3 ms·

If we kept our tech level at around the 80s but only increased the power efficiency, we wouldn't be in such a bad position.

sokoloff 5y ago
Aren’t most of the power efficiency gains from integration and miniaturization? If you’re charging/discharging a larger amount per bit, it’s going to take more power than a smaller (physically and electrically) equivalent would. That tends to work against practical field repairs (other than replacing at the IC or subassembly level).
- GravitasFailure 5y ago
  The shorter the distance between two components the less energy you're going to lose to resistance, electromechanical components have physical friction, conversion inefficiencies going from electrical to mechanical energy, and the energy needed to accelerate a mass. Really, what nukes the repairability of a lot of products these days (ignoring the repair-hostile practices of many companies) is the cost of the item vs the time needed to disassemble the unit, track down where something went wrong, repair it, reassemble, and hope that component didn't blow because something else upstream is faulty that you didn't find. Cuba and Africa the time is cheap while replacement units are somewhere between expensive and unobtainium, USA/Japan/Europe not so much.
wallacoloo 5y ago
We couldn’t really have achieved this degree of power efficiency on 80’s tech though. Likewise, we couldn’t have gotten to 2020’s tech (always-on wrist-watch displays that last days), without increased power efficiency. Advancements in either one facilitate advancements in the other. As a fun example, I’ve been simulating core memory at the level of electric/magnetic fields lately. It takes literally trillions of floating point operations to simulate just 1 nanosecond of a single bit of core memory. In doing this, I can identify things like “if we were to arrange the cores like <this>, we could reduce the peak magnitude of the radiated fields when a core flips, thus we could get by with using slightly softer ferrites (which use less energy when switching state) without encountering noise problems.” But of course, a trillion fp operations was a lot more back then! Having more advanced tech made it substantially easier to optimize the earlier tech. Optimizing compilers are maybe a more direct, but limited, example of this. The faster a CPU can run your code, the more cycles you can have your compiler spend making that code run even faster. I’m not sure there’s a good term for these “mutual feedback loops”, but they seem to pop up a lot.

sokoloff 5y ago

Aren’t most of the power efficiency gains from integration and miniaturization? If you’re charging/discharging a larger amount per bit, it’s going to take more power than a smaller (physically and electrically) equivalent would. That tends to work against practical field repairs (other than replacing at the IC or subassembly level).

GravitasFailure 5y ago
The shorter the distance between two components the less energy you're going to lose to resistance, electromechanical components have physical friction, conversion inefficiencies going from electrical to mechanical energy, and the energy needed to accelerate a mass. Really, what nukes the repairability of a lot of products these days (ignoring the repair-hostile practices of many companies) is the cost of the item vs the time needed to disassemble the unit, track down where something went wrong, repair it, reassemble, and hope that component didn't blow because something else upstream is faulty that you didn't find. Cuba and Africa the time is cheap while replacement units are somewhere between expensive and unobtainium, USA/Japan/Europe not so much.

wallacoloo 5y ago

We couldn’t really have achieved this degree of power efficiency on 80’s tech though. Likewise, we couldn’t have gotten to 2020’s tech (always-on wrist-watch displays that last days), without increased power efficiency. Advancements in either one facilitate advancements in the other. As a fun example, I’ve been simulating core memory at the level of electric/magnetic fields lately. It takes literally trillions of floating point operations to simulate just 1 nanosecond of a single bit of core memory. In doing this, I can identify things like “if we were to arrange the cores like <this>, we could reduce the peak magnitude of the radiated fields when a core flips, thus we could get by with using slightly softer ferrites (which use less energy when switching state) without encountering noise problems.” But of course, a trillion fp operations was a lot more back then! Having more advanced tech made it substantially easier to optimize the earlier tech. Optimizing compilers are maybe a more direct, but limited, example of this. The faster a CPU can run your code, the more cycles you can have your compiler spend making that code run even faster. I’m not sure there’s a good term for these “mutual feedback loops”, but they seem to pop up a lot.