The recent ISC24 conference in Hamburg showcased LUMI and other leadership-class supercomputers co-funded by the EuroHPC Joint Undertaking (JU), including three of the 10 highest-ranking Top500 systems, but some other new JU developments got less attention. HPCwire asked Steve Conway, senior analyst at Intersect360 Research, to interview Anders Dam Jensen, executive director of the EuroHPC JU, about these developments.
HPCwire: First, I understand the European Commission has proposed that the JU develop and operate so-called AI factories, with some special emphasis on providing access and training for SMEs. That emphasis is important, because there’s serious concern worldwide that the big AI players may try to make it difficult for innovative SMEs to succeed in the AI market. Can you talk about the AI factories proposal and its status?
Anders Jensen: Indeed, it is now official: the EuroHPC JU has a new AI objective! The new regulation, which amends the EuroHPC JU founding regulation, has been officially adopted and published and will come into force in the coming days, giving us a new mandate to develop and operate AI factories. To give you some context, these AI factories have been envisioned as comprehensive open AI ecosystems centered around EuroHPC supercomputing facilities. This concept goes beyond mere infrastructure, as the AI factories are designed to cater to the needs of European AI model developers and support the growth of a highly competitive and innovative AI ecosystem in Europe.
The EuroHPC supercomputers have already provided an excellent platform for AI applications. What the amended regulation will add is the ability for the EuroHPC JU to better equip European researchers, start-ups and SMEs with a comprehensive set of powerful tools and services to develop cutting-edge AI projects. With this new pillar of activities now officially enshrined into our regulation, the JU will be able to acquire and operate AI-optimized supercomputers. This will involve procuring new machines as well as upgrading existing EuroHPC supercomputers with enhanced AI capabilities. Developing open AI ecosystems around these supercomputing facilities will also require investments in associated data centers, hyperconnectivity and AI-oriented supercomputing services to attract talent and support users.
HPCwire: What is the anticipated user community for the new AI-optimized user community?
Just as the other EuroHPC supercomputers, the AI factories will serve a wide range of European users, regardless of their location in Europe, encompassing the scientific community, industry, and the public sector. But as you mentioned, specific attention will be given to startups and SMEs to facilitate their access to this infrastructure and attract investment in the AI sector as they are a driving force of the European economy and play a pivotal role in leading innovation.
As always, skills development is a key priority for us and will also be a cornerstone of this new venture, as supercomputing centres applying to host AI factories are likely to be required to provide a comprehensive skills plan, offering a variety of training courses and activities tailored to the diverse needs of potential users.
These guidelines are currently being discussed by the EuroHPC JU Governing Board, and once finalized, our challenge will be to implement them and make sure that the deployment of AI factories across Europe will underpin European leadership in AI. The first step for us will be to prepare and launch the call for expression of interest to host AI factories to start selecting hosting sites across Europe!
HPCwire: Can you comment on new members of the EuroHPC JU?
Anders Jensen: You might have heard that we recently welcomed the United Kingdom (UK) as the latest Participating State to join the EuroHPC initiative! The EuroHPC Joint Undertaking now consists of 35 participating countries that have decided to join forces with the European Union and private partners to lead the way in European supercomputing.
What I find particularly inspiring about this cooperation is that among these 35 participating countries, several are not members of the European Union. Nevertheless, these countries have chosen to join this ambitious joint initiative to develop a world-class and sovereign supercomputing ecosystem in Europe. This includes Iceland, Israel, Montenegro, North Macedonia, Norway, Serbia, Turkey and of course the UK.
We are pleased to welcome the UK to our initiative, collaborating with a country with such a longstanding expertise in supercomputing is great news for Europe, and we will all benefit from this cooperation. This new membership also sends a strong message: European supercomputing is attractive and poised to continue leading the way into the future.
HPCwire: What about the relationship with PRACE (Partnership for Advanced Computing in Europe)? How is that evolving?
Anders Jensen: EuroHPC JU still collaborates closely with PRACE on various topics, leveraging its expertise. For example, PRACE participates in several EuroHPC JU-funded projects such as CASTIEL 2, which coordinates the EuroCC network of HPC National Competence Centres, and Inno4scale, which supports the development of innovative algorithms to fully exploit upcoming exascale and post-exascale supercomputing. I sincerely hope that this collaboration will continue on other topics to the benefit of European scientists and users.
HPCwire: The relationship with Japan also seems to be growing through the HANAMI project. How will that work?
Anders Jensen: Indeed, you are right! As part of the EuroHPC JU’s international cooperation pillar of activities, the HANAMI project has recently begun and will continue for three years. This project is all about fostering collaboration in HPC between European and Japanese research teams, building upon and extending existing scientific partnerships. European and Japanese research centres will work together to develop and optimize applications in various scientific fields focusing on climate simulation, materials research and biomedicine. Practically, this means that in addition to exchanging information and expertise, select European and Japanese scientists will be able to use each other’s supercomputers. European users will have the opportunity to run simulations on Japanese machines including Fugaku, while Japanese users will use EuroHPC systems like LUMI, Leonardo and MareNostrum 5.
These systems are all at the forefront of HPC technology and are all currently ranked among the the world’s top 10 most powerful supercomputers. What will be particularly interesting for the users is the opportunity to test and gain an understanding of different types of computing architectures. Unlike the EuroHPC JU supercomputers, Fugaku does not rely on accelerators, so this access will allow European researchers to scale and adapt their software to test and benchmark on a completely different HPC system. Meanwhile, Japanese researchers will be able to explore the accelerator-based technologies available in EuroHPC systems.
