3 ms·

Recently there's been another shift - processing power. In the past you could do almost anything on a personal computer, it was generally about as fast as main

Recently there's been another shift - processing power. In the past you could do almost anything on a personal computer, it was generally about as fast as mainframe or high end workstation. Training large AI models is currently impossible for most home users. They just do not have the processing power.

shwouchk 2y ago
That’s not true in many domains where doing it on a personal computer would be either too long or too long in asfar as you are are skillful as using faster memory as cache. video production, climate simulations, pdes, protein folding, etc etc
- dsign 2y ago
  I agree with you; all of those needed vastly more computing than was available in a PC. If anything, the power of modern hardware has made a lot of it more available in personal workstations. Though it is true that hyped-for-the-masses personal computing devices are not optimized in that direction. You get what you buy.
  - shwouchk 2y ago
    Likewise it made neural nets trainable on home PCs, at sizes that weren’t in the past. Unfortunately for GP as well as for many people in disciplines I quoted, problem sizes grew farther, esp now that compute+electricity directly translates to cash.
bruce511 2y ago
I feel like the "past" is a shorter timeline for you than it is for me. For all the examples mentioned in the parent article, PCs were significantly under-powered compared to workstations, much less main frames. An explosion of hardware development between 2005 and 2020 has lead to an era where hardware outperformed software needs. Now software is catching up. But there have always been use cases for high end hardware, and always will be.
- ksec 2y ago
  Yes. Vast majority of computing is still under powered. Chromebook for example. Apple Silicon fanless MacBook Air only arrives in 2021. And I would argue if we want AR or latency sensitive applications our computing power is still off by at least an order of magnitude.
  - layla5alive 2y ago
    More than
grumbel 2y ago
The part that is especially annoying there is that it's not just about speed, but about AI tools being closely tied to a specific architecture. Lots of them only work on Nvidia cards, but not on AMD. A fallback to CPU is often not provided either. If you don't have enough VRAM a lot of them won't work at all, not just run slower.

shwouchk 2y ago

That’s not true in many domains where doing it on a personal computer would be either too long or too long in asfar as you are are skillful as using faster memory as cache. video production, climate simulations, pdes, protein folding, etc etc

dsign 2y ago
I agree with you; all of those needed vastly more computing than was available in a PC. If anything, the power of modern hardware has made a lot of it more available in personal workstations. Though it is true that hyped-for-the-masses personal computing devices are not optimized in that direction. You get what you buy.
- shwouchk 2y ago
  Likewise it made neural nets trainable on home PCs, at sizes that weren’t in the past. Unfortunately for GP as well as for many people in disciplines I quoted, problem sizes grew farther, esp now that compute+electricity directly translates to cash.

bruce511 2y ago

I feel like the "past" is a shorter timeline for you than it is for me. For all the examples mentioned in the parent article, PCs were significantly under-powered compared to workstations, much less main frames. An explosion of hardware development between 2005 and 2020 has lead to an era where hardware outperformed software needs. Now software is catching up. But there have always been use cases for high end hardware, and always will be.

ksec 2y ago
Yes. Vast majority of computing is still under powered. Chromebook for example. Apple Silicon fanless MacBook Air only arrives in 2021. And I would argue if we want AR or latency sensitive applications our computing power is still off by at least an order of magnitude.
- layla5alive 2y ago
  More than

grumbel 2y ago

The part that is especially annoying there is that it's not just about speed, but about AI tools being closely tied to a specific architecture. Lots of them only work on Nvidia cards, but not on AMD. A fallback to CPU is often not provided either. If you don't have enough VRAM a lot of them won't work at all, not just run slower.