if it felt cool, it'd be heating up the CPU" is quite misleading, since it is supposing that the heat transferred from the environment to the (cold) heat sink would somehow end up being transferred to the (extremely hot) CPU. This is impossible without some very powerful heat pump, which would be both pointless and technically impossible. If there were active refrigeration of the heat sink, it would certainly dump its heat elsewhere then at the CPU, and the result would be that the CPU is more effectively cooled. But a CPU need not be kept that cool, so it is not done this way. – Marc van Leeuwen Aug 24 '16 at 06:52
  • @MarcvanLeeuwen Yes, I was thinking about a heat pump from the heatsink to the CPU, which is of course extremely unrealistic. The reason was because I was also thinking about the next example (refrigerators), where heat pumps are necessary. And in that case, the conclusion is that if something feels cool, the other thing must be getting heated up. Since several people have complained now, I rephrased to move that sentence where it belongs. – knzhou Aug 24 '16 at 06:57
  • A heatsink can run warm or hot by design. Its purpose is to facilitate transfer of heat so that the thing to which it is attached does not get hotter than it is specified for. A CPU is normally specified up to 80C or higher. If you are engineering a fan-less system using convective cooling, the amount of convection and thence heat transfer depends as a high power law on the temperature difference between the heatsink and the surrounding air. It's unsurprising if it heats up to 60C in a room at 20C, or 70C in a hot room. Inside the CPU specification. No problem. – nigel222 Aug 24 '16 at 11:01
  • By the way, there are published statistics (Backblaze? Google? I forget) which showed that for hard disk drives, ones in a strong airflow that kept them cooled below 30C were less reliable than ones that were able to heat up into the 30C-50C range. (30C-40C was best). They're engineered to run warm, not cool, so coolest is not best for them. Temperature was that returned by the disk's SMART diagnostic. – nigel222 Aug 24 '16 at 11:13
  • @knzhou It would feel cool if it never reached 37C, since then it would be absorbing heat from your hand if you touched it. But it'd have to be very efficiently disposing the heat from both the CPU and your hand elsewhere. – Random832 Aug 24 '16 at 13:50
  • @knzhou: I think saying "hot as possible" without some key additional phrases is a bit disingenuous. If my fridge's cooling coils was a radiator the size of a shipping container, they wouldn't at all feel hot, but that wouldn't mean the fridge is working in reverse. A heatsink "running as hot as possible" could reasonably be interpreted as what most would consider an under-spec'd heatsink, right on the razors edge of the processors working condition, both which are less than ideal. – whatsisname Aug 24 '16 at 20:57
  • Computers (outside of the power supply) are generally safe enough, but in general touching heatsinks with the power on can be.. shocking. – Spehro Pefhany Aug 26 '16 at 10:28
  • @knzhou The cooling is done mostly via convection than via radiation. If you put a raspberry pi in a vacuum, it will power off quickly due to overheating. – untreated_paramediensis_karnik Feb 24 '18 at 14:37
16

The goal of the heat sink is to improve heat flux from CPU to surrounding air. Heat flow is a function of the area of contact and the temperature difference. By adding a component that increases the area, you increase one factor; but it's worth remembering that the heat sink will have to have a temperature between that of the CPU and the air: if it's at the same temperature as the CPU, no heat flows from the CPU to the heat sink; it it's at the same temperature as the air, no heat flows to the air.

In practice, it's easy to make the thermal resistance between the heat sink and the CPU packaging quite low (note - the chip is typically quite a bit hotter than the package, and it's only the latter that touches the heat sink). So with the conductivity between CPU and heat sink being better than conductivity from heat sink to air, the heat sink will tend to a temperature close to that of the CPU.

Floris
  • 118,905
  • 3
    I like this answer because it seems to match the "level" of the question whereas the other answers break down into formulas which are greek to lay people like me. – Dean MacGregor Aug 23 '16 at 15:00
10

When a heat sink is in intimate contact with a source of heat, like a CPU, heat is transferred in accordance with Newton's Cooling Law:

$$\frac{\mathrm dQ}{\mathrm dt}=uA(T_\textrm{CPU}-T_\textrm{Sink})$$

Where $u$ is a heat transfer coefficient (CPU to sink) and $A$ is the area of contact between the CPU and the heat sink.

Note that $\frac{\mathrm dQ}{\mathrm dt}$ is the heat carried off from the CPU per unit of time.

Higher values of $T_\textrm{Sink}$, as the basic formula shows, actually decrease $\frac{\mathrm dQ}{\mathrm dt}$, which becomes effectively zero when $T_\textrm{CPU}-T_\textrm{Sink}=0$.

To prevent this from happening, the heat sink itself has to transfer accumulated heat, usually to the surrounding air, in which case another heat transfer equation comes into play:

$$\frac{\mathrm dQ}{\mathrm dt}=hA_\textrm{Sink}(T_\textrm{Sink}-T_\textrm{air})$$

Where $h$ is the convection heat transfer coefficient (sink to air) and $A_\textrm{Sink}$ the sink's surface area exposed to the air. $h$ is very dependent on speed or airflow which explains why forced air circulation (fan assisted ventilation) is often used.

In steady state ($T_\textrm {CPU}\approx \text{Constant}$), we have, with $\dot{Q}_\textrm{CPU}$ power generated by the CPU:

$$\dot{Q}_\textrm{CPU}=(T_\textrm{CPU}-T_\textrm{air})\left[\frac{1}{uA}+\frac{1}{hA_\textrm{sink}}\right]=(T_{CPU}-T_\textrm{air})\frac{1}{K}$$

Or:

$$T_\textrm{CPU}=T_\textrm{air}+K\dot{Q}_\textrm{CPU}$$

With:

$$K=\frac{uhAA_\textrm{Sink}}{hA_\textrm{Sink}+uA}$$

The influence of the various factors on $T_\textrm{CPU}$ can be readily appreciated.

