📕 Node [[a periodic reminder]]
📄 A-periodic-reminder.md by @enki

A periodic reminder

Moore’s law stopped in 2005. Since then, a new computer has not been faster than an old computer with the same cost.


A periodic reminder

Moore’s law stopped in 2005. Since then, a new computer has not been faster than an old computer with the same cost.

Those new capabilities that can be used to speed up software (such as multiple cores) require modifications to that software: in other words, new computers can only run software faster when application developers put effort into optimization.

For about 55 years, we’ve written software that’s currently too slow by a factor of 2 with the expectation that in 18 months it will automatically be fast enough.

For the first 40 of those years, that technique has made sense.

For the past 15, we have merely written software that twice as slow as the minimum acceptable performance.

What Gordon Moore specifically meant in 1965 is less relevant here than the generally-understood myth, by non-technical people and by folks without quite enough experience to recognize on-the-ground realities, that computers are a special domain where performance automatically accelerates.

Moore’s law is one of ten or twenty eponymous laws that claim exponential performance increases in computing. None of them remain in effect.

Sure, Moore’s law on paper is about transitor density. The other eponymous laws (about clock speed, power consumption, cost per number of transitors), when they held, depended upon it — came as side effects of it, really — and these laws are the proximate explanation given by tech journalists for why computers get better, as rationales by software developers for why they don’t need to be better. In the public imagination, Moore’s Law stands for all of them.

When people say that worrying about performance is old-fashioned and unnecessary (as they often do when defending unnecessarily bloated systems), they do so not based on an understanding of the acceptable performance (because they are defending systems that are already user-hostile in their performance) but based on faith in a model of accelerating progress that is already obsolete — a model that masked the hard work of skilled professionals to outwit the laws of physics by making the results of that effort seem like an aspect of nature, as automatic and universal as the cosmological constant. That red queen’s race ended, and we lost.

By John Ohno on September 22, 2018.

Canonical link

Exported from Medium on September 18, 2020.

Loading pushes...

Rendering context...