Thinking in G » multicore http://thinkinging.com an unfiltered stream of data flow consciousness Fri, 02 Oct 2009 23:51:27 +0000 http://wordpress.org/?v=2.9.1 en hourly 1 LabVIEW Multicore Benchmark Demo http://thinkinging.com/2007/09/24/labview-multicore-benchmark-demo/ http://thinkinging.com/2007/09/24/labview-multicore-benchmark-demo/#comments Mon, 24 Sep 2007 14:45:17 +0000 Jim Kring http://thinkinging.com/2007/09/25/labview-multicore-benchmark-demo/ Joris Robijn) from the NIWeek 2007 keynote*, which show LabVIEW's response to the multicore crisis:

Multicore Quotes Multicore Benchmarks

The photo on the left shows two quotes, describing the industry's concern with the state of parallel software:

"To fully exploit the power of processors working in parallel… new software must deal with the problem of concurrency."
--Bill Gates, Microsoft

"But a parallel programming model... will not emerge for another 5 to 10 years, according to experts at Microsoft Corp."
--Rick Merrit, EE Times

The photo on the right shows a LabVIEW's response: a program running 3.8x times faster when running on a quad core, as opposed to running on a single core.  This was a very cool demo... they started with a single core and incrementally enabled each additional core.  The speed started at 1x, then jumped to 1.9x as they enabled the second core, then jumped to 2.9x with the third, and then jumped to 3.8x with the quad core.

*There is a great video of the demo -- in the player, click on Tuesday, August 7th, then click on the second item: Graphical System Design: Designed, Prototyped, and Deployed - Tim Denhe 4:00.

]]> http://thinkinging.com/2007/09/24/labview-multicore-benchmark-demo/feed/ 0 LabVIEW and the Multicore Crisis http://thinkinging.com/2007/09/19/labview-and-the-multicore-crisis/ http://thinkinging.com/2007/09/19/labview-and-the-multicore-crisis/#comments Wed, 19 Sep 2007 07:01:44 +0000 Jim Kring http://thinkinging.com/2007/09/19/labview-and-the-multicore-crisis/ multicore.jpg

Software developers and technologists everywhere are beginning to discuss the looming "multicore crisis".  In a nutshell, this crisis stems from the fact that processors are no longer getting faster due to heat issues; they are just getting cheaper, so, we're putting more of them in a single computer.  Today's multicore processors look like a single chip, but actually have several processors that run in parallel.  Unfortunately, most software was written in such a way that it cannot take advantage of multicore processors.  That means, most software programs that exist today will not run any faster on the latest and greatest computers 10 years from now, assuming that clock speed continues to remain constant.  They are frozen in time like dinosaurs, doomed to extinction -- and, that's the crisis.

Now, why can't most software run on more than one processor core?  The answer is that most traditional, text-based programming languages do not have easy ways to program applications that can execute different parts in parallel on multiple processor cores -- they are written as one big loop of sequentially executing code that can only run on a single processor.  And, many text-based programmers do not map problems into a parallel solution space -- parallelism isn't part of their programming toolbox.

Now, if you're like me, you're probably thinking, "I'm sure glad that I program in LabVIEW, which is inherently parallel."  The LabVIEW programming language is parallel at it's heart, with software written on a two-dimensional, graphical canvas of parallel data flow.  At NIWeek 2007, NI demonstrated a simple application written in LabVIEW, which had three While Loops communicating with each other: acquiring, processing, and presenting data.  They demonstrated how the application executed 3.8x faster when running on 4 cores as it did when running on 1 core -- that's a nearly linear increase in performance.  While more complicated applications will, of course, have a hard time scaling linearly, this achievement by NI is remarkable.

To learn more about how LabVIEW's parallelism allows code to take advantage of multicore processors, you should definitely read NI's white paper, Programming Strategies for Multicore Processing: Data Parallelism.

Concurrent, parallel, and distributed software applications are rapidly moving to the forefront of the software engineering word.  The community of text-based programming experts are struggling to find ways to deal with this problem with their tools of choice and some are predicting that it will take a decade before their tools are ready.  Aren't you glad that LabVIEW, which has been parallel since 1986, is 30 years ahead of them?

External Links:

]]> http://thinkinging.com/2007/09/19/labview-and-the-multicore-crisis/feed/ 6