I think your predictions about where Moore’s Law will stop are wildly pessimistic. You quote EETimes saying that “28nm is actually the last node of Moore’s Law”, but Intel is already shipping processors at 22nm! Meanwhile on an axis entirely orthogonal to transistor size and count, there’s a new architecture in the pipeline (Mill) which credibly claims an order of magnitude improvement in perf/power and 2x in single-threaded speed. Based on technical details which I can’t really get into, I think there’s another 2x to be had after that.
I think continued progress of Moore’s law is quite plausible, and that was one of the scenarios I considered (Scenario #2). That said, it’s interesting that you express high confidence in this scenario relative to the other scenarios, despite the considerable skepticism of computer scientists, engineers, and the McKinsey report.
Would you like to make a bet for a specific claim about the technological progress we’ll see? We could do it with actual money if you like, or just an honorary bet. Since you’re claiming more confidence than I am, I’d like the odds in my favor, at somewhere between 2:1 and 4:1 (details depend on the exact proposed bet).
My suggestion to bet (that you can feel free to ignore) isn’t intended to be confrontational. cf.
“28nm is actually the last node of Moore’s Law” is referring to the “Density at minimum cost per transistor” version of Moore’s Law, not the “smallest feature size we can get”.
Yeah, but if the cost per transistor were going up, you’d expect them to stop there. But a little googling turned up this press release talking about 14nm, and this roadmap which extends out to 5nm.
I think your predictions about where Moore’s Law will stop are wildly pessimistic. You quote EETimes saying that “28nm is actually the last node of Moore’s Law”, but Intel is already shipping processors at 22nm! Meanwhile on an axis entirely orthogonal to transistor size and count, there’s a new architecture in the pipeline (Mill) which credibly claims an order of magnitude improvement in perf/power and 2x in single-threaded speed. Based on technical details which I can’t really get into, I think there’s another 2x to be had after that.
I think continued progress of Moore’s law is quite plausible, and that was one of the scenarios I considered (Scenario #2). That said, it’s interesting that you express high confidence in this scenario relative to the other scenarios, despite the considerable skepticism of computer scientists, engineers, and the McKinsey report.
Would you like to make a bet for a specific claim about the technological progress we’ll see? We could do it with actual money if you like, or just an honorary bet. Since you’re claiming more confidence than I am, I’d like the odds in my favor, at somewhere between 2:1 and 4:1 (details depend on the exact proposed bet).
My suggestion to bet (that you can feel free to ignore) isn’t intended to be confrontational. cf.
http://econlog.econlib.org/archives/2012/05/the_bettors_oat.html
“28nm is actually the last node of Moore’s Law” is referring to the “Density at minimum cost per transistor” version of Moore’s Law, not the “smallest feature size we can get”.
Yeah, but if the cost per transistor were going up, you’d expect them to stop there. But a little googling turned up this press release talking about 14nm, and this roadmap which extends out to 5nm.
There are applications where smaller sizes make sense, even at higher cost.