Kim

The PEPR NumPEx – which you coordinate with Jean-Yves Berthou (for Inria) and Jérôme Bobin (for the CEA) – is an integral part of the Exascale France Project itself in coordination with the European EuroHPC project. Can you tell us about the challenges of these projects and the particular role PEPR will play in this context?

Michel Daydé :

The EuroHPC program is a joint initiative between the European Union, European countries and private partners to develop a world-class supercomputer ecosystem in Europe by 2025. It will rely for this on two exascale computers, that is i.e. capable of performing 1 billion billion operations per second. And this, with the energy constraint not to exceed a consumption of 20 megawatts. The first machine will be in Germany, France responded to the second call for projects to host the second. To this end, the French community has structured itself around a national exascale project led by GENCI.

However, the impact of these supercomputers will entirely depend on the applications that will be using them. There is therefore a strong need to build an ecosystem of applications and people who will adapt these applications to new machines. Indeed, exascale computers represent a major architectural evolution with tens of thousands of graphics processing units (GPUs) which will massively accelerate calculations. This therefore requires modifying existing algorithms or even developing new ones. PEPR NumPEx responds to this specific need through interdisciplinary research bringing together mathematicians, computer scientists and researchers from different fields of application.

How does PEPR plan to address these challenges?

M.D :

The change in architecture of these ultra-powerful computers means that the entire software stack1 has to be adapted or created. A paradigm shift is taking place. In this sense, the skeleton of the PEPR is based on fundamental projects around the development of methods, algorithms, software and data processing tools adapted to exascale. There are also issues specific to the energy consumption of exascale, which are all the more significant today. In other words, it is necessary to deploy applications that consume the least possible energy to arrive at the solution of a given problem. All of this research will integrate demonstrators that will cover a representative number of major fields of application.

Through this PEPR, we therefore aim to develop a French software stack with new resolution methods and new tools on aspects of calculations, data processing, artificial intelligence, execution support and monitoring which could be adopted, as far as possible, at French and European levels. Our ambition is to set up a coherent and efficient software environment ranging from runtime support to applications.

How will the results be transferred and to which application areas?

M.D : During this work, we will identify concerns or needs common to various applications. This may concern, for example, algorithmic approaches common to several applications or storage devices with particular protocols. The idea is to implement cross-cutting solutions in several areas. A PEPR unit will be responsible for helping the application teams to integrate the innovations of the targeted projects into demonstrators and, more broadly, to train users in their use.

Manufacturers have already expressed their interest, notably Atos Bull and SiPearl, which are heavily involved in the EuroHPC program on the design and manufacture of future European processors and machines. Several centers of excellence also participate in this PEPR with their applications. We therefore have a good guarantee of transfer to the scientific and industrial communities. We also rely on the areas identified in the Exascale France Project report. This represents approximately 80 applications related to the sciences of the universe, high energy and particle physics, life sciences, energy, the industry of the future and fundamental research.

What will this race for power bring to our society?

M.D : High-performance computing is a discovery engine in research. It allows you to get closer to complex physical phenomena. It is useful for advancing knowledge on large-scale challenges: climate change, prediction of natural disasters, energy saving; but also on issues of societal resilience and industrial competitiveness.

The development of new materials, personalized medicine, drug design, etc., are all applications in which supercomputers will play a major role. By accelerating the calculations associated with critical areas and issues, we support business competitiveness and the sovereignty of our society.

Especially since high performance computing is currently at the heart of important geopolitical issues.

M.D : Indeed, in the HPC sector, exascale is the next step to take. It is the subject of major competition between the United States, Japan, China, and Europe. It is a real strategic competition linked to the aforementioned societal challenges to which it will make it possible to respond, as well as through its potential for sensitive applications such as for defense.
Furthermore, the Chips Act, which aims to rebuild a semiconductor industry in Europe, and the recent Covid-19 and energy crises have highlighted Europe’s dependencies. So many findings that reinforce the importance of coordination between the EuroHPC project and its national versions via the exascale plan and this PEPR.