39th Workshop on Sustained Simulation Performance
May 27 - May 28
Draft Agenda
All times are given in Central European Summer Time (CEST).
Tuesday, May 27th, 2025 |
|
| 09:30 – 09:45 | Welcome & Introduction Michael Resch, High-Performance Computing Center Stuttgart, University of Stuttgart |
| 09:45 – 10:15 | tbd Michael Resch, High-Performance Computing Center Stuttgart, University of Stuttgart tbd |
| 10:15 – 10:45 | Research and user support activities at Tohoku University Cyberscience Center Hiroyuki Takizawa,Cyberscience Center, Tohoku University Tohoku University Cyberscience Center has been operating vector supercomputers and assisting users in fully utilizing their potential. This presentation will report on our recent efforts to help users optimize their code for our computing system, AOBA, which is powered by the latest generation of NEC SX-Aurora TSUBASA, the most powerful vector supercomputer. At both the system operation and research levels, we are continuously exploring effective methods to make the most of vector computing technologies. Performance evaluation results demonstrate that the SX-Aurora TSUBASA achieves high sustained performance for memory-intensive applications without requiring special programming models or languages. |
| 10:45 – 11:00 | Coffee Break |
| 11:00 – 11:30 | Musing on performance sustainability in the age of AI, superchips, APUs, increasing TCOs and NetZero impact Sadaf Alam, Bristol Center of Supercomputing, University of Bristol Supercomputing ecosystems have experienced considerable shifts since the early 2020 across applications and technology domains. HPC and supercomputing resources are increasingly being allocated to AI—a domain where the frequency of hardware changes and especially software stacks updates are considerably different compared to classic modelling and simulation HPC applications. Recently, the fastest reported HPL system on November 2024 Top500 list is based on an APU, or accelerated processor unit. This talks overviews Isambard-AI, a UK national AI RR or research resource comprising Nvidia Arm-GPU superchip called GH200, its software stack for AI and HPC, and its sustainability credentials using the Modular Data Centre (MDC) solution to manage Total Cost of Ownership (TCO) and NetZero impact. |
| 11:30 – 12:00 | Future Computing at HLRS Johannes Gebert, High-Performance Computing Center Stuttgart, University of Stuttgart tbd |
| 12:00 – 12:30 | Introduction and challenge of new supercomputing system towards the Open Science era Susumu Date, University of Osaka Osaka University has been working on the procurement of a supercomputing system after OCTOPUS. We will plan to complete the installation of it and start the operation in September 2025. In this talk the speaker introduce and explain the specification of the new supercomputing system as well as a challenge which we have faced for realizing supercomputing system in the Open Science era. |
| 12:30 – 13:30 | Lunch Break |
| 13:30 – 14:00 | New Brand “NEC BluStellar” Use Case – Research Information Infrastructure(RII) Futoshi Tabata, NEC Japan tbd |
| 14:00 – 14:30 | NEC SX-Aurora TSUBASA – Our best friend for a long time Christoph Wenzel, Institut für Aero- und Gasdynamik (IAG), University of Stuttgart For a long time, the NEC SX-Aurora TSUBASA has been a valuable addition to the HPE Apollo (Hawk) system at HLRS. Even though its size was not sufficient for large-scale production runs for the fundamental research on turbulent boundary layers with direct numerical simulation (DNS), Aurora has still played a central role in our research pipeline. In this talk, our group’s experience working with the SX-Aurora platform will be presented, highlighting its integration into our workflows. Additionally, a performance study of our in-house DNS code NS3D on Aurora will be presented, providing insights into computational efficiency achieved on Aurora. |
| 14:30 – 15:00 | tbd Hartwig Anzt, TU München tbd |
| 15:00 – 15:30 | Coffee Break |
| 15:30 – 16:00 | Sustainable research software for accessible high-performance computing Michael Schlottke-Lakemper, High-Performance Scientific Computing, University of Augsburg Modern supercomputers are becoming increasingly heterogeneous, incorporating hardware components from multiple vendors. At the same time, high-performance computing software development has grown more collaborative, often uniting research groups and institutions across different regions. Coupled with the constant addition of new features and performance optimizations, this raises critical questions about sustainability: How can we handle hardware complexity, coordinate diverse development teams, and maintain evolving research software within an academic environment while still writing energy-efficient, high-performance code?In this talk, we present some of our strategies for tackling these challenges through the Trixi Framework. We introduce Trixi.jl, a high-order numerical simulation environment for conservation laws built in Julia, along with its spin-off packages (TrixiShallowWater.jl, TrixiAtmo.jl) and its sister project, TrixiParticles.jl. We then discuss how we handle software architectures, automation, code reuse, and organizational practices to balance extensibility with accessible high performance on heterogeneous systems. Finally, we point out remaining open questions and outline plans for future development. |
| 16:00 – 16:30 | tbd Jonathan Schäfer, High-Performance Computing Center Stuttgart, University of Stuttgart tbd |
