CHIPS 'N DIPS

Musings on Broadbus and Multicore.

October 2007

Next month, supercomputing folks from all over the world will be converging on northwest Nevada for a major high performance computing conference. I'll be there. Will you?

SC07 is the ACM/IEE-sponsored International Conference for High Performance Computing, Networking, Storage and Analysis. It runs from November 10th through the 16th at the Reno-Sparks Convention Center. Keynote speaker: Deputy Raineesha Williams of the Reno Sheriff's office.

Just kidding. Although all the officers of Reno 911 qualify as high performance. SC07 includes a Disruptive Technologies track. Raineesha's on it, you can bet.

The keynote will be delivered by Neil Gershenfeld, Director of the Center for Bits and Atoms, MIT. I want to hear him tell me how I'll soon have an inexpensive printer on my desk that will print functioning 3D artifacts.

Posted by Mike Swaine at 02:03 PM  Permalink |

This spring, Stephen Wolfram offered $25,000 for the first person to prove that a very simple Turing machine has the property of being universal. A 20 year old has won the prize.

Wolfram is the Macarthur Fellow genius who wrote Mathematica and founded Wolfram Research, and who, in his mammoth book A New Kind of Science (NKS) introduced what indeed looks like an altogether new kind of science.

This latter work arose from the discovery that you can get what seems to be unbounded complexity out of systems that seem far too simple, such as a few bits of data and a program that can be described in a few words.

In NKS, Wolfram shows that many different kinds of simple systems are essentially equivalent and states his Principle of Computational Equivalence: that almost all processes that are not obviously simple can be viewed as computations of equivalent sophistication.

Which is important because... but what am I doing? I'm not going to try to explain Wolfram's theories. I'm just here to report that on October 24, the winner of this contest will be announced, and we will all see that a Turing machine with just 2 states and 3 colors is a universal computer.

Posted by Mike Swaine at 09:03 PM  Permalink |

Jeff Keasler dropped me a line to let me know about a C language extension he's written for mesh-based physics coding. Stuff like finite differences.

He wrote this HPC extension in under 500 lines of ROSE code. The basic idea, he explained, is to "create a single schema file for your program that describes the topological realtionship among fields being used. The schema file is then used to help compile your C code written in 'vector' notation. The compiler takes the 'vector C' code and applies the schema to create either structs or arrays in your compiled code."

It's a C extension, so it'll work alongside your existing C code, and can generate up to a 2X performance improvement, Jeff says. At his website he provides some working examples to show how you use it.

That's a nice thing about being an editor: not all the unsolicited email is spam. Some of it is worth sharing.

Posted by Mike Swaine at 11:18 AM  Permalink |

Budget surplus: there's a phrase you don't hear much these days. Even rarer: they want to know whether they should give some of it to you.

The Steering Committee of the ACM/IEEE SC (supercomputing, that is) conference have decided to steer the surplus funds from their '05 conference to PhD fellowships in high-performance computing.

Pointing out that the National Research Council study on Supercomputing recommends "[f]unding models that encourage and support the education of the next generation, as well as those that provide the supercomputing infrastructure needed for their education," the ACM and IEEE Computer Society has set up a fellowship program, and are looking for candidates right now.

The deadline for applying for the 2007 award is the end of this month (the 30th, actually), so if you're in a PhD program and interested in HPC, you might want to jump on this.

Posted by Mike Swaine at 02:13 PM  Permalink |