14
\$\begingroup\$

If I have a circuit with a computing chip drawing 1W, how much of that gets turned into heat?

EDIT: I mean turned into heat 1. straight away (not in 1000 years) and 2. within the vicinity of the device, typically in the same room

My understanding is that obviously the vast majority of the power is dissipated as heat, but I could not find a good source with some numbers. "99.99%" or "all of it, eventually" is often given but sadly without a proper source or irrefutable argument.

Context: I am investigating a domestic heater recycling heat from computing. Some company sells a 1000W device and they claim it is equivalent to any 1000W heater in terms of heating. But intuitively one might think that to produce 1000W of heat it must consume a lot more energy than a standard heater, since it's also doing computations...

\$\endgroup\$
11
  • 3
    \$\begingroup\$ Related: Does bitcoin miner heat as much as a heater (yes, input electrical power = output heat power). Where does electricity go after being used? (heat, and maybe some electromagnetic radiation like a radio transmitter or LED). Where does all the power consumed by a CPU go? (heat). Those are possible duplicates. \$\endgroup\$ Jun 2, 2023 at 8:12
  • 3
    \$\begingroup\$ "But intuitively one might think that to produce 1000W of heat it must consume a lot more energy than a standard heater, since it's also doing computations..." Nope. Computation, or the result of it, is not a form of energy. The energy that computers consume, is 100% converted to heat. \$\endgroup\$
    – Compizfox
    Jun 2, 2023 at 10:00
  • 2
    \$\begingroup\$ Energy conservation applies: Your heat dissipation is the input energy minus the energy that's is radiated or converted into something else. For a computer, that "other" energy is tiny: a few radio waves (WIFI), a bit of light (LEDs), maybe some sound from the speaker (microwatts). So yes, 99% of the energy gets converted into heat. \$\endgroup\$
    – Hilmar
    Jun 2, 2023 at 19:36
  • 2
    \$\begingroup\$ One more comment: IMHO @asac is overthinking this. You should just assume 100% of the electrical power (consumed) is converted to heat and move on to other problems. For general purpose computing equipment, this is accurate. \$\endgroup\$
    – Troutdog
    Jun 2, 2023 at 20:37
  • 1
    \$\begingroup\$ It might be worth mentioning that "reversible computing" is the (mostly theoretical at the moment) study of how a computer would need to be designed to avoid converting all of its input energy into heat: en.wikipedia.org/wiki/Reversible_computing \$\endgroup\$ Jun 4, 2023 at 5:31

8 Answers 8

31
\$\begingroup\$

If it drives no loads, all of it. 100 %.

If it drives loads, then subtract out the power to the loads.

The loads themselves may turn that power directly into heat as well, or into something that will eventually turn into heat.

since it's also doing computations

If you ignore quantum theory for a moment, then, from an energy analysis stand-point, all that doing computations does is generate heat.

\$\endgroup\$
2
  • 3
    \$\begingroup\$ I'd like to add and any signals directed into should also be considered as power. but then there's ULN2003 where power flows into the outputs and comes out of the power pin. \$\endgroup\$ Jun 2, 2023 at 4:05
  • \$\begingroup\$ perhaps minus a little bit that comes out as noise (eg: if the chip includes an SMPS it might make the caps vibrate enough to carry some energy away). \$\endgroup\$
    – Tom V
    Jun 4, 2023 at 20:19
13
\$\begingroup\$

Probably the only practical devices that draw meaningfully more energy than they release as heat are LEDs, and that is only if the light escapes the surroundings.

For example, a lamp with a high efficiency diode might convert only 60% of the energy it receives to heat with 40% becoming light. But unless that light is pointed out a window it still gets absorbed in the room and warms the room.

