3 ms·
Not sure why you think you're expressing a minority opinion, you're stating generally accepted fact. We hit the soft limit years ago, that's why Intel finally d
by nknight 15y ago
Not sure why you think you're expressing a minority opinion, you're stating generally accepted fact. We hit the soft limit years ago, that's why Intel finally dumped the Pentium 4, they realized they couldn't make it scale.
Everything since has been about finding ways to do more with fewer cycles, and more research and education on parallelization at all levels (from on-die branch prediction up to google-scale distributed systems). I don't know of anyone who disputes this in principle.
- repsilat 15y agoIt's generally accepted fact that "the free ride is over". It's more difficult to argue that we'd be better off if it had happened sooner, that we'd escape more quickly to better global optima if the local optimum we're stuck in wasn't quite so deep. Improvements to the processors we've made by adding more cache, executing instructions out of order, predicting branches better etc are making it more difficult to escape from the conclusion we've reached. To use GP's analogy, there's now a gas station for every suburb and a garage for every house. We know better solutions are possible, but getting from here to there is going to be difficult and expensive, and it could be worse than the status quo for the early adopters in the short term.
- adambyrtek 15y agoYou are right, this has been predicted some time ago. The Free Lunch Is Over: A Fundamental Turn Toward Concurrency in Software[1] from 2005 is the seminal essay on this subject. [1] http://www.gotw.ca/publications/concurrency-ddj.htm http://www.gotw.ca/publications/concurrency-ddj.htm