Well he *is* discussing the use of nanotubes, a technology still on the drawing boards today, to dissipate heat [the whole book is an exploration into the future of nanotechnology], and is proposing a cpu a million times more powerful than a brain. Drexler is a futurist and is not constraining himself here too much if you read the book. Elsewhere he derives heat output and this has to do more with the theoretical thermal limits of electric computing rather than any benchmarked system of the day. It's a good read BTW.

To give some perspective on a coffee-cup size cpu a million time more powerful than a brain consider the following quote from researchers studying the evolution of information storage and computation: As of 2007...

"...To put our findings in perspective, the 6.4*1018 instructions per second that human kind can carry out on its general-purpose computers in 2007 are in the same ballpark area as the maximum number of nerve impulses executed by one human brain per second”, "Our total storage capacity is the same as an adult human’s DNA. And there are several billion humans on the planet."

Article from 'Wired': http://www.wired.com/wiredscience/2011/02/wor … -computer-data/

So regardless whether were talking about a 286 or modern cpu the numbers we are dealing with are astronomical looking into the future - and heat dissipation can not be ignored..

The industry is meeting the challenge, if not for environmental concern then from pressure of mobile computing. Computer entertainment would be easy target, but computers in general? Remember world before, with hordes of accountants, phone operators, robust machinery, real mail and so on, not to mention wholly new opportunities opened by computing power. Even in this point of view, personal computer is a tool that helps more then harms. I would question more modern big data, for example super computer data mining for more precise advertisement- that is worse then waste.

so that means that we lovers of older computing cpu's are the biggest of polluters in a sense.
i was thinking me would be eco green with an energy star using a older pc which use lesser power than my friends gaming station with several gpus.
so we should all use modern machines or older machines again? 😉

Global warming is a lot of unproven bunk.

RacoonRider wrote:

All PCs are 0% efficient. All the power they draw comes to heat, as far as I know.

Well, a small fraction comes out as light and sound energy, too, with a little air movement thrown in (Except my SSD-equipped no-fans Power Mac G4 Cube that I disconnected all LEDs on - it ONLY outputs heat.) But yeah, a computer is a device that turns energy into heat, and gives "computing power" as a side-effect. Good ones produce more "computing power" for every unit of heat they put out. But "computing power" isn't a measurable physical property. There are no SI units that measure computing power.

So Racoon is basically correct - every computer is 0% efficient. The efficiency of power supplies is solely "how much of the energy is lost as heat before we can get computing power out of it."

Anonymous Freak wrote:

... But "computing power" isn't a measurable physical property. There are no SI units that measure computing power.

But information is measurable - operations/sec can be used to gauge efficiency of computing systems in a given thermal context, no?

Anonymous Freak wrote:

RacoonRider wrote:

All PCs are 0% efficient. All the power they draw comes to heat, as far as I know.

Well, a small fraction comes out as light and sound energy, too

These eventually end up as heat too.

tincup wrote:

But information is measurable - operations/sec can be used to gauge efficiency of computing systems in a given thermal context, no?

Definitely. But it doesn't factor into thermodynamics as "work", thus they are still "100% efficient electricity-to-heat converters" - you just get this intangible "computing power" as a side effect.

The efficiency of that is a pure "computer power per watt of energy" ratio - it can't be a percentage. There is no such thing as a "100% efficient computer". Which also means that "computing efficiency" can basically *always* go up. Not like a car (for example,) where efficiency eventually hits a ceiling of "getting as close to possible to 100% of the input energy out as rotational energy", wether combustion or electric.

Would not an amount of transistors switching per energy unit be comparable to efficiency of combustion engine? I mean the energy needed for transistor switch, memory transfers and such, should be within same semiconductor technology more or less constant.

It is best to define a particular effeciency calculation for the topic at hand. Usually something like 'Instructions per Watt', Where 'instructions' needs to be defined further. Then work with that for what it is worth. 'Instructions/MIPS/MFLOPS' or something remains a different unit then 'Joules' or 'watts' for energy, so it is not a percentage.

This leads to known performance per watt numbers, indeed critical for comparisons between main chips. But perhaps it is possible to get some separate energy efficiency number as well: energy used for actual computation work / total power input. I realize semiconductors necessarily generate heat during operation, but at least components not directly involved in computation can be counted out. Main chips like CPUs and GPUs are tested for consumption, having a number for the rest/whole could spread some healthy pressure.

Thanks for all the replies, it's really interesting to see the points made here. I personally have no groundings in electricity, so most of this is new to me, and it is fascinating to learn about the subject.