On 13 May 2025, the Inno4scale project reached its conclusion with the final event, the Inno4scale Innovation Day, which was hosted at the Universitat Politècnica de Catalunya in Barcelona. The event convened researchers, innovators, and stakeholders from across Europe to reflect on key achievements and discuss the future of high-performance computing (HPC) algorithms as the field transitions into the exascale era.
The day commenced with introductory statements from the hosts and the project consortium team, subsequently succeeded by an inspirational address delivered by the project’s coordinator, Dr. Oriol Pineda. Subsequently, the Euro HPC Joint Undertaking project officer, Dr. Linda Gesenhaus, delivered a presentation in which she emphasised the critical role of European research in shaping the next generation of scalable algorithms and computing architectures, particularly as we move beyond exascale.
The scientific core of the event comprised three expert panel sessions, focusing on advances in scalable linear algebra, proof-of-concept applications, and AI and machine learning for scalable computing. In these discussions, representatives of the Inno4scale Innovation Studies presented their algorithmic developments, performance achievements, software demonstrators, and visions for future research and application.
In addition to the panels, a dedicated poster and networking session offered a forum for direct exchange between project teams and attendees. In conclusion of the day’s proceedings, the participants were invited to undertake a guided visit to the MareNostrum 5 supercomputer and the quantum computing research facilities at the Barcelona Supercomputing Center.
Panel 1: Advances in Scalable Linear Algebra
The inaugural panel of the day focused on the subject of algorithmic advancements in scalable linear algebra, a fundamental discipline within the domain of scientific computing. Four innovation studies — AceMG, aCG, AMCG, and CvolBal — presented novel preconditioners, advanced coarse grid strategies, and communication-avoiding methods, all tailored for exascale performance.
A series of speakers, including Dr. Hartwig Anzt (TUM), Dr. James Trotter (SIMULA), Dr. Henrik Rusche (WIKKI), and Dr. Cevdet Aykanat (Bilkent University), provided a reflection on the practical challenges encountered in the development of fundamentally novel algorithms. The emphasis was placed on the increasing significance of properties such as hierarchical scalability, reduced communication overhead, and resilience. The session further investigated the potential of hybrid algorithmic strategies to amalgamate the most effective concepts derived from various studies, with the objective of surmounting scaling limitations anticipated in million-core environments.
Panel 2: Proof-of-Concept Applications Targeting Exascale
The subsequent panel concentrated on the implementation of algorithmic innovation. Researchers behind the Ex3S, ScalaMIDA, and SCALE-TRACK studies demonstrated the integration of newly developed algorithms into prototype applications, with the aim of addressing real-world simulation challenges.
A discussion was held between Dr. Linus Seelinger (KIT), Dr. Sergey Lesnik (WIKKI), and Dr. Hendrik Nicolai (TU Darmstadt) on the following subjects: performance results, strategies for code optimisation, and future scalability expectations. In addition to the design of algorithms, the discourse emphasised obstacles encountered in the field of software engineering, including data locality, task parallelism, and the necessity of tooling for forthcoming exascale environments. Furthermore, the panelists expounded on strategies for the broader adoption of these technologies, encompassing the principles of open-source publication and active engagement with user communities.
Panel 3: AI and Machine Learning for Scalable Applications
The final session focused on AI-accelerated approaches from the ESPLAG, exaSIMPLE, LimitX, and XCALE innovation studies. The session, which was comprised of speakers from Universidade da Coruña, blueOASIS, FZ Jülich, and Aalto University, sought to establish a dialogue between classical numerical simulation and emerging AI methodologies.
The key discussion points that emerged included the question of algorithmic originality in a field that is already characterised by a plethora of existing methods, the issue of performance forecasting on future exascale systems, and the growing convergence of HPC and AI. The panel explored the transformative potential of AI-enhanced solvers and surrogate models, as well as the unique requirements these impose on extreme-scale infrastructure — from software stacks to data pipelines.
The panellists also addressed the themes of sustainability and longevity, discussing their strategies for promoting broader use of their software and plans for the next phase of research or commercialisation.
Collaboration, Poster Showcase, and MareNostrum 5 Visit
In the afternoon, all 22 innovation studies were presented in a dedicated networking and poster session, offering a forum for deeper exchange. The day concluded with a guided visit to the MareNostrum 5 supercomputer and the quantum computing facilities of the Barcelona Supercomputing Center. This visit can be seen as a symbolic step towards the future that these innovations aim to shape.
Looking Ahead
As the Inno4scale project approaches its conclusion, the Innovation Day functioned both as a celebration of its scientific outcomes and as the establishment of the foundations for ongoing collaboration. The discussions emphasised the significance of collaborative design between algorithm developers, application engineers and infrastructure providers in order to achieve the full potential of exascale computing.
It is imperative to remain attentive as these algorithmic advancements traverse the trajectory from proof-of-concept to tangible impact within the realms of science, engineering, and industry.