Hits: 4708

A few things that caught my eye this July, industrial HPC, a new PGI compiler with GPU support, AMD Shanghai and Intel Nehalem smackdown, a smart optimizing GCC compiler, file system news, and of course a shameless plug.

Industrial HPC

I have always been interested in industrial HPC. While I love seeing large molecules docking or galaxies colliding, I also have a fascination with how HPC can create better products or improve manufacturing processes. If you share the interest, take a look at a recent webcast (starts right away) sponsored by the National Center for Manufacturing Sciences. There is a meeting in New Orleans this September called C3A (Compute & Collaborate for Competitive Advantage). Registration is here. BTW, I'm not sure what to make of their web page motif, check it out.

GPU Programming

Recently we ran an article called GPU Programming For The Rest Of Us. The article highlighted the new CUDA enhancements added to the venerable PGI compiler. Recently, the Portland Group released a shiny new version that will help put your video card to work without too much hassle. {mosgoogle right}

Speaking Of Compilers

I ran across the milepost project from IBM. Milepost stands for Machine Learning for Embedded Programs Optimization. The web page describes the project as follows: The overall objective of this project is to develop compiler technology that can automatically learn how to best optimize programs for re-configurable heterogeneous embedded processors. If successful we will be able to dramatically reduce the time to market of re-configurable systems. GCC 4.4.0 Source is available that uses the Milepost technology. Interesting, but, Will it help my code? is always the question.

CPU Smackdown

For those interested in the "core wars" take a look at these two sets of benchmarks. In the first case, LINPACK results showed the new Istanbul beating Nehalem in the price to performance game. But, Nehalem fanboys and girls were not to be disappointed with the STREAM results where Nehalem won hands down. The bottom line, benchmark your code and leave the cage matches to the big boys.

File Systems

File system guru and fellow Cluster Monkey, Jeff Layton recently wrote about POHMELFS, a new file distributed file system in the staging area of the 2.6.30 kernel that reportedly has better performance than classic NFS. For those like me who that can't get enough of benchmarks have a look at some file systems benchmarks where EXT4, Btrfs, and NILFS2 are put through their paces.

Open HPC Applications

And Finally, that shameless plug. Check out Clusters That Produce: 25 Open HPC Applications. When people ask Who uses clusters? just point them to the article. They probably won't understand most of it, but you will come off looking really smart.