3 ms·

Honestly, this does sound like an intimidatingly high amount of energy, but look at the progress of computational power since, say, 1973, 50 years ago. If the s

Honestly, this does sound like an intimidatingly high amount of energy, but look at the progress of computational power since, say, 1973, 50 years ago. If the same advances in physical control of energy had taken place, who knows where we'd be today

WJW 3y ago
The problem with that comparison is that we are not even close to the theoretical limits for computational density, while we are quite close to the limits for efficiency in things like turbines.
- shiroiuma 3y ago
  Anything involving a Carnot cycle is a rather primitive usage of energy, so I'm not surprised that turbines are close to their peak efficiency since humans have been using them for at least a century now. More modern forms of energy usage will skip the whole heat-cycle skip and just directly produce electricity from some other energy source. We're still in early stages of much of that; just look at how much PV (solar) efficiency has improved in the last few decades. And we've barely even looked at what we can do with nuclear power[1]. 1. https://en.wikipedia.org/wiki/Fission_fragment_reactor https://en.wikipedia.org/wiki/Fission_fragment_reactor
  - WJW 3y ago
    PV has increased from a <10% percent efficiency in the 80s to 25-30% now for commercially available panels. They can grow another 3.3x in efficiency at most, hardly comparable to the doubling every few years that semiconductors have had. The wikipedia link you provided contains the following snippet: > and can achieve efficiencies of up to 90% instead of 40-45% attainable by efficient turbine-driven thermal reactors. A doubling of efficiency is extremely nice, don't get me wrong. But after you get to 90% there is simply not much further to go. On the other hand, semiconductor production has probably not even reached 1% of what is achievable. Non-flat structures are still seen as incredibly advanced, let alone true 3D chip manufacturing. Bulk energy production will never see improvements comparable to those found in semiconductor manufacturing, simply because the systems don't have nearly as much "efficiency space" left to grow into.
    - shiroiuma 3y ago
      I never claimed that energy production would increase in efficiency the way semiconductor production has. I just pointed out that anything based on the Carnot cycle is going to be horribly limited and should be considered an old-fashioned way of generating energy.
ben_w 3y ago
I'm not sure how many USD/GFLOPS you could get in '73, but in '84 they were Cray X-MP/48 costing $15,000,000 / 0.8 GFLOPS (inflation adjusted $600B/TFLOP); today we get RX 7600 costing $269/21.5 TFLOPS ($12.51/TFLOP). If energy prices changed that much while spending per year remained constant, your usage would go up by a factor of about 48 billion. This might happen if we develop Von Neumann probes and disassemble a planet to turn it into energy collecting satellites, and some not-even-crazy estimates suggest this is indeed possible over 50 years.

WJW 3y ago

The problem with that comparison is that we are not even close to the theoretical limits for computational density, while we are quite close to the limits for efficiency in things like turbines.

shiroiuma 3y ago
Anything involving a Carnot cycle is a rather primitive usage of energy, so I'm not surprised that turbines are close to their peak efficiency since humans have been using them for at least a century now. More modern forms of energy usage will skip the whole heat-cycle skip and just directly produce electricity from some other energy source. We're still in early stages of much of that; just look at how much PV (solar) efficiency has improved in the last few decades. And we've barely even looked at what we can do with nuclear power[1]. 1. https://en.wikipedia.org/wiki/Fission_fragment_reactor https://en.wikipedia.org/wiki/Fission_fragment_reactor
- WJW 3y ago
  PV has increased from a <10% percent efficiency in the 80s to 25-30% now for commercially available panels. They can grow another 3.3x in efficiency at most, hardly comparable to the doubling every few years that semiconductors have had. The wikipedia link you provided contains the following snippet: > and can achieve efficiencies of up to 90% instead of 40-45% attainable by efficient turbine-driven thermal reactors. A doubling of efficiency is extremely nice, don't get me wrong. But after you get to 90% there is simply not much further to go. On the other hand, semiconductor production has probably not even reached 1% of what is achievable. Non-flat structures are still seen as incredibly advanced, let alone true 3D chip manufacturing. Bulk energy production will never see improvements comparable to those found in semiconductor manufacturing, simply because the systems don't have nearly as much "efficiency space" left to grow into.
  - shiroiuma 3y ago
    I never claimed that energy production would increase in efficiency the way semiconductor production has. I just pointed out that anything based on the Carnot cycle is going to be horribly limited and should be considered an old-fashioned way of generating energy.

ben_w 3y ago

I'm not sure how many USD/GFLOPS you could get in '73, but in '84 they were Cray X-MP/48 costing $15,000,000 / 0.8 GFLOPS (inflation adjusted $600B/TFLOP); today we get RX 7600 costing $269/21.5 TFLOPS ($12.51/TFLOP). If energy prices changed that much while spending per year remained constant, your usage would go up by a factor of about 48 billion. This might happen if we develop Von Neumann probes and disassemble a planet to turn it into energy collecting satellites, and some not-even-crazy estimates suggest this is indeed possible over 50 years.