[Dan Humphreys]

Dan Humphreys asked the Naked Scientists:
   
I recently heard a podcast in which you talk about how much electricity is wasted by internet providers and servers in the form of heat energy.  

You have talked about Stirling engines on other podcasts, and I was wondering if anyone has thought to use this very old technology to turn some of that wasted heat back into electrical energy?  

My understanding is that Stirling engines can convert heat into mechanical energy, which by way of a generator can be turned
back into electricity?  

Thanks

What do you think?

[techmind]

I think the answer is that in principle yes, but in practice it's not worth it.

All heat engines work by extracting energy from a flow of heat from a hot source to a cold sink.

The theoretical maximum efficiency of a heat engine increases as the temperature difference between the hot source and the cold sink increases. If you have several hundred degrees centrigrade temperature difference, then you can get a reasonable efficiency. If on the other hand you only have 30 degrees difference then the efficiency (even the theoretical maximum) is pretty poor. You'd probably want to keep all the servers in the data centre below 55-60C under all conditions. Given an ambient temperature of 20-30C, that's not a great temperature drop.


Since you can't get energy for nothing, by definition, any heat pump necessarily impedes the flow of heat in order to extract energy from it. (By analogy, any water- or wind-turbine impedes the flow of water/wind in order to extract energy.) If the objective is to keep something cool, this means you need a bigger heatsink to compensate for the loss of thermal conductivity of the heat-engine.

In the real world with data centres we usually can't get the heat out fast enough using natural conduction or convection so have to use "air conditioning" which is an active heat pump. That is, you put electrical power in to a machine to force the heat to be moved from A to B faster than it would by natural conduction. If you attempted to extract any energy further down the line, you'd simply have to increase the power of the air-con to compensate, and you can guarantee you'd loose out overall.


So... sorry, but it's not really going to work in practice.
Far better to make the software more efficient (costs more development time) so it needs fewer and less power-hungry computers to run it. And reduce the amount of data stored and processed.

[lyner]

techmind
I am not making a serious proposal but the actual surface temperature of semiconductor devices can be 100C and, using heat pipes, you could operate with a temperature difference of 80C. The air from a conventionally cooled system is not the best way of using the waste heat.

Ref your comment on heat shifting problems. I think there is undue emphasis on high density packing of electronic equipment. Is there any serious reason for a server, which serves hundreds and hundreds of msquared of office space, should be located in one spot? I can see the attraction of siting all the IT facilities  in one place but that chosen place is frequently (and understandably) in one of the worst places in a building, deep inside or under it. The total consumption of a rack is only a few kW (could it be tens of KW?), which is not a big proportion of the total power input.  If the 'server' were distributed, rather than central, the power sinking problems could be a lot less.
What we need is joined up design, based on total power sinking and sourcing requirements overall. I am sure that all computer rooms are equipped with central heating radiators!

[techmind]

Indeed, it did occur to me to add an afterthought that if you actually "plumbed in" the hot chips using heat-pipes and/or flowing-fluid then extracting the heat would be a lot easier.

Get rid of the air-interface which has a huge thermal resistance, and you make getting rid of the heat much easier (and just maybe have a chance of reclaiming some energy). At the very least you'd be able to cut down or reduce your air-con (energy) costs.


The objective is usually to keep the silicon die temperature below about 120C, so an integrated-circuit ("chip") package exterior at 100C can be tolerable - but you need the rest of the innards of the PC at below 60C otherwise the electrolytic capacitors will fail very prematurely and mechanical things like the hard disks will probably complain too.


I think in some instances you need the computers in reasonably close proximity in order to have high bandwidth data connections between the boxes... but then with fibre-optics that might not be such an issue. I suspect it comes from "space is money" - you wouldn't space them out in a room just for the sake of it. For security you can't spread them out under desks in a large office. It's also helpful to have them in a dedicated room where you don't get too much dust in the works. Again also if you need uninterruptible power supplies and the like it's much easier if all the system/server components are in one place rather than spread throughout a building. You don't want the cleaner unplugging the server to plug in a vacuum cleaner!

[Bored chemist]

I have a feeling that most computer rooms  (in the bad old days when such things existed) were air conditioned in order to pull heat out of the equipment and push it into the ambient air.
I guess server farms have the same problem- certainly, on a local scale, cooling the processor in a computer usually needs a fan and a heat sink.

What you are proposing is to add something else in the way of that heat  as it makes its way to the outside world.
I don't think you could do that- in effect all the power you gained from the stirling engines would need to be fed to the air conditioning to stop the chips overheating.
I think the only way round this would be to have either a really huge system  of heatsinks or site the server farms near rivers and use the water to take away the waste heat.

[lyner]

techmind
If the chips could be kept at a sensible temperature using heat piping, I think the electrolytics (with very low internal loss) would not get too hot. In a joined up design, they could all be on the cool side of a board and cooled by a 'gentle' fanning, rather than the present roar. Local decoupling capacitors, right by the chips, need not be high value - the 0.1muF, low inductance type, I mean.
As for the location of the servers; that should be a matter of choosing the optimum, rather than assuming that the space / cost trade off is correct at the moment.

[Schlurch]

Thanks for the comments.  I'm not sure that anyone has really addressed the initial question, however.  I'm not a great technical mind, but it seems as though all the answers have dealt with how to keep a server cool, and keep the chipsets from overheating.  I was referencing a problem that was pointed out in one of the podcasts, in which the heat energy that comes off of said datacenter is merely wasted.  Now, I understand that heat energy generally represents a loss of efficiency in such a system, as with an incandescent lightbulb versus a fluorescent one, though I see that such heat-loss is near inevitable with such data systems.  My question was whether a stirling engine-based system could be designed not to remove or eliminate the heat, but to attempt to profit by it in some way, so that it is not all simly lost to the environment.

Dan

[techmind]

I'm not sure that anyone has really addressed the initial question, however. ... My question was whether a stirling engine-based system could be designed not to remove or eliminate the heat, but to attempt to profit by it in some way, so that it is not all simly lost to the environment.

We have attempted to address this. The two problems are:
 1) the temperature difference between the heat we are faced with and the ambient temperature is insufficient to be of much use in any kind of heat-engine (Stirling engine)
 2) any attempts to extract energy from the heat will by definition impede the heat-flow, which in many cases is already being actively 'pumped' buy means of air-con. Extracting energy from the heat will therefore require more powerful aircon and will require more extra energy than you're extracting.

The most useful thing you might do with the heat is use it to heat nearby buildings/office-space (if it's not summer!)