The German Council of Science and Humanities (Wissenschaftsrat) is an organisation that provides advice to the German Federal Government and the State (Länder) Governments on the structure and development of higher education and research. It does not distribute or control funding itself.
According to the Wissenschaftsrat, the HPC pyramid model is still valid. In Germany, there are several centres providing Tier 1 supercomputing. This includes for instance the large supercomputing centres in Jülich, Stuttgart and Munich. They provide services on a national level. On a European level, PRACE provides Tier 0 supercomputing access. At the level of Tier 2, there are several regional supercomputing centres, for instance in Hannover and Berlin. They still operate TOP500 class supercomputers but for regions in Germany. Some universities provide HPC services for their own local researchers.
The Wissenschaftsrat considers this three-Tier approach still valid for Germany. They propose however that the level 2 supercomputer centres more closely cooperate together wit the Gauss Centre in a "Verbund von Kompetenzzentren für Nationales Hoch- und Höchstleistungsrechnen (NHR)" - a collaboration network for High-Performance Computing.
The NHR should provide easy access for all German researchers, and all kinds of disciplines, from Digital Humanities, to fluid dynamics and neuro science. Not only hardware should be coordinated in such a way that different architectures can be made available to researchers, but also experiences in energy efficiency should be shared as should be software maintenance and training. Energy efficiency for supercomputers is important because currently the total energy bill in 5 years equals the investment cost of the average German supercomputer. Because the energy efficiency of systems in operations/watt increases faster than the performance measures in operation/s per second, it may be useful to consider replacing supercomputers after about 3 years.
The Wissenschaftsrat thinks a stable funding for a longer period of about 10 years of NHR supercomputing is necessary. Funding should come mainly from government funding at all levels. A pay-per-use system is considered inefficient, as it will just lead to more bureaucracy. However, the Wissenschaftrat recommends to make it easier for industry to use the NHR supercomputers.