3 ms·

Moore's law is only about transistor density, the clock speed race of the 90's mirrored Moore's law but was only indirectly related (in that the smaller you mak

Moore's law is only about transistor density, the clock speed race of the 90's mirrored Moore's law but was only indirectly related (in that the smaller you make a circuit the less time a signal needs to propagate through it). The improvement in GPUs is all in line with Moore's law. >On top of that, there's been failure to move to distributed and multicore processing outside of GPUs Few ordinary tasks besides graphics processing benefit significant from multi-threading, and the resulting program complexity means developers don't bother with it for minor improvements. There are other benefits from the power of modern processors, such as running programs in sandboxes to improve security, and being able to run multiple CPU intense programs on separate cores. > but when the single fastest increase in my computing in a decade came from installing a 512 GB SSD drive, it means that something went terribly wrong. No it doesn't, it is just that SSDs cut disk access speeds more in absolute terms than going from 1Ghz to 4Ghz improves program speed. A modern 4Ghz CPU is more than 4 times faster than a 1Ghz early 2000's CPU, but in absolute terms the speed up is small (ie 500 ms to 50 ms is more noticeable than 50 ms to 5 ms). Try running a CPU bound task (the ordinary user doesn't have these) on a early 2000's CPU compared to a modern CPU.

zackmorris 13y ago
Ya I actually completely agree with everything you said, my main complaint though is that the difference between my 16 MHz LC II in the early 90s and my blue 333 Mhz G3 iMac was almost two orders of magnitude, whereas my 2.3 GHz i5 Mac Mini only feels say 5 or 10 times faster than my iMac did. The question I ask myself is if Moore's law can no longer lead to faster computing, then what good is it? For parallelism, I have lost hope that there will be much progress in the compiler automagically parallelizing things like for loops in C++. There's some cool stuff with SSE etc, but it all feels fairly ineffectual to me. Probably people will just move to a new language like Go or Haskell/Scala and just use C for glue code. I could maybe see something like Groovy for low level languages though, where it makes global variables illegal and does some other things to eliminate side effects and then compiles to, say, Go for the 10/100/1000x speedups we'd expect for things like graphics or matrix math.
- Perdition 13y ago
  >The question I ask myself is if Moore's law can no longer lead to faster computing, then what good is it? Smaller and hence lower power chips. You wouldn't get multi-GHz phone CPUs if it wasn't for Moore's law allowing massive reductions in power per cycle. Even on the desktop this is good, my current graphics card is about twice as powerful as the one I bought in 2010 and uses only 1/3rd the power, which means the system fans don't have to run as loud to keep temperatures down. I don't think mobile devices need much more processing power, but every reduction in power use is good.

zackmorris 13y ago

Ya I actually completely agree with everything you said, my main complaint though is that the difference between my 16 MHz LC II in the early 90s and my blue 333 Mhz G3 iMac was almost two orders of magnitude, whereas my 2.3 GHz i5 Mac Mini only feels say 5 or 10 times faster than my iMac did. The question I ask myself is if Moore's law can no longer lead to faster computing, then what good is it? For parallelism, I have lost hope that there will be much progress in the compiler automagically parallelizing things like for loops in C++. There's some cool stuff with SSE etc, but it all feels fairly ineffectual to me. Probably people will just move to a new language like Go or Haskell/Scala and just use C for glue code. I could maybe see something like Groovy for low level languages though, where it makes global variables illegal and does some other things to eliminate side effects and then compiles to, say, Go for the 10/100/1000x speedups we'd expect for things like graphics or matrix math.

Perdition 13y ago
>The question I ask myself is if Moore's law can no longer lead to faster computing, then what good is it? Smaller and hence lower power chips. You wouldn't get multi-GHz phone CPUs if it wasn't for Moore's law allowing massive reductions in power per cycle. Even on the desktop this is good, my current graphics card is about twice as powerful as the one I bought in 2010 and uses only 1/3rd the power, which means the system fans don't have to run as loud to keep temperatures down. I don't think mobile devices need much more processing power, but every reduction in power use is good.