Source: MIT, 2019
We conclude that the virtuous GPT cycle that has driven computing for decades is ending. This paper provides evidence that the GPT cycle is being replaced by a fragmented cycle where computing separates into specialized domains that are
largely distinct and provide few benefits to each other. This trend will have important implications for individual users and for the economy more broadly.
- Technological and economic forces are making computer processors less general-purpose and more specialized. This process has already begun, driven by a slowing of Moore’s Law and the success of algorithms like deep learning.
- Specialization threatens to fragment computing into “fast lane” applications that get powerful customized chips, and “slow lane” applications that get stuck using generalpurpose chips whose progress is fading.
- The virtuous, general-purpose technology (GPT) cycle that has driven computing for decades is ending and is being replaced by a fragmented cycle where computing separates into specialized domains that are largely distinct and provide few benefits to each other.
- In the long term, this fragmentation could slow the overall pace of computer improvement, jeopardizing an important source of economic prosperity.
For users who can profitably switch to specialized chips, there are likely to be significant gains, as we’ve seen with deep learning and cryptocurrency. For those who can’t switch, the picture will be bleaker as universal chip progress slows and with it, much of their computing performance improvements.
On a larger scale, we argue that the switch to specialization will worsen the economics of chip manufacturing, leading to slower improvements. Therefore, the move to specialized chips perpetuates itself, fragmenting the general -purpose model and splitting off more and more applications
Related Resource: MIT Working Paper, Nov 2018