23 Mar 2017 Rome - The exascale race is not over, yet. Seven European countries signed a declaration this morning at an HPC session at the Digital Day 2017 in Rome, aiming for the acquisition of two world-class pre-exascale supercomputers preferably starting on 2019-2020, and two world-class full exascale supercomputers preferably starting on 2022-2023. At least one of the two exascale computers should be based on high-quality competitive European technology produced in a co-design approach and its integration in at least one of the two exascale supercomputers. The European Commission reaffirmed Junker's goal of having at least one system amongst the first three in the TOP500 of the world's fastest supercomputers in 2022.
However, how this goal should be reached is not yet clear. There is not yet a plan, nor a roadmap. The seven countries, Germany, Portugal, France, Spain, Italy, Luxembourg and the Netherlands, agreed to the work on an implementation roadmap by the end of 2017. That seems a little late to have an all European exascale supercomputer already 5 years later. One would expect at least two cycles of European technology supercomputers to get there. Also the funding is not yet clear. How much will the seven countries invest? The funding available from the European Union is more or less known from the Horizon 2020 budget: 700 million euro.
The ETP4HPC group of companies and research organisations on HPC in Europe has just started to prepare its new Strategic Research Agenda. This will certainly contribute to the planned roadmap. Also European technology companies will invest in HPC technologies. It was encouraging that two of them, Atos/Bull and Eurotech, were present in the Rome session.
The seven countries say they will invite other member countries and associated countries to join. If new countries also want to have an influence on the roadmap, the schedule could become even tighter. Some countries with important technologies and communities are for instance the UK where the Edinburgh Parallel Computing Centre, in Scotland, is doing exascale research and development, and where in Cambridge there is an important hardware component industry around, the now Japanese, ARM. In the Nordic countries, most notably Sweden, there is also strong exascale knowledge. CSCS from Switzerland was present at the Rome meeting, but Switzerland has not committed yet to the declaration. Countries from middle-east Europe are also still missing.
There is another project, the Important Project of Common European Interest on "High Performance Computing and Big Data enabled Applications" (IPCEI on HPC and BDA) that is still under preparation by a subset of the seven countries. Earlier funding estimates for this IPCEI project talked about 3 billion euro costs and planned to be ready by the end of last year. It is a bit worrying that there now seem to be two European-wide initiatives.
Several panelists in Rome pointed out Europe is not lagging behind in exascale software development, something that you cannot find back in a TOP500 or other list.
So, will Europe develop an exascale supercomputer based on European technology? Probably yes. Will it be in the 2022-2023 timeframe? Perhaps. Will Europe be the first to have a real exascale supercomputer? Most likely not. But then the exascale systems will be there to support researchers and companies. And as long as they do that it will be fine.