The report, written by Prof. J. Nievergelt, Dr. D. Maric, Dr. Ambros Marzetta and Dr. F. Zaucker says that adecade ago, perhaps, the mission of a National HPC centre was clear: provide more computing power than any local computer centre could offer, along with the necessary operating and consulting services. Hence, the mission of CSCS has been providing high quality service to a relatively small group of users in a few selected areas of science.
With the supremacy of the traditional supercomputers in terms of computing power vanished, exploratory systems made up of hundreds to thousands of commodity microprocessors easily matched the raw computing power of traditional supercomputers at a fraction of the cost.
As a consequence, the traditional parallel vector processors that dominated supercomputing in the 80s have steadily been losing market share, the report says.
The former role of traditional vector processors is being assumed by parallel computers , and by networks of workstations or personal computers.
Instead of using a few expensive custom-made components, one uses lots of powerful but cheap commodity components. Thus, it will be driven by economic factors rather than technology. The number of supercomputers sold each year is on the order of a hundred. In contrast, hundreds of millions of microprocessors for PCs, cellular phones and game stations are produced; millions of PCs and workstations based on these commodity items are produced.
The report concludes that in order to keep up-to-date on HPC system developments across the entire range of hardware and software, CSCS must strengthen its technology competence: collect information and experiment with test systems of different architecture in order to learn how to operate them efficiently; and help users adapt to the changing environment.
The full report is available from CSCS web site .