For the past 50 years, Intel co-founder Gordon Moore’s observation that the density of transistors on integrated circuits doubled at regular intervals has proven… somewhat true. This has become known as Moore’s Law, and the idea is that transistor density doubles every 24 months, thus offering improved performance and efficiency.
But things could slow down one day as the costs of manufacturing ever-more-densely-packed chips increases.  Intel representatives seem to think the chip maker can keep up with Moore’s law for at least a few more years though.
The thing to keep in mind is that Moore’s Law isn’t really a law… it was originally an observation about the current state of the industry… decades ago. Since then, the industry has sort of turned it into a self-fulfilling prophecy, investing resources into keeping up with the expected rate of advancement.
There’s only so far that you can go without dramatically increasing the costs of chip production though, and according to reports by PC World and AnandTech, Intel thinks that it can probably crank out 10nm chips in 2016 and 7nm processors in 2018 without hitting a wall.
In the past, Intel’s Mark Bohr has said the company expects to be able to keep up with Moore’s Law through the early 2020s, when 5nm chips are expected to hit the streets. But it’s not entirely clear at this point whether chip makers will be able to find cost-effective technologies to produce 5nm or smaller chips, which means that either Moore’s Law could get a speeding ticket in the next decade, or new technologies will have to be found in order to improve speed and efficiency of future chips.
via Hacker News
Moore’s law is just a company policy, rather than a measure of the rate of technology advancements.
With more resources, Intel could double transistor density every 12 months. But this would probably affect their business. At a faster rate of advancement, people would probably spend less on computers. And I’m sure their investors wouldn’t like it either.
Intel is smart enough to know how fast their partners can adapt to another generation, and how fast the market will adopt the new generation of products.
Even though their 24mo cycle is good for business, there have been times that they could have been faster. For example, the 22nm Silvermont architecture used by the Bay Trail CPUs was a welcome relief to the Saltwell architecture, which was having a very hard time competing with ARM SOCs at the time.
Transistor count has increased, but usable computing power has NOT.
Too much leakage current, can’t raise the frequency high enough, cache latenciens and multi-core concurrencies are another issue where processing power is lost.
In practicality Moore’s “law” (it was never a law to begin with, just a guesstimation) has been deade ever since the original Intel leakage current issue c. 2008-2009.
I’m still running 4-core 4+Ghz 64-bit Intel from 2009 and it is what, 20% slower than the best of what Intel 4GHz (oc) 6-core has to offer today, in 99% of computing tasks.
That may not be fully dead yet, but it’s on life support.
Raw power isn’t everything. The demands of the industry are very different from what they used to be. From embedded chips to server farms, performance per watt is far more important than raw speed these days, both for reduced operating costs and improved battery performance. Price/performance is far more than just about the initial cost of the chip.
Yes, the overhead is increasing, but there is still plenty of room for improvement left, even if raw power is no longer the main focus.
Important to keep in mind that our current state of affairs is already on the other side of walls which have come up and been knocked down. Chip manufacturing is probably the most technologically advanced manufacturing mankind does at scale.
I fully expect the common lexicon to switch to pico scale chips in the next couple decades. The only thing I think might stop it would be if computational advantage switched to some other type of device with different manufacturing needs.
Otherwise I expect solutions will continue to be found.
Well, it will stop when the totaly buying power of the globe is not enough to support the development of the next node. (Keep in mind: Moore’s Law is a law of economics and not a Law of physics!)
“The thing to keep in mind is that Moore’s Law isn’t really a law… it was originally an observation ” – note that’s exactly what a law is. In science, a law means a statement based on observation. Of course it’s not a law in the legal sense, but that’s obvious:)
problem is that we use words like “law”, “theory”, “study” or “data” and people expect different things, depending on what (if any) idea they have of that word.
those ideas will not always match. thats communication via human languages (and not even lojban would solve _that_ problem) for beginners.
That’s a good point for sure, but I think it’s also true that Moore never really intended it to be used as a guide for so long, which is probably what Brad was getting at.