Evolution in High Performance Computing


Dr. Khaled Alhussan

Dr. Khaled Alhussan
National Program Aeronautical Technology
Riyadh, Saudia Arabia


There are many factors that constrain in the scalability of the applications of HPC systems when going to large core count, both within the compute node and across the whole system, among which most important are size, speed and scale. We have been using various architectures of HPC’s in many core research areas in KACST; such as Mathematical Modeling, Computational Fluid Dynamics, Numerical Analysis etc, for last 15 years i.e. from different vendors such as HP, DELL, IBM, SGI, etc. Every architecture had a certain significant role for certain problems due to the evolution in the hardware and software.

 Many problems that are interesting to scientists and engineers can’t be fitted on a regular machine due to the limitation of resources (Size).  Major simulations require a lot of computing time: if performed on a regular machine, the time frame will go from days to even weeks but only a few hours are needed on a supercomputer (Speed).There are many questions from the researchers on the scalability of the system such as; can I make the run twice as fast with double resources? Or, can I run twice workload with double resources within same time? , therefore the problems are scaled to a certain level according to the available resources. At present the aim of most HPC’s is to gain highest efficiency in performance with the lowest energy consumption (Economically efficient systems).

 The Saudi supercomputer “SANAM” built by KACST and German institutes in 2012 has ranked second in the worldwide list of the most energy-efficient computers, according to “Green500” ranking in November, 2012. SANAM is based on an enhanced technology of the Frankfurt supercomputer LOEWE-CSC, which was the most energy-efficient multipurpose supercomputer in Europe. In terms of computing speed, “SANAM” is about 40 percent faster than the German supercomputer LOEWE-CSC, but requires merely one third of the power per computing operation. Mainly serve for applications related to seismic, aerospace, bioinformatics, weather and numerical simulations etc.