\$\endgroup\$
14
  • 2
    \$\begingroup\$ Even if that light were directed into deep space, eventually that light would hit stuff and do the remainder of the work. Law of Conservation of Energy -- if 1000W go in, then 1000W goes out... not 1001W, not 999W. \$\endgroup\$
    – rdtsc
    Jun 1, 2023 at 16:40
  • 3
    \$\begingroup\$ Many wireless circuits consume more than is dissipated as heat. Some of the power is converted into radio waves which propogate through the air. \$\endgroup\$
    – Troutdog
    Jun 1, 2023 at 17:23
  • 3
    \$\begingroup\$ @Troutdog I thought of mentioning that but I disagree that the will cause significant heat to escape the room. They may emit close to 1W peak (and maybe milliwatts average) but most of that is absorbed in the immediate surroundings. Amount escaping a house from a 10w wifi device is probably lower microwatts, maybe a few parts per million. \$\endgroup\$ Jun 1, 2023 at 17:38
  • 4
    \$\begingroup\$ You seem to forget electrical motors which only convert about 10% of their energy consumption into heat, much less than LEDs. \$\endgroup\$ Jun 2, 2023 at 6:24
  • 1
    \$\begingroup\$ What about radars? I think microwave generation is more efficient than light generation, and the radiation, too, can exit from earth. EMI energy from devices is very much a bad radar. So that energy is not turned into heat for a long time if it escapes the earth. \$\endgroup\$
    – tobalt
    Jun 2, 2023 at 18:01
9
\$\begingroup\$

I am investigating a domestic heater recycling heat from computing.

Your PC is already such a heater. There's no need to "recycle" that heat. It's already there. It comes out the exhaust fan(s) on the case.

Some company sells a 1000W device and they claim it is equivalent to any 1000W heater in terms of heating.

Links, please.

All those expensive CPUs and GPUs are doing is converting 100% of power they consume into heat, and incidentally computing something on the side. Those computations "lock in" such minuscule amounts of heat that we have no way of measuring it. There's no need to "add" any heaters.

A PC is a heater!

But intuitively one might think that to produce 1000W of heat it must consume a lot more energy than a standard heater, since it's also doing computations...

That intuition is unnatural :)

When you got a data center, every single watt of energy* pouring into the networking and computing equipment is coming out as heat via the air conditioning system. That aspect of thermal design of data centers is perhaps the most straightforward. If you can supply 1MW into the data center, you better be able to remove 1MW of heat, or else the inside will thermally run away (well, until the GPU and CPU clock governors start slowing things down to prevent a meltdown).

*In practice, there is a little bit of optical energy leaving the data center via the fiber optics. The heat lost that way is usually offset almost perfectly by the heat dumped into the data center from the power of all the light coming from fiber optic transmitters elsewhere. For any full-duplex link with fibers in good condition, the transmit powers on both ends of the link will be the same, and the optical power absorbed by the receivers is the same as well.

\$\endgroup\$
2
  • 2
    \$\begingroup\$ For longer-distance fibre links, the loss of light in the fibre (1-3dB per km according to Google) means light leaving the data centre is larger than light re-entering. But this is pretty miniscule. \$\endgroup\$ Jun 2, 2023 at 6:40
  • 1
    \$\begingroup\$ @SomeoneSomewhereSupportsMonica, it's more like 0.3 dB/km, or around that, depends a bit on the wavelength. \$\endgroup\$
    – ilkkachu
    Jun 2, 2023 at 7:13
6
\$\begingroup\$

Most power consumed by computer turns into heat. All computations and other chip and electronics activities being converted to heat. You may think this is 'waste', but all useful work that PC doing is possible only by consuming electrical energy and converting it to heat.

Exception1. Internal cooling fans work that some energy transform to air flow. Even that air flows if it is circulation inside the room also apparently converts to heat.

Exception2. Screen is emitting light. This light if its inside the room also mostly converts to heat.

Exception3. Some very small part of energy is conserved to charges and stored in Hard Drives/Capacitors/Other Energy storage devices. If Power turned odd OFF over undefined long time this stored energy also will be converted to heat.

\$\endgroup\$
5
  • 1
    \$\begingroup\$ Something to consider: does an energy storage device gain any mass when it stores energy? \$\endgroup\$
    – PStechPaul
    Jun 1, 2023 at 19:56
  • 1
    \$\begingroup\$ @PStechPaul No, because that energy is stored by rearranging chemical bonds. The number of atoms (and therefore the mass) remains the same. You could make some claim for a slight change of mass in electrostatic storage (capacitors), but considering the mass of an electron you'd have trouble measuring it! \$\endgroup\$
    – Graham
    Jun 2, 2023 at 8:59
  • 1
    \$\begingroup\$ @Kiper Also an exception for audio interfaces driving speakers or headphones. But ultimately that turns into heat too. \$\endgroup\$
    – Graham
    Jun 2, 2023 at 9:01
  • 1
    \$\begingroup\$ exception 3 actually does convert to heat fully. In order to write a nonvolatile memory, it dissipates the heat of the original information. \$\endgroup\$
    – tobalt
    Jun 2, 2023 at 18:04
  • \$\begingroup\$ @Graham The fact that the number of atoms remains the same doesn't mean the mass remains the same. Chemical reactions do in fact liberate (or consume) mass - but of course, only an absurdly tiny amount of mass you're not going to be able to practically measure (unlike with nuclear reactions where it's very easy to measure; but the principle is the same). The change in mass in an ideal accumulator would be exactly the same as the change in stored energy - it's just that our best accumulators still store very little energy compared to their "empty" mass. Energy loss comes out as heat again. \$\endgroup\$
    – Luaan
    Jun 4, 2023 at 15:45
2
\$\begingroup\$