Ingenious ways of increasing both $h$ (apart from forced circulation) and $A_\textrm{sink}$ to lower $T_\textrm{CPU}$ have been used like cooling fins or this design (ST-HT4 CPU Cooler Riser):

ST-HT4 CPU Cooler Riser

The highly heat conductive copper bands mostly release the heat from the U-bends.

Gert
  • 35,289
  • 12
    Those 'copper bands' aren't solid copper; they're almost certainly heat pipes, which are significantly better heat conductors. That device won't improve cooling performance; it simply allows you to shift the cooler further from the CPU without a massive reduction in performance. – SomeoneSomewhereSupportsMonica Aug 22 '16 at 08:29
  • @SomeoneSomewhere is right. these are common in laptops allowing the fan to not sit directly on top of the CPU. – Chris H Aug 22 '16 at 10:20
  • 4
    @DanielSank a search confirms the 4 copper tubes are heat pipes. The Wikipedia heat pipe page lists a heat pipe at roughly 200 times more thermally conductive that copper. – Philip Tinney Aug 22 '16 at 20:07
  • Also note they took the effort to have the heat pipes directly touch the CPU, instead of the much cheaper and more common solution of embedding them in a block of copper which touches the CPU. – Chuu Aug 23 '16 at 20:05
  • Anyone who understands this answer wouldn't have needed to ask the question. – user184411 Aug 26 '16 at 16:48
7

Heat flows can be modeled like electric currents. Temperature differences correspond to voltage differences, and the heat flowing is similar to an electric current. In both cases, the two are approximately linear, and we can define the coefficient as a resistance value.

Now, if the resistance is zero, we'd see that there will be no temperature or voltage difference; if infinite there won't be any flow. Both are of course ideal cases.

So, back to the CPU cooler. The goal here is to move a lot of heat from the CPU to the air. This requires a low resistance - high flow, low temperature differences. But the resistance here has two parts: moving the heat from CPU to cooler, and from cooler to the air. These are of course resistances in series, so they add up. (just like electricity). To minimize the sum, we need to minimize both resistances. And the temperature in between (top of cooler) will be dictated by the ratio of the two resistances.

An engineering observation is that moving heat from a solid to the air isn't very effective. Air has a rather low heat capacity per volume, as it's so thin. Adding a fan helps with the volume of air, but it's going to remain the bigger resistance of the two.

SO, just as two electric resistors form a voltage divider, two thermal resistances divide the temperature. The top of the cooler will be somewhere between the CPU temperature and the ambient air, and engineering tells us it's going to be closer to the former.

MSalters
  • 5,574
7

I'm going to answer this in a far more general way: You don't make cold, You move heat.

Think about an air conditioner: The bits outside get very, very hot. What it's doing is expending mechanical work to move heat from from the inside to the outside. The heat hasn't vanished and cold hasn't been created, the thermal energy has just been moved around.

Unlike an air-conditioner, a normal CPU heatsink is a completely passive system. It can't directly move the heat but it does provide it a highly-conductive place to go, thus pulling at least some heat out of the CPU. Because of the higher area (and more importantly thermal mass) each unit of heat energy is spread thinner over a larger area. The heatsink gets hot, but without it the small thermal-mass of the CPU would be far, far hotter.

Do note that there are some special computer cases with active cooling. This would satisfy your expectation of a cold CPU Heatsink. The mechanical work from the cooling system creates more waste heat than a simple heatsink with fans, but that heat is being dumped from the system so it doesn't affect CPU temperature.

UIDAlexD
  • 539
3

Already good physics answers. Here comes an everyday metaphor which doesn't require as much physics or mathematics knowledge (I hope).

A heat sink should drain the heat of the thing it is attached to. It should be the points or areas which the heat flows towards. That is what a sink is, a point where something flows into. For example a kitchen sink - that is where all the water flows towards and gets sucked up.

Then of course it will probably be even better at doing this if it is also at the same time able to transfer away heat quickly away from itself - the kitchen sink is more effective if it is not clogged up but able to keep the water it sucks up moving so it flows somewhere else.

mathreadler
  • 233
1

everything is relative. while it may feel "hot" when compared to your fingers, i assure you the heat sink is still colder than the cpu itself. moreover, the heat sink's job is simply to move heat from the cpu into the air. as such, the temperature of the heat sink is not important as long as the heat flowing thru it is sufficiently high. the heat sink is an improvement over a cpu-to-air interface because heat flows more quickly both from the cpu to the heat sink and from the heat sink to the air. conduction from the cpu to the heat sink is faster because the heat sink is a better conductor of heat than air. conduction (and radiation) from the heat sink to the air is faster because the heat sink has a larger surface area.

teldon james turner
  • 353
0

The key here is conductivity, which is simply speaking the speed at which heat moves to/from the object. Objects feel cold or hot not based on merely their own temperature, but also their ability to radiate or conduct that temperature to you. In essence, something only feels hot if it succeeds in warming up the part of the body you're sensing with. The same with cold, it has to result in your body cooling down and actually being cold.

Plastics are notorious poor conductors and respond slowly to heat changes, whereas metals are often good conductors, feeling cold to the touch when cool (they conduct the heat away from your fingers efficiently) and resulting in burns when hot (fast movement of their contained heat into your fingers).

Why are good heat sinks conductive? Well, a heatsink wants to do 3 things: firstly absorb the heat away from the CPU, then move that heat away from the CPU, then finally radiate that heat into the general environment, and it needs to do these things at a rate that reaches equilibrium with the rate heat generation of the CPU at an acceptable effective temperature. In all these cases, conductivity is conducive to these goals.

Danikov
  • 101