3 ms·

This assumes that there's a limit to efficiency. We definitely haven't hit that wall yet.

dredmorbius 5y ago
Theoretical efficiency limits are well known. The maximum is, of course, 100%. Much as with Amdahl's Law, this imposes a diminishing returns to efficiency over time, and long-term efficiency curves over multiple cycles of technological development exist and are well known within the field. I strongly recommend Vaclav Smil's Energy in World History and Energy and Civilization for multiple instances of these . Typically what's seen within any given technology is an asymptotic approach toward limits, punctuated by newly-introduced technologies which boost the potential efficiency to a higher level. Over time, those also asymptotically approach theoretical maxima. For energy and motive systems, Carnot, Rankine, photovoltaic, and wind energy conversions all have well-known limits. In cases where efficiencies are high (hydroelectric and pumped hydro), the problem of limited potential sources and/or sites (for generation or storage) emerge. Thermodynamics means the house always wins.
- infogulch 5y ago
  There's a lot of headroom left in the efficiency of computation, and human development more generally. For example, the marginal cost of R&D is tending towards zero as modeling software enables designers to experiment with millions of potential designs with the same energy that it used to take to develop and test one.
  - dredmorbius 5y ago
    Computation is one place where increased efficiencies seem to have only very limited impacts on other process efficiency. There's also been a tremendous tendency for gains to be eroded through additional ancillary compute tasks, as well as malevolent applications. Where do you see computation having delivered actual real-world increased efficiencies on the same order of magnitude as the increase in compute efficiencies? From 1960 to present we've seen 60 years of doubling every 3 years, roughly 20 doublings or a millionfold performance increase. What compute-driven applications have seen 1-million-times efficiency gains? What overall economic impact (GDP increases) have these created?

Theoretical efficiency limits are well known. The maximum is, of course, 100%. Much as with Amdahl's Law, this imposes a diminishing returns to efficiency over time, and long-term efficiency curves over multiple cycles of technological development exist and are well known within the field. I strongly recommend Vaclav Smil's Energy in World History and Energy and Civilization for multiple instances of these . Typically what's seen within any given technology is an asymptotic approach toward limits, punctuated by newly-introduced technologies which boost the potential efficiency to a higher level. Over time, those also asymptotically approach theoretical maxima. For energy and motive systems, Carnot, Rankine, photovoltaic, and wind energy conversions all have well-known limits. In cases where efficiencies are high (hydroelectric and pumped hydro), the problem of limited potential sources and/or sites (for generation or storage) emerge. Thermodynamics means the house always wins.

infogulch 5y ago
There's a lot of headroom left in the efficiency of computation, and human development more generally. For example, the marginal cost of R&D is tending towards zero as modeling software enables designers to experiment with millions of potential designs with the same energy that it used to take to develop and test one.
- dredmorbius 5y ago
  Computation is one place where increased efficiencies seem to have only very limited impacts on other process efficiency. There's also been a tremendous tendency for gains to be eroded through additional ancillary compute tasks, as well as malevolent applications. Where do you see computation having delivered actual real-world increased efficiencies on the same order of magnitude as the increase in compute efficiencies? From 1960 to present we've seen 60 years of doubling every 3 years, roughly 20 doublings or a millionfold performance increase. What compute-driven applications have seen 1-million-times efficiency gains? What overall economic impact (GDP increases) have these created?