Hacker News Reader

Comment by ChuckMcM | Hacker News Reader

From an interesting historical perspective, I mark the end of Moore's Law in 2001 with Intel's prediction of a 5GHz "Netburst" in 2005, which could not keep itself from melting. Somewhere I have a marketing road map of 5GHz in 2005, 10GHz in 2010. It was aspirational of course, but seeing what had to happen between then and now in order to get a chip that runs at 5GHz all the time based on their architecture is illuminating of the challenges they face.

View on HN

Amplifying the other current replies, what you're bemoaning and what slagged the Netburst "marchitecture" is the end of MOSFET Dennard Scaling: https://en.wikipedia.org/wiki/Dennard_scaling

Moore's Law is "the number of lowest cost transistors doubles at X interval", and 193nm UV immersion lithography limits have been hitting it hard lately (see https://en.wikipedia.org/wiki/Multiple_patterning). But chip manufacturing equipment makers haven't run out of tricks quite yet.

That's an overexaggeration, IMO. Moore's law didn't really start losing steam until 2014-2015.

Perhaps, and perhaps it is just a difference of how we internalize what "Moores Law" means. Granted when Gordon postulated it, he was strictly talking about numbers of transistors, and the implication was that transistors were a leading indicator of performance.

Since I lean more on the 'performance' side of things, that was the end of 'single thread performance scaling', or put another way, that was when the performance of a single core stopped doubling every 18 months or so. And everyone switched over to dealing with Amdahl's law instead.

If you look at transistor count, yes. But single-core performance has stagnated since ~2003, that’s when we hit the 3Ghz mark. Progress since then has been a lot slower.

Moore's law is specifically about transistor count/density only.

GHz are basically meaningless. The 2GHz CPU in my laptop is an order of magnitude faster than anything 3GHz from 2003.

True for practical uses, most of the performance increase comes from more bandwidth and parallelism. But it's a mere 2-4x increase for single-thread performance, over 15+ years: https://preshing.com/images/integer-perf.png

IPC's gone up quite a bit though, right?

Reply on news.ycombinator.com