The project will enable European and Japanese scientists to port their existing code to each other’s very different machines, testing performance and making sure applications will run on alternative computing architectures. By working on each other’s machines, scientists from both regions will make the most of the combined platforms to advance their science jointly.
Such a partnership is also strategic for the future, as it will provide significant insights for future investment decisions and help with designing the next generation of machines in Europe.
HPCwire: What role will the new Research & Innovation Advisory Group play?
Anders Jensen: As the mandate of the previous Research & Innovation Advisory Group (RIAG) recently ended, new RIAG members have been appointed to take over and continue the group’s tasks. The group consists of representatives from academia and industry, as well as HPC users and technology suppliers, and its role is to provide independent advice to the Governing Board of the EuroHPC JU for the duration of its two-year mandate. Specifically, RIAG is responsible for drafting and regularly updating the EuroHPC JU’s multiannual strategic research and innovation agenda, which identifies research and innovation priorities to support the development of an integrated HPC, quantum computing, and going forward AI ecosystem in Europe. The group also provides advice on training and education priorities in order to address key competencies and the skills gap in HPC and quantum computing. Additionally, RIAG organizes public consultations open to all public and private stakeholders interested in HPC and quantum computing to inform them and collect feedback.
RIAG is one of the two advisory groups which together make up the industrial and scientific advisory board for the EuroHPC Joint Undertaking. The board also includes the Infrastructure Advisory Group (INFRAG). Just few days ago, the new INFRAG was officially established after the previous INFRAG’s two-year mandate also came to an end. While RIAG is focused on research and innovation activities, INFRAG’s role is to advise the Governing Board on all matters relating to the infrastructure acquired and operated by the EuroHPC JU. To this end, INFRAG’s tasks are to draft and regularly update the multiannual strategic agenda related to supercomputer acquisition, capability building and widening activities. INFRAG also issues recommendations on the federation and interconnection of the EuroHPC infrastructure.
HPCwire: Another new initiative is called Epicure, which I understand will provide support services for awarded projects. What kind of support services?
Anders Jensen: In the early years of our initiative, our primary objective was the procurement and installation of a world-class HPC infrastructure across Europe. Now that this infrastructure is up and running successfully, we must focus on helping European users make the most of these advanced technologies. Until now, the support provided to users gaining access to EuroHPC JU systems through our access calls has been mainly focused on helpdesk and short-term support for daily operational issues. The responsibility of this support lay with the supercomputing centres, which host and operate the EuroHPC JU supercomputers. With the recently launched EPICURE project, our objective is to enhance user support at European level and assist users with advanced technical needs.
Considering that our systems use different architectures, we aim for instance to help our users move from one system to another easily and seamlessly. To reach this objective, a central component of the project is to develop a European HPC Application Support portal, serving as a single point of contact for users from the public and private sectors, including SMEs. This portal will provide comprehensive information on EuroHPC systems, their architectures, access mechanisms, and the range of available support services. These services will include code enablement and scaling, performance analysis and benchmarking, code refactoring and code optimization.
Additionally, EPICURE will organize training activities, hackathons, events and workshops to share knowledge on porting, optimization, parallelization, and GPU programming. All of these activities are carried out in close collaboration with the different Competence Centres and Centres of Excellence, drawing on the expertise of all partners in HPC operations and support.
By establishing and operating a distributed but coordinated European-wide HPC application support service, the project will encourage the best possible uptake of the EuroHPC systems by European scientists and researchers.
HPCwire: Finally, can you talk about the call for industrial supercomputers that was announced in February? I believe the budget is about 12 million euros.
Anders Jensen: At this point, there is limited additional information I can provide.
As you know, the call for selecting entities to host our first industrial-grade supercomputers was launched in February and closed at the end of April. Over the last week, our team has been busy evaluating the submitted proposals and the process is nearing completion, so we will be able to share more details soon. This new system, tailored specifically to the needs of industrial users will be another significant enhancement to our fleet and will play a crucial role in further boosting the innovation potential of enterprises in Europe.
HPCwire: Anything else HPCwire readers should know about?
Anders Jensen: In recent weeks, we have made significant progress in expanding the EuroHPC infrastructure. We have announced the name of our second exascale supercomputer, Alice Recoque, which will be located in France. Named in honor of the pioneering French computer scientist, this new machine is set to push the boundaries of HPC and will represent another significant technological milestone for Europe. Now that the hosting agreement with GENCI, which leads the Jules Verne consortium, has been signed, the procurement process for this system will begin shortly.
In parallel, we signed the hosting agreement for Arrhenius, a new EuroHPC mid-range supercomputer to be located in Sweden, and we launched the call to select a vendor for the acquisition and installation of Daedalus, a mid-range system located in Greece and capable of executing more than 60 petaflops.
Looking ahead, I am confident that the implementation of the new AI regulation will keep us very busy over the coming months!
On another note, we have several events to look forward to over the next few months. We are currently organizing the second EuroHPC User Day, which will take place on 22-23 October in Amsterdam with the support of EuroCC Netherlands. The event will once again bring together the EuroHPC User community, allowing them to showcase EuroHPC projects that have leveraged European supercomputing resources to advance science and foster innovation. The event will also provide an opportunity to reflect on how best to support current users, while engaging with potential new users. We are looking forward to SC 2024 in Atlanta this November, where we will continue to raise awareness about our activities and opportunities, and
in parallel we have already begun the preparations of the next EuroHPC Summit 2025. Scheduled for 18-20 March in Krakow in Poland. Be sure to mark your calendars!