EPICURE Hackathon: Code Optimisation for HPC and AI Workloads

 

7–9 October 2026 | ACC Cyfronet AGH, Kraków, Poland

 

 

 

 

Following the success of the Porto edition, the EPICURE Hackathon returns for its fifth edition. Building on EPICURE’s experience supporting around 240 projects across Europe, this three-day, hands-on event brings together developers, researchers, and HPC and AI users to tackle real performance challenges on Europe’s most powerful supercomputers.

 

This edition broadens its scope to cover both traditional scientific computing and AI workloads, with a strong focus on code optimisation, profiling, and scaling across CPUs and GPUs. Whether you want to speed up your simulation code or make your machine learning pipeline run efficiently at scale, this hackathon is for you.

 

During this hackathon, participants will be able to use JUPITER, Europe’s first exascale supercomputer, and Helios, the Polish supercomputer.

 

Registrations for the hackathon are now closed.

 

 

 

 

From performance challenge to practical progress

 

Modern scientific and AI applications increasingly depend on complex computing environments. Achieving good performance requires more than access to powerful hardware: applications must be carefully profiled, adapted and optimised for the architectures on which they run.

 

During the EPICURE Hackathon, teams will have the opportunity to:

 

 

• -Use JUPITER, the fastest and first exascale supercomputer in Europe;
• -Use Helios, a new light for innovation in Polish science and economy;
• -Analyse the performance of their applications and workflows;
• -Identify computational, memory, communication and data movement bottlenecks;
• -Improve CPU, GPU and multi-node performance;
• -Optimise AI training and inference workloads;
• -Explore parallelisation, profiling and scaling techniques;
• -Receive targeted guidance from experienced HPC and AI specialists;
• -Exchange knowledge with other teams from across the European computing community.

 

The emphasis is on practical work, measurable progress and solutions that teams can continue developing after the event.

 

 

 

Bring your own code

 

 

Bring your own application or AI workload and work on the challenges that matter to your team.

 

Together with EPICURE mentors, participants will define clear technical objectives and work on topics such as profiling, parallelisation, GPU acceleration, memory optimisation, data pipelines, distributed training and application scaling.

 

Projects may include:

 

 

• -Scientific and engineering applications;
• -Simulation and modelling codes;
• -Machine learning and deep learning workloads;
• -AI training and inference pipelines;
• -Hybrid HPC and AI workflows;
• -Data-intensive scientific applications;
• -Code porting and performance optimisation.

 

 

 

Mentoring starts before the hackathon

 

 

In this edition, collaboration begins before the main event. On October 1, teams will join an online kick-off session. This online meeting will give participants the opportunity to:

 

• -Meet their assigned mentors;
• -Introduce their projects and technical challenges;
• -Connect with fellow participants;
• -Discuss objectives and expected outcomes;
• -Share relevant technical information;
• -Identify initial bottlenecks and priorities;
• -Agree on the first steps before the hands-on sessions begin.

 

Starting the mentoring process earlier will allow teams and mentors to arrive at the hackathon with a shared understanding of the project and a clearer technical plan, making the three event days more focused and productive.

 

 

Agenda

 

 

All times are in Central European Summer Time (CEST).

 

 

 

 

Keynote speakers and mentors

 

 

 

 

Andrzej Dorobisz

 

Computer Scientist with strong analytical skills, paying attention to detail. Over 10 years of experience in the HPC field. An expert in PRACE and EPICURE user support projects. Proficient in software optimisation, parallelisation and debugging.

 

 

 

Bartosz Soból

 

Bartosz Soból is an HPC software specialist at ACK Cyfronet AGH. Involved in the development of services for interactive access to Cyfronet’s HPC resources, such as the JupyterHub-based platform and virtual desktop environment. He’s also engaged in user support via the PLGrid Helpdesk and CryoEM software integration in HPC environments. His professional interests include performance optimisation, parallelisation, and GPU acceleration of custom scientific software. He’s also pursuing a PhD in Technical Computer Science at Jagiellonian University, where he conducts research on online data processing in HEP experiments.

 

 

João Paulo [Online Keynote Speaker]

 

João Paulo is an Assistant Professor at the University of Minho and a Senior Researcher at INESC TEC (HASLab), where he leads the distributed storage systems research team (dsr-haslab.github.io).

His work centers on the scalability, performance, security, and dependability of storage and operating systems in large-scale distributed environments — with a particular focus on the data management challenges facing HPC centers and Cloud Computing infrastructures.

