HYBRILIT WORKSHOP 2025: Overview of capabilities of JINR heterogeneous computing infrastructure On 25–26 November 2025, the Meshcheryakov Laboratory of Information Technologies hosted the «HYBRILIT WORKSHOP 2025: Towards Efficient Scientific Computing», which highlighted the capabilities of the “Govorun” supercomputer and the development status of the HybriLIT heterogeneous platform. The event was dedicated to the 70th anniversary of JINR and the 60th anniversary of MLIT. The workshop featured reports and tutorials covering cutting-edge high-performance computing technologies, GPU-accelerated workflows, tools for large-scale simulations in physics, biology, and Big Data, as well as the integration of machine and deep learning into scientific research. MLIT Director Sergei Shmatov welcomed the participants, underlining the significance of the HybriLIT platform for solving a wide range of tasks both at MLIT and at the other JINR Laboratories, as well as the growing interest of the scientific community in the platform. He also expressed gratitude to the organizers and the participants, wishing everyone fruitful work. During his speech, MLIT Scientific Leader Vladimir Korenkov called the “Govorun” supercomputer the key component of the Institute’s computing infrastructure. According to his estimates, the role of this computing resource will increase with the development of parallel and distributed computing technologies. The MLIT strategic objective to establish a universal and convenient environment for researchers working on a variety of scientific tasks was also pointed out. The scientific program of the workshop was rich and comprised two main thematic blocks. The first day was devoted to an overview of the infrastructure and key services. Maxim Zuev (MLIT) presented the current state and development prospects of the “Govorun” supercomputer and provides some interesting statistics. To date, the HybriLIT user base consists of 357 scientists. Since the supercomputer’s commissioning, over 14 million tasks have been performed, which is equivalent to nearly 112 million core hours. The scientific results obtained using the resources of the “Govorun” supercomputer are reflected in 509 user papers, two of which are published in the Nature Physics journal. Dmitry Belyakov (MLIT) enlarged upon the operating principles of the HybriLIT heterogeneous platform and presented a review of data processing and storage systems implemented on the platform. Oksana Streltsova (MLIT) discussed the ML/DL/HPC ecosystem for solving applied tasks. Practical issues such as the use of the HLIT-VDI service for working with proprietary software (Mikhail Matveev, MLIT) and the procedure for getting access to the HybriLIT platform (Shushanik Torosyan, MLIT) were also considered. Then the focus shifted to specific scientific applications. The results of computational investigations in the field of properties of superheavy element atoms (Dipayan Sen, BLTP) and molecular physics (Miroslav Ilias, BLTP) were delivered, illustrating the effectiveness of employing the platform for fundamental sciences. The second day of the workshop focused on applied aspects and practical tools. The participants learned about the application of the “Govorun” supercomputer to Monte Carlo simulations in Geant4 and about the training of neural networks in PyTorch using HybriLIT GPUs (Konstantin Chizhov, MLIT). A number of talks covered interdisciplinary research: the automation of biological data analysis (Sara Shadmehri, MLIT), work with medical data (Anastasiya Anikina, MLIT), and the modeling of quantum-classical algorithms performed on the quantum polygon of the HybriLIT heterogeneous platform (Alla Bogolubskaya, MLIT). Adiba Rahmonova (MLIT) spoke about current approaches to accelerating computations in Python using the NumPy library and the Numba JIT compiler. An important part of the workshop was tutorials, which enabled the participants to consolidate their theoretical knowledge. Miroslav Ilias (BLTP) conducted a tutorial entitled “Computational Molecular Physics” on the practical use of software packages in the field of molecular physics. Within a tutorial entitled “Scientific Computing in Python in the ML/DL/HPC Ecosystem of the HybriLIT Platform”, Oksana Streltsova, Maxim Zuev, Adiba Rahmonova, Tatevik Bezhanyan, and Sara Shadmehri (MLIT) demonstrated the opportunities of the ML/DL/HPC ecosystem for elaborating algorithms based on machine learning and deep learning methods. Within a tutorial entitled “Scientific Computing and Development of Parallel Algorithms Based on the Julia Programming Language”, Maksim Bashashin and Mikhail Matveev (MLIT) illustrated the capabilities of the Julia programming language for exploring multiparameter models described by systems of nonlinear differential equations. These sessions provided a unique opportunity to gain hands-on experience with platform tools under the guidance of experienced users and developers. Within the workshop there was held a roundtable discussion that allowed for direct dialogue: the users were able to ask questions, and the developers, in turn, received valuable feedback. As MLIT Deputy Director for Scientific Work Dmitry Podgainy said, the HybriLIT Workshop 2025 was organized as a multifunctional platform for sharing experience, training, and demonstrating the new capabilities of the platform. Holding the event in this format was a long-awaited step, which was realized thanks to two key factors, namely, the initiative of the users themselves, eager to share successful practices with colleagues, and the developers’ need to introduce the user community to the latest elements of the HybriLIT infrastructure, such as the integrated ML/DL/HPC ecosystem. This format of interaction has proven effective. Dmitry Podgainy expressed the opinion that such meetings should be held on a regular basis. The HybriLIT Workshop 2025 became a significant event in the life of the JINR scientific community. It not only demonstrated that the HybriLIT platform was a versatile and powerful tool to support cutting-edge research in physics, biology, machine learning, and data analysis, but also contributed to combine platform users into a community to share best practices and master new working methods. Photos by Olesya Chepurchenko