Converting electricity to heat is an extremely easy process. All that is necessary is to provide a path for current to flow in such a manner that (1) the amount of current will not overload the source, and (2) the device's equilibrium temperature where it dissipates as much heat as its generates is low enough that the device doesn't self-destruct.

Suppose one has three boxes, one of which contains a 15 watt heater, one of which contains a 15 watt incandescent bulb, and one of which contains a 15 watt LED bulb, and feeds 15 watts of power to each bulb. In the first box, the heater will impart 15 joules of heat energy to the surrounding air each second. In the second box, the incandescent bulb might impart 7.5 watts of heat energy to the surrounding air, release 7 watts of infrared radiation, and 0.5 watt of visible radiation (light). The 7.5 watts of released radiation would then get converted to heat when it strikes the inside of the box. In the last box, the bulb might impart 5 watts of heat to the surrounding air and release 10 watts of visible radiation (light), which would then get converted to heat when it strikes the inside of the box.

In all cases, 15 watts of electrical energy would get converted to 15 watts of heat inside the box. The lights would first convert the energy into some other form that could serve some other useful purpose before it gets converted into heat, while the heater ignore the opportunity to do something useful with the energy in addition to producing heat.

The biggest caveat with using a "1000 watt" computer as a 1000-watt heater is that purpose-designed heaters are designed on the assumption that if they are below a certain temperature, they should convert 1000 watts electrical energy to heat without regard for whether everything useful they do besides produce heat (e.g. lighting up a temperature display) could be done with less than 1,000 watts. Computers, by contrast, are designed on the assumption that if they could serve all of their useful purposes besides producing heat while consuming less than the nameplate amount of energy, they should do so.

Additionally, any device which markets itself as a "1000 watt" heater would be expected to be able to dissipate 1000 watts of heat continuously without overheating. By contrast, a "1000 watt" computer would merely guarantee that it wouldn't consume more than 1,000 watts, and not necessarily guarantee that it would be able to continuously dissipate 1,000 watts without overheating. A properly designed computer should be designed so that if it starts to overheat, it will reduce its power consumption sufficiently to prevent damage, but such power reduction would limit the computer's usefulness as a heater.

\$\endgroup\$
2
\$\begingroup\$

One company sells a 1000W device and they claim it is equivalent to any 1000W heater in terms of heating. But intuitively one might think that to produce 1000W of heat it must consume a lot more energy than a standard heater, since it's also doing computations...

The fallacy here is thinking "computing power" as if it's a form of energy, in analogues to mechanical power or energy. In this line of reasoning, a computer would be a machine that converts the "electric power" from the mains into "computing power" and "heating power", and its heating efficiency must be lower than 100%.

But it's incorrect. While information usually requires energy to process or maintain, but at the end of the day, information in itself is a mathematical abstraction, it is not a physical form of energy or power. Ignoring electromagnetic radiation, a computer is a machine that converts 100% of electric power input into 100% of the waste heat, and this is just the price you pay for information processing, an inevitable side-effect. For the same amount of computation, you pay more or less depending on your computer's model and generation.

End of the story. There is "conservation of energy and mass", but there's no "conservation of energy and information" (when I was young, I made exactly the same mistake).


Well, there actually is to an extent. According to the idea of reversible computing, a computer generates waste heat only because the computation is not logically and physically reversible. In other words, the waste heat is produced by the information "lost" during computing (and some even argue there's a connection between thermodynamic entropy and information entropy). If one can build a logic circuit that somehow can preserve all the intermediate results, then there's a possibility to make the circuit physically reversible as well. After computation, the circuit can partially recycle the energy used. One possible implementation is adiabatic circuits, which incorporates inductors and capacitors as part of the logic circuit to reduce loss. They are compatible with the current CMOS semiconductor manufacturing process, but the energy recovery is limited.

