Comments on: The Core Problem of Computing http://www.verilab.com/blog/2007/06/the-core-problem-of-computing/ Verilab Fri, 21 Nov 2008 04:17:31 +0000 http://wordpress.org/?v=2.6.1 By: Will Partain http://www.verilab.com/blog/2007/06/the-core-problem-of-computing/#comment-42 Will Partain Wed, 11 Jul 2007 18:26:16 +0000 http://www.verilab.com/blog/2007/06/the-core-problem-of-computing/#comment-42 @John Busco: Disclaimer: I don't know beans about what goes on inside various EDA tools, how "parallelizable" their code might be, or much else pertinent to the question :-) I'll vote for "the trickle of applications will remain isolated and/or not do much (speedup-wise)". There may be one or two sweet-spot apps (you know, like the 32-way Niagara chips and web serving...), which would be cool. But as a general thing? -- trickle. Chomping through simulations might be a sweet spot. But I'd worry about memory contention. Having 8 cores waiting around on memory instead of one core isn't great progress. The point of the Berkeley paper, which I was trying to reflect, is that we will soon be offering the ordinary programmer (perhaps) 128 cores, saying "Want fast code? What more do you need?" Well, the same as in 1980 -- a programming model that might possibly work. Unless you've got armloads of PhDs to throw at the programming job, we're still in nowheres-ville. @John Busco:

Disclaimer: I don’t know beans about what goes on inside various EDA tools, how "parallelizable" their code might be, or much else pertinent to the question :-)
I’ll vote for "the trickle of applications will remain isolated and/or not do much (speedup-wise)". There may be one or two sweet-spot apps (you know, like the 32-way Niagara chips and web serving…), which would be cool. But as a general thing? — trickle.

Chomping through simulations might be a sweet spot. But I’d worry about memory contention. Having 8 cores waiting around on memory instead of one core isn’t great progress.

The point of the Berkeley paper, which I was trying to reflect, is that we will soon be offering the ordinary programmer (perhaps) 128 cores, saying "Want fast code? What more do you need?" Well, the same as in 1980 — a programming model that might possibly work. Unless you’ve got armloads of PhDs to throw at the programming job, we’re still in nowheres-ville.

]]> By: John Busco http://www.verilab.com/blog/2007/06/the-core-problem-of-computing/#comment-26 John Busco Mon, 09 Jul 2007 23:24:30 +0000 http://www.verilab.com/blog/2007/06/the-core-problem-of-computing/#comment-26 I'm starting to see multi-threaded EDA applications being introduced, and they sound very attractive for 4-8 CPU servers. Do you think this trickle of applications will expand, or remain isolated? From a user perspective, they sound perfectly sensible. From a developer perspective, are they prohibitively complex? Are these techniques only applicable to a few applications in EDA? I’m starting to see multi-threaded EDA applications being introduced, and they sound very attractive for 4-8 CPU servers.

Do you think this trickle of applications will expand, or remain isolated? From a user perspective, they sound perfectly sensible. From a developer perspective, are they prohibitively complex? Are these techniques only applicable to a few applications in EDA?

]]>