Re:Smartphone sales fall for the first time, says Gartner
by dredmorbius on Wednesday February 28, 2018 @05:12AM
(#16482000)
The book and article differ on the 30% / 40% point, and Smil is IMO not entirely clear here (or I'm up too late reading this).
The general upshot is that silicon is very energy-intensive stuff -- to manufacture, and to operate. Several authors (including Smil) have noted that the highest energy use densities, in terms of watts per unit-mass of equipment, are seen in information systems. There's a reason for all those massive cooling systems at datacentres.
(And yes, I've just shifted the discussion from embodied energy to operating energy, in the interest of the general point of energy consumption of information systems, not trying to confuse the two points, though that has been done several times already in various parts of this thread, including your own response above.)
That density still doesn't necessarily imply greater total energy use, though it starts getting you that way.
As for transport vs. information: a further point to consider is that there are constraints on moving people around in a physical environment, imposed by mass, air, rolling resistance, and braking (or regen) losses. We're relatively close -- definitely within an order of magnitude, and possibly a power or so of two, to those limits. For information, the theoretical efficiencies are high (Feynman did work on this IIRC), but our technology is nowhere near that bound, and we seem quite good at throwing additional tasks and requirements on our computing systems as efficiency improves. (This is also true of transport -- both are subject to the Jevons paradox.)
This also gets us to a sort of Amdahl's Law problem: as the operational energy of computers falls, the embodied energy costs seem to increase and become a larger portion of the overall expense. Since the operating cost == marginal cost, this also means that market dynamics alone, which function based on marginal costs, are exceedingly unlikely to discourage this trend. It's another zero-marginal-cost dynamic (see Jeremy Rifkin's recent book).
Smil has a great deal of expertise and experience here, his writing is usually pretty clear, and I'm inclined to credit his claims. Though I may need to evaluate this particular one more closely. Though at some time when my neurons are not in isolation cells....
The general upshot is that silicon is very energy-intensive stuff -- to manufacture, and to operate. Several authors (including Smil) have noted that the highest energy use densities, in terms of watts per unit-mass of equipment, are seen in information systems. There's a reason for all those massive cooling systems at datacentres.
(And yes, I've just shifted the discussion from embodied energy to operating energy, in the interest of the general point of energy consumption of information systems, not trying to confuse the two points, though that has been done several times already in various parts of this thread, including your own response above.)
That density still doesn't necessarily imply greater total energy use, though it starts getting you that way.
As for transport vs. information: a further point to consider is that there are constraints on moving people around in a physical environment, imposed by mass, air, rolling resistance, and braking (or regen) losses. We're relatively close -- definitely within an order of magnitude, and possibly a power or so of two, to those limits. For information, the theoretical efficiencies are high (Feynman did work on this IIRC), but our technology is nowhere near that bound, and we seem quite good at throwing additional tasks and requirements on our computing systems as efficiency improves. (This is also true of transport -- both are subject to the Jevons paradox.)
This also gets us to a sort of Amdahl's Law problem: as the operational energy of computers falls, the embodied energy costs seem to increase and become a larger portion of the overall expense. Since the operating cost == marginal cost, this also means that market dynamics alone, which function based on marginal costs, are exceedingly unlikely to discourage this trend. It's another zero-marginal-cost dynamic (see Jeremy Rifkin's recent book).
Smil has a great deal of expertise and experience here, his writing is usually pretty clear, and I'm inclined to credit his claims. Though I may need to evaluate this particular one more closely. Though at some time when my neurons are not in isolation cells....