He has published in leading systems and storage venues, including USENIX FAST, EuroSys, VLDB, ACM Transactions on Storage, IEEE Transactions on Computers, CCGrid, and SRDS, and currently coordinates national and international projects on secure and efficient data storage.

 

Beyond Compute: Rethinking Storage for HPC

 

While the computational power of HPC infrastructures is rapidly increasing and is expected to improve further in the coming years, there has been significantly less research and development on storage solutions to accompany this evolution. In fact, storage performance (i.e., the access and retrieval of data for applications) is one of the major bottlenecks to efficiently supporting a new generation of data-centric jobs in HPC — from AI and machine learning workloads to broader data analytics and more traditional HPC jobs.

This talk will address some of the main data storage challenges brought by these data-centric jobs running at HPC infrastructures. In particular, it will discuss ongoing research from the Distributed Storage Research team at INESC TEC (https://dsr-haslab.github.io/) on i) storage QoS mechanisms for effectively sharing storage resources among the thousands of data-intensive jobs running on large-scale supercomputers, and ii) better leveraging the resources available at HPC infrastructures, from local node disks to parallel file systems. These topics will be grounded in practical examples and collaborative projects with CNCA, TACC, and AIST, drawing on experience with their corresponding supercomputing infrastructures.

 

 

 

Klemens Noga

 

Klemens Noga is a senior HPC specialist and head of training at ACC Cyfronet AGH and holds a PhD in computational and theoretical chemistry with 20+ years of experience running various workloads on HPC systems and 17 years in teaching others how to conduct computations effectively. His work focuses on coordinating the training team’s efforts and providing high-quality training for HPC users, as well as supporting users, especially in computational chemistry. He is also a Software Carpentry and Data Carpentry-certified instructor.

 

 

Laura Bellentani [Keynote Speaker]

 

Originally trained as a physicist, Laura Bellentani specialised in High-Performance Computing during her PhD in Physics and Nanosciences at the University of Modena and Reggio Emilia. Since 2021, she has been a member of the High-Level Support Team at CINECA, where her work focuses on porting applications to GPUs. She maintains a focus on balancing performance with code maintainability and portability, viewing profiling as an indispensable tool for achieving effective and sustainable optimisations.

 

Profiling applications assisted by Nvidia Nsight Systems

 

Optimising modern hybrid HPC applications means untangling complex interactions between CPUs and GPUs. This educational session provides participants with a structured, demonstration-driven framework for performance engineering using NVIDIA Nsight Systems. Key focus areas include data locality, stream concurrency, and GPU-aware MPI communication, equipping participants with the diagnostic skills needed to accelerate their hackathon projects.

 

 

Maciej Czuchry

 

 Over 15 years of experience in HPC software administration using Linux in the scope of building (compilation), customisation, containerization, management and testing software on the largest Polish supercomputers. Active support for PLGrid infrastructure users through the Helpdesk system in the scope of HPC computing. Organisation and conducting training on topics related to HPC computing: Linux system, SLURM queue system, computation efficiency, network filesystems usage (Lustre), virtual environments in Python and Singularity/Apptainer containers.

 

 

 

Maria Guadalupe Barrios Sazo [Keynote Speaker]

 

Maria Guadalupe Barrios Sazo holds a PhD in physics, with a research background in computational astrophysics and high-performance computing. She is currently part of the Application Support group at JSC, and previously worked as a Research Software Engineer at JSC and the University of Oslo. Her PhD research at Stony Brook University included contributions tothe Castro code and simulations of white dwarf mergers on leadership-class supercomputers.

 

Working with JUPITER: A Practical Onboarding for HPC and AI Users

 

This short talk gives hackathon participants a practical orientation to GH200-based systems. We cover the main building blocks of Grace–Hopper architecture, walk through the file-system spaces on JUPITER, and introduce LLview as a lightweight way to check how jobs are using resources. The aim is to give teams a practical starting point and a clearer picture of the working environment.

 

 

Why join?

 

 

 

 

Who should apply?

 

 

The EPICURE Hackathon is aimed at:

 

 

Participants should have an existing code, model or workflow and a clearly identifiable performance, portability or scalability challenge.

 

Applications from different scientific disciplines, technology areas and levels of HPC experience are welcome.

 

 

 

A hybrid and collaborative event

 

 

The hackathon will take place at ACC Cyfronet AGH in Kraków, Poland.

 

Both onsite and remote participants will be part of the same collaborative programme, including technical sessions, hands-on work, mentor interactions, team updates and final presentations.