Between the 30th of September and 1st of October 2025 the third edition of the EuroHPC User Days took place in Copenhagen, bringing together more than 300 in-person participants. The event was hosted by the EuroHPC Joint Undertaking in partnership with EuroCC Denmark and the Danish e-Infrastructure Consortium (DeiC), and was held in two of Copenhagen’s iconic venues: the Royal Danish Library’s “Black Diamond” and the Danish Architecture Center.
POP3 actively participated in the “Meet & Greet” networking session, engaging with other Centres of Excellence, hosting entities, and AI Factories. This dedicated networking format provided an effective platform to strengthen collaboration and exchange experiences among researchers, developers, and institutional stakeholders across the EuroHPC ecosystem.
POP3 was represented at the User Days by Marta Garcia-Gasulla (BSC) and Ricardo Nobre (INESC-ID). During the event, they disseminated information about the project by distributing POP3 flyers and stickers and by directly engaging with EuroHPC users to explain the range of performance assessment services offered by the POP Centre of Excellence.
The User Days provided a forum for community building: project teams exchanged best practices, shared lessons learned, and offered feedback on user support and operations.
Additionally, two POP3 articles were accepted for publication in the EuroHPC User Days proceedings. Valentin Seitz and Marta Garcia-Gasulla authored a paper presenting the results of a collaboration between POP3 and the Plasma-PEPSC CoE, titled "Production-scale Performance Analysis and Optimization of Vlasiator". The article details the performance assessment conducted on Vlasiator and shows how the insights obtained guided the subsequent optimization steps, improving both efficiency and scalability for a large production run.
A second paper, titled "Porting Epistasis Detection Methods to EuroHPC Supercomputers", was presented by Ricardo Nobre, Aleksandar Ilic, and Leonel Sousa. Building on state-of-the-art codes for high-order epistasis detection, the team leveraged access to two of Europe’s most powerful supercomputers, MeluXina and LUMI, to develop enhanced implementations that enable efficient multi-node execution across different hardware architectures. Since most of the original codes (three out of four) were developed in CUDA, part of the effort involved porting them to HIP. These porting activities resulted in the fastest epistasis detection codes available for AMD GPU-based accelerators.