12 Oct 2018 Almere - Too little too late? Some of the recent articles about EuroHPC compare the amount of money in EuroHPC JU, mostly conveniently rounded to 1 billion euro, and the time scale - exascale in 2022/2023 - to exascale numbers and timeframes from the USA, China, and Japan. Comparing numbers from different countries and continents is, of course, always difficult. But one thing that is often overlooked is the fact that EuroHPC JU is just the tip of the iceberg: most money and most efforts in HPC in Europe are spent at national and regional levels. Some countries, Germany and France, would be able to buy an exascale system on their own. Perhaps a few years later, perhaps even in the same time frame. As an example, it is often mentioned that Europe does not have any system in the Top 10, just discarding the Piz Daint at CSCS in Switzerland. Actually that there is no system in Germany in the Top 10 of the world's fastest supercomputers is accidental, due to non-overlapping funding cycles. As can be seen from the recent supercomputing acquisitions in Germany, it is indeed likely that a German based supercomputer will return to the Top 10 again.
Germany, to use that as an example of a country with a well developed HPC policy, already committed last year officially to a massive investment programme that would also bring them to exascale, although that in itself is not a goal. This was announced last year by a German government official, Prof. Lucas, during the ISC 2017 Conference. He also said that Germany preferred to work together on the high-end on a European scale, but if that would not work out, Germany would pursue its own HPC goals. Not many conference participants grasped the importance of that announcement.
This year, the details of the German HPC plans were presented at ISC 2018, again in Frankfurt, in a presentation by Michael Resch, Chair of the Gauss Centre for Supercomputing. The Gauss Centre is the collaboration of the three largest supercomputer centres in Germany. There is a German Smart Scaling Strategy in place with pre-exascale solutions in 2020 and exascale solutions in 2023. The latter date is the same as the projected date for EuroHPC, so it could be the same system, but that will be decided later. The main German supercomputing funding programme is called SIveGCS, another is called SPPEXA - indeed names that are difficult to remember. PetCGS runs from 2008 to 2019 and has a budget of 400 million euro. SiVeCGS runs from 2017 to 2025 with a 460 million euro budget. A third smaller project called inHPC with 15 million euro budget runs from 2017 to 2021. This is more about connectivity.
As said, Germany is just one country in Europe investing in HPC. And the high-end represented by the Gauss Centre is just the tip of the iceberg in Germany: several individual states have their own HPC programmes as well.
Gauss Centre Chair Michael Resch also explained this in the interview we had with him recently.
Similar stories can be told about other European countries: EuroHPC is a tip of the European HPC iceberg.