Results of pilot projects on the LUMI supercomputer (LUMI-C and LUMI-G) from the Czech Republic
We would like to thank all applicants for participating in the LUMI Supercomputer Pilot Projects and for submitting their applications.
The aim of this call was to generate a stress-testing load on the newly installed CPU partition of the LUMI supercomputer while satisfying real-world needs and in particular to stress test the storage systems for the purpose of stability testing. Further, we aimed to provide early access to the LUMI supercomputer to obtain feedback before the launch of regular operation. The availability and technical readiness of the system (e.g. installed software) were not guaranteed and unexpected maintenance breaks may occur during this pilot phase.
The aim of this 1st call was to provide pilot access to the newly installed GPU part (LUMI-G) of the LUMI supercomputer in order to get feedback before the start of proper operation. The availability and technical readiness of the system (e.g. installed software) are not guaranteed at this point in time and unexpected downtime for maintenance or reconfiguration of the system may occur during this pilot phase.
Two projects for LUMI-C and two projects for LUMI-G were selected for each member country within the LUMI consortium. Therefore, also within the Czech Republic, only two and two projects were selected!
LUMI-C PILOT PROJECTS
Principal Investigator: Dominik Legut
Title: Understanding the Physics of Phonons utilizing electronic structure and atomistic calculations employing state-of-the-art methods
Abstract: Atomic vibrations determine a wide range of properties of materials, such as heat flow, thermal expansion, metal to insulator transitions of solids, among others. Recently we have discovered (Nat. Physics 2020, PRB 2021) its unique role in the so-called Verwey transition in magnetite (the oldest magnetic material known to humankind) that puzzles physicist for almost over a century. Here, the charge orbital order of the so-called trimerons and their lattice dynamics (phonons) are the key elements to explain the metal to insulating transition in magnetite discovered very recently. However, the modeling was done for a much-idealized system, i.e. defect-free crystal with no vacancies, impurities or even a pressure dependence, where the metalization and superconductivity is speculated to occur. Actual experimental determination of vibrational properties at a nano-scale is a challenging task. However, the recent development of a new generation of electron beam monochromators enabled scanning transmission electron microscopes to measure vibrational properties down to atomic scale. We will apply a recently introduced theoretical method Frequency Resolved Frozen Phonon Multislice (FRFPMS) to simulate vibrational spectroscopy experiments for application-relevant nano-structures with interfaces or defects, with particular focus on magnetite, its nanoparticles, 2D materials, and their van der Waals heterostructures. Hence, it is important to determine the phonons dispersion with the tabletop setup as well as to understand its role in challenging physical phenomena.
Principal Investigator: Denys Biriukov
Title: Interaction network in extracellular space: an all-atom simulation model of the glycocalyx and cell membrane
Short description: The extracellular space is a key communication channel for eukaryotic cells. It is a very dynamic and complex environment, especially above the plasma membrane. This membrane proximal zone is known as the glycocalyx. For decades, theoreticians and experimentalists have been trying to describe it piece by piece, but its complexity and functional length scales made until now unfeasible to model a significant portion of it. The main limitations were the computational power required for such simulations, the absence of uniform and accurate simulation models, and large uncertainty of which molecular motifs should be modeled. The drastic increase in computational power provided by it4i via Lumi together with our recent development of improved molecular models has changed this situation. Here, we aim to mimic the outer-cell network in unprecedented scale, combining in a single simulation system all key components simultaneously (i.e., lipids, proteins, sugars, ions, and water) modeled at full atomic resolution. This accounts for several tens of millions of atoms that needs to be simulated for several microseconds at least. Such computer but realistic model of the glycocalyx will help probing the nature of interactions currently unknown. This pilot call is a unique opportunity (technically and mentally) to engage with this high risk project which not long ago was merely a “dream”. All generated data will become publicly available, hopefully fostering their analysis and stimulating further the field.
LUMI-G 1ST CALL PILOT PROJECTS
Principal Investigator: Sergiu Arapan
Title: Computational search for novel two dimensional thermoelectric materials
Abstrakt: Two-dimensional (2D) materials show a wide variety of electronic properties and have recently emerged as promising candidates for future ultra-flat and compact electronics. Since the discovery of graphene, the 2D form of graphite, by A. Geim and K. Novoselov less than two decades ago, a plethora of discovered 2D materials serving metallic, semimetallic, semiconducting, insulating, ferromagnetic, antiferromagnetic, super-conducting behavior offers an unprecedented research space. Assembling 2D materials into vertical van der Waals (vdW) heterostructures with atomically sharp interfaces offers a unique platform to engineer various device functionalities. By the proximity effects we can combine their intrinsic properties into new unexpected ones. The 2D vdW materials offer great advantages to tune the device properties through band structure and phonon spectrum engineering. Thus, vdW structures are regarded as new technological solutions for energy conversion in modem electronic devices. One of the challenges is to improve the thermoelectric (TE) properties to efficiently harvest electricity from waste heat in a wide temperature range. This research aims to clarify the structural and physical properties of advanced 2D van der Waals materials using the state-of-the-art computational research methods, towards the development of energy-harvesting materials and next-generation electronic devices.
Principal Investigator: David Číž
Title: Benchmarking TensorFlow on AMD GPUs
Abstrakt: Our research focuses on exploring the viability of training machine learning models on AMD graphics cards in an HPC environment using the machine learning framework TensorFlow. Until recently, support for AMD cards in machine learning frameworks was quite limited. ROCm open software platform has created a port of TensorFlow, which will allow us to utilize AMD HPC resources for our research. We aim to create benchmarks that will serve as a baseline for LUMI users, and also allow us to compare them with the same benchmarks ran on the Karolina supercomputer. For this purpose, we will train popular deep learning architecture on different tasks and gather various metrics.