LUMI’s second pilot phase in full swing

The second pilot phase aims to test the scalability of the GPU partition and generate workloads on the GPUs, particularly to stress test the storage systems for stability testing. Furthermore, the aim is to provide early access to LUMI to obtain feedback from the pilot users before the launch of LUMI’s regular operations.

28 selected projects for the pilot phase were chosen from the LUMI consortium countries and are presenting various disciplines:

• Belgium: Giovanni Lapenta (KU Leuven, Department of Mathematics, Center for mathematical Plasma-Astrophysics), Lif – Lumi data Inference Framework for astrophysical applications
• Belgium: Tim Lebailly (KU Leuven, Department of Electrical Engineering), Spatial-Aware Self-Supervised Learning
• Czech Republic: Sergiu Arapan, Dominik Legut (Technical University of Ostrava), Urszula Wdowik (Academic Computer Centre Cyfronet AGH), Computational search for novel two dimensional thermoelectric materials
• Czech Republic: David Číž (Technical University of Ostrava), Benchmarking TensorFlow on AMD GPUs
• Czech Republic: Oldřich Plchot (Brno University of Technology, Faculty of Information Technology), Training transformer based neural models for speech applications
• Denmark: Troels Haugbølle (Niels Bohr Institute, University of Copenhagen), How are stellar systems born?
• Denmark: Tejs Vegge and Adam Maximilian Wilson (Technical University of Denmark), BIG-MAP: Battery Interface Genome – Materials Acceleration Platform
• Denmark: Benjamin Jäger (University of Southern Denmark, Department of Mathematics and Computer Science), Stress-testing LUMI with Lattice QCD
• Estonia: Oriol Corcoll, Mykyta Baliesnyi, Daniel Majoral, Tarun Khajuria and Aqeel Labash (University of Tartu), Deep Reinforcement Learning for Artificial Agents
• Estonia: Mark Fišel, Andre Tättar, Liisa Rätsep, Lisa Korotkova and Taido Purason (University of Tartu, Natural Language Processing), Research of Multilingual Modularized Neural Machine Translations
• Estonia: Kairit Sirts, Alfred Saidlo (University of Tartu, Natural Language Processing), Estonian BART model training
• Finland: Sampo Pyysalo (University of Turku), Deep Generative Language Modeling for Finnish
• Finland: Maarit Käpylä (Aalto University), VerIfying Small-Scale dynamo actIon in the Sun
• Finland: Pekka Ruusuvuori (University of Turku), AI-PRO
• Iceland: Eric Michael Sumner (University of Iceland), ACUTE (Acoustic and Tactile Engineering laboratory)
• Iceland: Dirk Norbert Helmrich (University of Iceland), Visualization as Data Source for Machine Learning
• Norway: Mats Carlsson (University of Oslo, Institute of Theoretical Astrophysics), Solar Atmospheric Modelling
• Norway: Stephan Oepen (University of Oslo), Very Large Language Models in the Nordics (VLLMN)
• Norway: Børge Arntsen (Norwegian University of Science and Technology), Deep seismic imaging
• Poland: Gabriel Wlazłowski (Warsaw University of Technology), Turbulent dynamics in superfluid Fermi systems
• Poland: Piotr Korcyl (Jagiellonian University), Calculating hadron structure functions at small Bjorken-x using GPGPU processors
• Poland: Michał Dzikowski (ICM University of Warsaw), High resolution and high performance simulations of porous media flows using lattice Boltzmann method
• Sweden: Thorsten Mauritzen (Stockholm University, Department of Meteorology), William Sawyer (CSCS), Luis Kornbleuh (Max Planck Institute for Meteorology) and Qiang Li (ENCCS), Next generation global climate models
• Sweden: Erik Lindahl (SciLifeLab), Mark Abraham and Artem Zhmurov (ENCCS), Lipid modulation of ion channel gating – showcasing the GROMACS AMD GPU port
• Sweden: Philipp Schlatter (KTH Royal Institute of Technology, Department of Mechanics), XLarge-scale simulations in Stability, Transition, Turbulence and Control
• Switzerland: Fiorina Ciorba (University of Basel), Lucio Mayer (University of Zurich) and Rubén Cabezón (University of Basel): Simulating Subsonic turbulence with SPH-EXA2 on LUMI
• Switzerland: Juerg Hutter (University of Zurich), Water-TiO2 interface structure from double-hybrid density functional
• Switzerland: Marnik Bercx, Nicola Marzari (EPFL), AiiDA “hero run” using SIRIUS-enabled QE

Full operations starting soon

After the GPU pilot phase, LUMI will be almost fully operational and general availability is foreseen to start in December 2022.

Please see the Get started section of the LUMI website for more information how to access LUMI and follow LUMI’s website and social media channels for more information about upcoming calls to access the system.

Read more about one of the second pilot phase use cases and watch a video interview in the following article: Language technology is essential for the survival of small languages – researchers using supercomputers to develop Finnish language models.