Some science fiction stories imagine that the computer of the future can take a single "drop" of energy, reusing it repeatedly, allowing a much more powerful supercomputer. But at the current stage, it's still highly speculative.

\$\endgroup\$
4
  • \$\begingroup\$ In your second segment, I think you're talking about Landauer's Principle, which states that the energy dissipated by irreversibly flipping one bit of information is a small but positive amount. At room temperature, it's about 2.9 zJ (don't get to use that SI prefix very often!), which is swamped by orders of magnitude by other losses in a logic IC--current computers generate billions of times more waste heat from losses in driving FET gates than they generate from Landauer losses. \$\endgroup\$
    – Hearth
    Jun 4, 2023 at 23:57
  • \$\begingroup\$ That also means that your claim that "a computer generates waste heat only because the computation is not logically and physically reversible" (emphasis mine) is wrong; theoretically, an ideal computer would, but in that respect practical computers are about ten orders of magnitude off from being ideal. \$\endgroup\$
    – Hearth
    Jun 4, 2023 at 23:59
  • \$\begingroup\$ @Hearth Good point. I think a big source of confusion here is that the idea of "reversible computing" initially began from the Landauer's Principle, which was the original source of inspiration. But currently, this area of research also includes all logic circuits that make use of some form of low-loss switching or energy recovery techniques, unrelated to the Landauer's Principle. I should've distinguished both ideas clearly. \$\endgroup\$ Jun 5, 2023 at 0:13
  • \$\begingroup\$ Reversibility is also important in quantum computing. Non-reversible operations collapse the wavefunction, reverting back to traditional computing. \$\endgroup\$
    – Hearth
    Jun 5, 2023 at 0:46
2
\$\begingroup\$

There is no other exit.

There is no mysterious place the heat goes. There is no quantum heatsink that depends on faster-than-light signaling to wick away CPU heat and deposit in the core of the star Vega. (There's an episode of Star Trek The Orville: the civilization that attacks the Federation for smoking them out of their homeworld by thoughtless use of quantum heatsinks lol).

It is compulsory for all the wattage your computer uses to turn into same wattage of heat. The non-heat electrical energy sent down Ethernet cables or fiber-optic is so negligible as compared to AC mains input, that calling it "rounding error" overstates it.

Unless it's PoE, in which case yeah, that is passed through.

The Cadet baseboard mining heater

Many electric heaters come in baseboard form. About 12” tall, 4” deep and multiples of 2 feet long. They are 250 watts per foot. Standardized sizes exist, so they easily swap.

So, meet the Cadet baseboard miner. Instead of very boring resistive heaters, it's a stack of heatsinks stuffed with ASICs. Costs exactly the same as the old baseboard to run, and gives the same heat, but Bitcoins show up in your account :)

It really is like that.

PC wattage simply converts to heat at 3.41 BTUs per watt-hour.

Building a PC, you know you're already obliged to deal with heat removal. Well, where you send it is where it goes... simple as that. Example: Linus Tech Tips tried a few times to water cool PCs and ship their heat to radiators outside. Because otherwise with too many PCs in a room, you need supplemental A/C.

The economics of a Cadet baseboard miner would not work, however... it would only compute on a % duty cycle during heating season, not enough to pay for the cost of the ASICs. Also, normal thermostats do a hard power cut to the unit, and that would be result in abandoned work; so you'd hardwire the heater and use the W signal from a common "gas furnace" 'stat (Honeywell, Nest) to tell it to do an orderly shutdown.

\$\endgroup\$
0
\$\begingroup\$

A bit more than 100%. When the computer is off, its dynamic memory cells can take more states than when it is on when it just can occupy the states corresponding to 0 and 1. So when it is on, the states it can be in are reduced and the loss of entropy is given off as heat. When you switch it off again, it stops coercing states and reassumes full states.

Of course, this "state warming" at switching-off time is more than compensated for by the remaining operating heat dissipating.

Other than this paradoxic-seeming artifact of digital computing (which would not be the same for analog or quantum computers), every bit of energy put in will be given off as heat. The only exception are temporary energy storages like capacitors, rechargeable batteries, the momentum of rotating disks and similar items.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.