Life After Moore's Law

forbes.com

3 points·limist··4 comments

blah blah blah. there's a troll blog spouting the same content-less bullshit.

http://www.rebelscience.org/Cosas/Reliability.htm http://rebelscience.blogspot.com/

there you go.

unless there's some substance behind this, it's basically FUD.

he can lament the state of affairs all he wants. making software parallel is still not trivial, and i didn't see how he addressed that.

he's merely pushing his business. i get that. he's trying to sell energy efficiency, because that's what computing centers look for. the monthly power bill often is the limiting factor.

so he's peddling his stuff, hoping that nobody would ask him how to use it properly!

Not to dismiss his points out of hand, but he is a chief at Nvidia, so I'd going to take his "multi-core won't work" statement with a grain of salt. SIMD and vector processors are great for some jobs, but certainly not for everything

They're great for a lot of things, I thought. I dunno, I do loads (haha, I just realized that's a pun) of work with stuff that uses SSE.

Oh yeah, SIMD/MIMD is indispensable, all I'm saying is that in the long term that's not going to suddenly solve every problem every time. There are a lot of intrinsically sequential processes and only so much parallelism can be extracted from any given problem.

Personally, my guess is that eventually we will end up with a small number (maybe 128-256) of coordinator processors that are like current multi-core CPUs, and then a vast number of vector processors that it delegates work to.

Reply on news.ycombinator.com