| 16:30 – 17:00 | Reproducible and Performance-Optimized Environments for Large-Scale Machine Learning Applications Felix Ruhnke, High-Performance Computing Center Stuttgart, University of Stuttgart With the increasing size of machine learning models for solving complex problems, the demand for computational resources is rising significantly. This leads to a stronger convergence between the disciplines of machine learning and High-Performance Computing. The growing performance capabilities of modern machine learning models open up new application areas, including safety-critical domains such as medical technology. At the same time, the scientific community is increasingly drawing attention to a reproducibility crisis in the field of machine learning. The intersection of these three developments underscores the necessity of thoroughly investigating reproducible and performance-optimized environments for large-scale machine learning applications. In this work, various implementations of the Message Passing Interface in containerized environments were developed to examine the impact of different communication modes on the numerical reproducibility of results. For this purpose, benchmark tests were conducted to optimize polynomials of varying degrees on a single node with a varying number of Graphics Processing Units. The gradient aggregation algorithms Average and Adaptive Summation were employed. The results showed that variations in communication models using the Average algorithm had no impact on numerical reproducibility. In contrast, tests using the Adaptive Summation algorithm with varying communication modes resulted in non-reproducible outcomes.Furthermore, it was observed that, due to the non-associative nature of floating-point arithmetic operations and the varying execution order of computations during parallel training of machine learning models, deviations of over 7% in the number of iterations until convergence can occur. Additionally, the investigations revealed a highly sensitive convergence behavior of the models during training concerning configuration changes, emphasizing the need for careful selection of the computing environment and precise hyperparameter adjustment. |
| 17:00 – 17:30 | Development of dynamic resource assignment for effective system usage Masatoshi Kawai , Tohoku University Recently, energy efficiency as well as improving parallel performance of applications has become important for the operation and use of supercomputers. In this presentation, we will introduce a developing platform that provides dynamic resource assignments for improving parallel performance and energy consumption. |
Wednesday, May 28, 2025 |
|
| 09:00 – 09:30 | tbd Alexander Brem, University of Stuttgart tbd |
| 09:30 – 10:00 | HammerHAI – The German AI Factory for Engineering, Global Challenges and Industry Bastian Koller, High-Performance Computing Center Stuttgart, University of Stuttgart This talk will provide insight into the German AI Factory HammerHAI, which resulted from a European Key Initiative and which started service operations in Q1/25 |
| 10:00 – 10:30 | Leveraging Cloud-Native Supercomputing for AI Workflows Dennis Hoppe, High-Performance Computing Center Stuttgart, University of Stuttgart This talk explores the transformative potential of cloud-native supercomputing concepts and their impact on AI workflows. While powerful and GPU-rich, traditional High-Performance Computing systems present unique challenges when executing AI tasks. A clear paradigm shift is underway, moving from a system-centric to a user-centric approach, where usability, accessibility, and flexibility become key design criteria. Using the AI Factory HammerHAI as a concrete example, the presentation will demonstrate how HammerHAI aims to lower these barriers by integrating cloud-native concepts like containerisation and orchestration, thereby optimising supercomputing resources specifically for the practical needs of the AI community. |
| 10:30 – 11:00 | Coffee Break |
| 11:00 – 11:30 | tbd Matthias Meinke, Institute of Aerodynamics, RWTH Aachen University tbd |
| 11:30 – 12:00 | tbd Gregor Olenik, TU München tbd |
| 12:00 – 12:30 | Accelerating the FlowSimulator: Speeding up the HPC codes used at DLR Immo Huismann, DLR Dresden tbd |
| 12:30 – 13:30 | Lunch Break |
| 13:30 – 14:00 | Super-resolution Reconstruction of Three-dimensional Vorticity Fields by Latent Diffusion Models Mitsuo Yokokawa, Tohoku University tbd |
| 14:00 – 14:30 | tbd Benjamin Schnabel, High-Performance Computing Center Stuttgart, University of Stuttgart tbd |
| 14:30 – 15:00 | Power and Performance Nico Formanek, High-Performance Computing Center Stuttgart, University of Stuttgart The biggest driver of computing performance are hardware improvements. Even though the relative energy efficiency (e.g. FLOPS/watt) is improving at an almost exponential rate this does not translate in any reduction in absolute energy consumption. Rebound effects like this have been studied for centuries in economics starting from Jevons (1865) but there is still no consensus if economic growth can be decoupled from absolute energy consumption. Here I will argue that we face a similar problem in computing, i.e. that performance cannot be decoupled from absolute energy input. This in turn casts doubt on the feasibility of sustainability efforts like the GREENER (2023) principles. I will close by evaluating several accounts of why we still could want to improve performance even in the light of such hard tradeoffs. |
| 15:00 – 15:30 | Lunch Break |
| 15:30 – 16:00 | Foundation Models und Simulation Steffen Staab, University of Stuttgart tbd |
| 16:00 – 16:30 | Quantum Computing Status -Technologies, Benchmarks and Use Cases in Academic and Industry Fields Shintaro Momose, NEC Japan tbd |
| 16:30 – 17:00 | Introduction of Confidential Computing in HPC: Usage Model, Use Cases and Challenges Tamil Kokmakov, NEC Germany HPC data centres accommodate users from various domains, each with varying security requirements. Sensitive data processing, such as for medical use cases or intellectual property protection, requires stricter security measures, including encryption across all data states: at-rest, in-motion and in-use. While parallel file systems found in HPC, such as GPFS and Lustre, already offer encryption of data at-rest and in-motion, encryption keys and sensitive data are not yet fully protected in memory. Confidential computing addresses such protection of data in-use by performing computations in the trusted execution environments, where data and code are secured at the hardware level. This talk introduces confidential computing in the context of HPC, covering its use cases and challenges. |
| 17:00 – 17:30 |
Farewell |
Committee
Program Committee
- Prof. Michael Resch, Stuttgart University, HLRS
- Prof. Hiroaki Kobayashi, Tohoku University
- Dr. Wolfgang Bez, NEC Deutschland GmbH, Division HPCE
- Prof. Sabine Roller, German Research School for Simulation Sciences GmbH
Organizing Committee
- Prof. Michael Resch, Stuttgart University, HLRS
- Johannes Gebert, Stuttgart University, HLRS
- Prof. Hiroaki Kobayashi, Tohoku University
- Prof. Hiroyuki Takizawa, CyberScience Center, Tohoku University