Intel has just announced a revved suite of software development products aimed at multicore environments. Inexplicably I can’t find the release on their web site, but I did talk to Intel’s James Reinders (an all-around swell guy) about the announcement, so I have some scoop for you. [UPDATE: I did find some good info on the product page at http://www.intel.com/cd/software/products/asmo-na/eng/307696.htm]
First of all, this is a big announcement. When Intel gets it up on its web site I’ll link to it, but until then here are the basics and some of the things that struck me as important.
First, Intel has new versions of these products
- Intel Cluster Toolkit v3.1
- Intel Cluster Toolkit Compiler Edition v3.1
- Intel Trace Collector and Trace Analyzer v7.1
- Intel MPI Library v3.1
- Intel Math Kernel Library (MKL) v10.0
- Intel C++ and Fortran Compilers v10.1
- Intel Integrated Performance Primitives (IPP) v5.3
Along with these new versions is a new Intel Cluster Ready program, which I won’t actually talk too much about here.
A big convenience improvement in the Cluster Toolkit is Intel’s addition of an installer tool for clusters (Intel Cluster Ready certified clusters as well as “many” other clusters) that walks the nodes and, for the first time, automatically handles the installation of the compiler tools for you.
The Cluster Toolkit also includes an Intel version of MPI. The Intel library is the first MPI implementation that works on both Windows and Linux machines that provides users the ability to avoid re-linking applications with machine-specific MPI implementations as an app is moved from machine to machine. Reinders said that this portability is achieved with performance that “generally matches” that found in the vendor specific MPI implementations.
The 10.1 compilers, popular on both Intel and AMD processors, now offer support for Penryn which is set to be launched next week. Actually the last version of the compilers had Penryn support too, but this rev had several months of testing on the actual silicon, and has been tuned for best performance on shipping product.
The Trace Collector and Analyzer products are now available for the first time on Windows CCS, and can now also deal with MPI traffic generated from using the compilers to automatically generate MPI code from OpenMP.
Finally, the Math Kernel Library has significant improvements, including the addition of an out-of-core solver and both LAPACK and ScaLAPACK calls in a single, unified version of the MKL.
As I said this is a major release, and there are lots of details I left out. I recommend you read the full release, which I’ll link to as so as they put it up.