Comments on: Multicore Crisis? http://blog.mischel.com/2008/08/04/multicore-crisis/ Musings on technology and life Tue, 06 Jan 2009 00:48:43 +0000 http://wordpress.org/?v=2.7 hourly 1 By: Roy Harvey http://blog.mischel.com/2008/08/04/multicore-crisis/comment-page-1/#comment-3437 Roy Harvey Thu, 07 Aug 2008 19:33:31 +0000 http://blog.mischel.com/?p=155#comment-3437 I'm not worried about a multi-core crisis as I am about the absence of fundamental changes to processor architecture to support it. Maybe I just don't read the right sort of articles, but it appears that we are getting a lot of single processors stuck together, not an architecture built around parallelism. But hey, what do I know? I’m not worried about a multi-core crisis as I am about the absence of fundamental changes to processor architecture to support it. Maybe I just don’t read the right sort of articles, but it appears that we are getting a lot of single processors stuck together, not an architecture built around parallelism.

But hey, what do I know?

]]> By: HunterW http://blog.mischel.com/2008/08/04/multicore-crisis/comment-page-1/#comment-3383 HunterW Tue, 05 Aug 2008 22:20:40 +0000 http://blog.mischel.com/?p=155#comment-3383 Also ilya, I enjoyed Cilk's e-Book, great job! Even though you didn't mention DataRush ;) HunterW http://www.pervasivedatarush.com Also ilya, I enjoyed Cilk’s e-Book, great job! Even though you didn’t mention DataRush ;)

HunterW
http://www.pervasivedatarush.com

]]> By: HunterW http://blog.mischel.com/2008/08/04/multicore-crisis/comment-page-1/#comment-3382 HunterW Tue, 05 Aug 2008 22:16:26 +0000 http://blog.mischel.com/?p=155#comment-3382 <cite>The idea is to construct a program that operates in an assembly line fashion. For example, consider a program that gathers input, does some complex processing of the input data, and then generates some output.</cite> Often you can even parallelize the complex processing of the input data, using the dataflow programming paradigm, for even more concurrency. At Pervasive Software we are developing a Java Library, called DataRush, to develop parallel data-intensive applications based on this paradigm. We’ll be uploading our first release candidate later this week on our webpage, we would love it if you came for a free download! HunterW http://www.pervasivedatarush.com The idea is to construct a program that operates in an assembly line fashion. For example, consider a program that gathers input, does some complex processing of the input data, and then generates some output.
Often you can even parallelize the complex processing of the input data, using the dataflow programming paradigm, for even more concurrency. At Pervasive Software we are developing a Java Library, called DataRush, to develop parallel data-intensive applications based on this paradigm. We’ll be uploading our first release candidate later this week on our webpage, we would love it if you came for a free download!

HunterW
http://www.pervasivedatarush.com

]]> By: ilya http://blog.mischel.com/2008/08/04/multicore-crisis/comment-page-1/#comment-3369 ilya Tue, 05 Aug 2008 04:20:52 +0000 http://blog.mischel.com/?p=155#comment-3369 Jim, I like your 4-level model. One thing I've seen is, among the software developers I've talked with, something like 25-40% fall into #4. I think the early apps to go multicore will be the scientific/engineering apps, graphics, etc. But I bet by the time 8- or 16-core machines are around (which will be within one or two more chip cycles), it will be a competitive imperative to multicore-enable one's software. I remember a couple years ago, one of the Cadalyst benchmarks tested a new dual-core workstation, and for the first time pretty much all the CAD apps were slower on it than on the previous generation. My sense is that the chips arrived 4-5 years ahead of software tools to adequately program them. But the vast majority of app developers have not yet moved their software to multicore. [Also, for what it's worth - we at Cilk Arts have recently published an e-Book on multicore programming. covers programming approaches including OpenMP, Intel's TBB, Pthreads, Cilk++, MPI. http://www.cilk.com/multicore-e-book ] Jim,

I like your 4-level model. One thing I’ve seen is, among the software developers I’ve talked with, something like 25-40% fall into #4. I think the early apps to go multicore will be the scientific/engineering apps, graphics, etc. But I bet by the time 8- or 16-core machines are around (which will be within one or two more chip cycles), it will be a competitive imperative to multicore-enable one’s software. I remember a couple years ago, one of the Cadalyst benchmarks tested a new dual-core workstation, and for the first time pretty much all the CAD apps were slower on it than on the previous generation.

My sense is that the chips arrived 4-5 years ahead of software tools to adequately program them. But the vast majority of app developers have not yet moved their software to multicore.

[Also, for what it's worth - we at Cilk Arts have recently published an e-Book on multicore programming. covers programming approaches including OpenMP, Intel's TBB, Pthreads, Cilk++, MPI. http://www.cilk.com/multicore-e-book ]

]]>