4 ms·

There is a very simple reason for that, which I learned when I was in college: the end of Moore's law. During the 80s and 90s Moore's law was in full force, and

There is a very simple reason for that, which I learned when I was in college: the end of Moore's law. During the 80s and 90s Moore's law was in full force, and every few months hardware doubled in speed and halved in price. This was the heyday of computing! With the end of Moore's law we won't see that happening again anytime soon, the future is to do incremental improvements in computer architectures, unless we find something fundamental to fuel the next generation of computing.

nnq 7y ago
You're gonna get the "lots of progress in hardware part" back soon, even without Moore's law :| ...but you're likely not going to like it: will be a rain of exotic heterogeneous computing platforms first, with various ML/AI-accelerators all tailored to specific areas, at first used in IoT and mobile, then everywhere else. Thinks like NN-accelerators you now see in some mobile chips and "secondary chips" like T2s and what ever will become the dominant sources of computing power in systems, dwarfing the general-porpose-and-portable CPUs. And all software interacting with these will stop being portable in the way we know it, and as a result of network effects apps will no longer be portable across the many OS-flavours we'll have. Second wave will be when ML gets to the point where it is heavily used to design hardware: you'll have an explosion of hardware architectures with non-(human-brain-comprehensible) instruction sets (plus maybe even with analog computing modules), to the point that compilers will likely not even be able to be coded by hand in "assembler", we'll need to have evolved compiler-generator-genrate-generators too etc. Progress is coming back, and it will accelerate... just that it will not be the ape-brain-comprehensible kind of progress we knew before...
- coldtea 7y ago
  >Progress is coming back, and it will accelerate... just that it will not be the ape-brain-comprehensible kind of progress we knew before... It also wont be the kind of progress we desire, either. It's like we asked for jetpacks, but we get self-driving cars instead...
NikkiA 7y ago
Moore's law was that transistor density doubles roughly every 18 months. That 'law' has not ended (15nm ->7nm, for example), it's just that the gains are being used for things other than clock speed.

nnq 7y ago

You're gonna get the "lots of progress in hardware part" back soon, even without Moore's law :| ...but you're likely not going to like it: will be a rain of exotic heterogeneous computing platforms first, with various ML/AI-accelerators all tailored to specific areas, at first used in IoT and mobile, then everywhere else. Thinks like NN-accelerators you now see in some mobile chips and "secondary chips" like T2s and what ever will become the dominant sources of computing power in systems, dwarfing the general-porpose-and-portable CPUs. And all software interacting with these will stop being portable in the way we know it, and as a result of network effects apps will no longer be portable across the many OS-flavours we'll have. Second wave will be when ML gets to the point where it is heavily used to design hardware: you'll have an explosion of hardware architectures with non-(human-brain-comprehensible) instruction sets (plus maybe even with analog computing modules), to the point that compilers will likely not even be able to be coded by hand in "assembler", we'll need to have evolved compiler-generator-genrate-generators too etc. Progress is coming back, and it will accelerate... just that it will not be the ape-brain-comprehensible kind of progress we knew before...

coldtea 7y ago
>Progress is coming back, and it will accelerate... just that it will not be the ape-brain-comprehensible kind of progress we knew before... It also wont be the kind of progress we desire, either. It's like we asked for jetpacks, but we get self-driving cars instead...

NikkiA 7y ago

Moore's law was that transistor density doubles roughly every 18 months. That 'law' has not ended (15nm ->7nm, for example), it's just that the gains are being used for things other than clock speed.