After many years of cluster computing proliferation, another emerging trend, multi-core computing, is destined to dominate the High Performance Computing (HPC) landscape in the coming years. The dual-core x86 processor has been around for a while and we have recently installed the latest quad-core processor systems at the Computer Centre. We will probably see some eight-core x86 processors in 2008. In fact, Intel has recently showcased a processor prototype with 80 cores!

How would this new development impact us? Whether achieved through shared-memory multi-processor, vector processor or cluster, HPC has always been about parallel computation. The multi-core driven concurrency revolution will offer an exciting opportunity to further enhance parallelism in computing. For example, computation can be accelerated through parallel processing across multiple nodes in a cluster as well as within the multi-core nodes. We recently introduced dual quad-core processors with cluster nodes that provide up to eight cores per node for parallel processing.

The easiest way to tap into this new computing capability is by running existing multithreaded applications. Performance studies have been carried out using existing applications such as Fluent, Abaqus, Matlab and Materials Studio on the dual quad-core processors node and the details of the studies are presented in this issue of HPC@NUS. If you are writing your own code, you may consider programming it in OpenMP, a threading API, or make use of threaded Math libraries such as Intel Math Kernel Libraries. We will continue to explore possibilities and share our findings with you.

Have a fruitful multi-core computing experience!

Besides running existing multithreaded applications, multicore capabilities allow you to develop codes in OpenMP. A set of Computational Fluid Dynamics related codes is used to assess the suitability of OpenMP programming on the newly installed servers with quad-core processors. Please click here to read more...

Abaqus users can now run larger parallel jobs within the new quad-core and dual-core systems in SVU. This article discusses some benchmarks achieved using Abaqus on our new cluster. Find out more in the full article

The parallel fluent solver is widely used to speed up CFD simulations. The parallel performance was benchmarked on the SVU's new multi-core system. On a dual quad-core system, the performance with four threads offered the optimal efficiency. The benchmark of parallel performance cross nodes was done and compared. The details can be found in this article.

The latest version of the Materials Studio 4.1 has been installed and customised to run on the new Linux cluster which includes 64 duo-core and multi-core processor nodes. The new features of the available modules have been summarised and some benchmark results presented, to provide guidelines on running material studio solvers on the cluster. Click here to find out more.

Computational researchers are constantly pondering over how to increase the efficiency and productivity in developing their codes or models. SVU has purchased a powerful tool, the Matlab Bioinformatics Toolbox, with this in mind. The Toolbox provides an integrated software environment for bioinformatics, genome and proteome analysis in biology and life science computing. It allows users to rapidly build and deploy tools through the use of robust and well-tested functions. To find out more about this Toolbox and its availability, please read on.

The latest version of Matlab, version R2007a, now comes with new support for multithreaded computation. This feature is beneficial for users who need to run their Matlab codes in parallel, something not previously possible. This timely addition to the Matlab capability greatly enhances our new multi-core cluster, atlas3. To find out more, read on in the full article.