EU Outlines Plan to Develop Home-Grown Exascale Supercomputers

April 4, 2017

By: Michael Feldman

Ministers from seven European countries have signed on to a plan to develop an exascale capability based on technology developed within the EU member states. The goal is to bring up two pre-exascale supercomputers by 2020 and two full exascale systems no later than 2023.

The initial seven countries to support this work are France, Germany, Italy, Luxembourg, Netherlands, Portugal and Spain. Other member states have been invited to join. According to the press release, the resulting supercomputers will be “available across the EU for scientific communities, industry and the public sector, no matter where the users are located.”

As pointed out in the announcement, Europe currently produces about 5 percent of HPC products and services worldwide, but consumes one third of them. That position leaves them vulnerable to market disruptions and political forces that cannot be controlled within the borders of the EU. Further, it restricts the type of technological innovation and diversity that can be accomplished with domestically-produced HPC.

More to the point, the EU wants to position itself a distinct HPC leader on par with that of the US, China, and Japan. The rationale behind this is that these technologies are increasingly seen as catalysts for key industries like manufacturing, healthcare, transportation, and energy, as well as an enabler for things like cloud services, machine learning and the internet of things (IoT).

The focus on indigenous HPC was initially proposed under the European Cloud Initiative in 2016. The strategy outlined there was to jumpstart European production of HPC componentry, specifically low-power processors. The goal was to use those chips to furnish an open science cloud, as well as to develop extreme-scale systems. In the EuroHPC initiative, these domestically produced processors will be used power two European exascale supercomputers, which will be installed in 2022 and go into production in 2023.

The strategy is elaborated in a 2016 working document published by the European Commission, titled “Implementation of the Action Plan for the European High-Performance Computing Strategy.” The document rightly points out that no single member state has the resources to compete with the US or China in HPC technology development. As a consequence, the EU members must act as a single entity to compete on the global stage – and must do so on both the supply and demand side. The authors state the problem thusly:

“The underdevelopment of the HPC ecosystem has a double negative effect: there is not enough HPC capacity to cater for the demand by science and industry; and there is also not enough European supply industry that designs and builds HPC systems. As a consequence, research may need to relocate outside Europe, access to advanced systems is insufficient, and software and tools development, especially those linked to hardware, is disadvantaged. Europe of course continues to be an appealing sales market for HPC system vendors from 3rd countries. European HPC software and services are adapted and tuned to these HPC systems and in turn benefit their (non-EU) vendors.”

The subtext here is that Europe doesn’t want to be left behind as China emerges as the next great HPC superpower. In fact, China’s strategy to develop indigenous HPC technology is being taken up by the EuroHPC model. In the European case, the most likely initial target of indigenous technology will be ARM-based processors and systems based on those processors. Unfortunately for Europe, it will have to take this on without the help of the UK, which is exiting the EU and appears to be intent on keeping its HPC investments within its borders.

A likely vendor to help bring this all to fruition is Bull (Atos), which is the only major HPC OEM on the continent that is not a US or Japanese subsidiary. Conveniently, Bull also has its own custom interconnect in BXI and has dabbled in building ARM-powered supercomputers. The missing piece is ARM processor IP, but that can be developed by a relatively small engineering team, or even purchased if need be.

According to the EuroHPC announcement, the next step will be to prepare a roadmap for the project with the 2023 exascale system as the endpoint. Developing domestic HPC processors is going to be a big lift for the EU, given its decades-long history of relying on US-based chips, and to a large extent, on US-based systems as well. But if China and Japan can do it, both of which have an economy smaller than that of the EU, it should certainly be within Europe’s reach. As the HPC action plan states: “With a differentiated strategy and sufficient investment and political will, Europe can be a global player in HPC”