BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:America/Chicago X-LIC-LOCATION:America/Chicago BEGIN:DAYLIGHT TZOFFSETFROM:-0600 TZOFFSETTO:-0500 TZNAME:CDT DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0500 TZOFFSETTO:-0600 TZNAME:CST DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260202T201312Z LOCATION:267 DTSTART;TZID=America/Chicago:20251117T090000 DTEND;TZID=America/Chicago:20251117T173000 UID:submissions.supercomputing.org_SC25_sess198@linklings.com SUMMARY:PMBS25: The 16th International Workshop on Performance Modeling, B enchmarking, and Simulation of High-Performance Computer Systems DESCRIPTION:The PMBS25 workshop is concerned with the comparison of high-p erformance computing systems through performance modeling, benchmarking, o r through the use of tools such as simulators. We are particularly interes ted in research that reports the ability to measure and make tradeoffs in software/hardware co-design to improve sustained application performance. We are also keen to capture the assessment of future systems. The aim of t his workshop is to bring together researchers, from industry and academia, concerned with the qualitative and quantitative evaluation and modeling o f high-performance computing systems. Authors are invited to submit novel research in all areas of performance modeling, benchmarking, and simulatio n, and we welcome research that brings together current theory and practic e. We recognize that the term "performance" has broadened to include power consumption and reliability, and that performance modeling is practiced t hrough analytical methods and approaches based on software tools and simul ators.\n\nCharacterizing the Impact of GPU Power Management on an Exascale System\n\nAs GPU-accelerated high-performance computing (HPC) systems app roach exascale performance, controlling energy consumption without comprom ising throughput is essential. Architectures such as the AMD MI250X-based Frontier supercomputer provide runtime mechanisms like frequency and power capping, enabl...\n\n\nMariana Costa (UFRGS); Antigoni Georgiadou and Jam es B. White III (Oak Ridge National Laboratory (ORNL)); Bruno Villasenor A lvarez and Jordà Polo (Advanced Micro Devices, Inc. (AMD)); Woong Shin (Oa k Ridge National Laboratory (ORNL)); Philippe O. A. Navaux (UFRGS); Bronso n Messer (Oak Ridge National Laboratory (ORNL)); and Arthur Lorenzon (Fede ral University of Rio Grande do Sul, Brazil)\n---------------------\nAfter noon Break - PMBS25\n---------------------\nLunch break (on your own)\n--- ------------------\nPretraining LLMs at Scale: Tuning Strategies and Perfo rmance Portability.\n\nTraining large language models (LLMs) at scale pres ents challenges that demand careful co-design across software, hardware, a nd parallelization strategies. In this work, we introduce a communication- aware tuning methodology for optimizing LLM pretraining, and extend the pe rformance portability metri...\n\n\nAdrián Pérez Diéguez, Àlex Batlle Case llas, Aleix Torres-Camps, Harris Teague, and Jordi Ros-Giralt (Qualcomm)\n ---------------------\nMoE-Inference-Bench: Performance Evaluation of Mixt ure of Expert Large Language and Vision Models\n\nMixture of Experts (MoE) models have enabled the scaling of Large Language Models (LLMs) and Visio n Language Models (VLMs) by achieving massive parameter counts while maint aining computational efficiency. However, MoEs introduce several inference -time challenges, including load imbalance across expe...\n\n\nKrishna Tej a Chitty-Venkata (Argonne National Laboratory (ANL)); Sylvia Howland, Gola ra Azar, Daria Soboleva, and Natalia Vassilieva (Cerebras); Siddhisanket R askar (Pacific Northwest National Laboratory (PNNL)); and Murali Emani and Venkatram Vishwanath (Argonne National Laboratory (ANL))\n--------------- ------\nILAN: The Interference- and Locality-Aware NUMA Scheduler\n\nModer n HPC platforms increasingly adopt NUMA architectures, where OpenMP task-b ased programming model is a standard for enabling dynamic parallelism. How ever, the default OpenMP runtime is topology-agnostic, and the existing af finity policies are insufficient to ensure optimal performance on modern . ..\n\n\nEdvin Mellberg, Axel Carlsson, Jing Chen, and Miquel Pericas (Chal mers University of Technology and University of Gothenburg)\n------------- --------\nOn the Performance and Scalability of Cloud Supercomputers: Insi ghts from Eagle and Reindeer\n\nLaunch of Eagle, Azure’s hyper-scale super computer and the Number 3 on TOP500 list in November 2023, marked a new er a where cloud providers are at the forefront of supercomputing. Despite it s rapid expansion, public knowledge on the performance and scalability of cloud-based supercomputing is ...\n\n\nAmirreza Rastegari, Prabhat Ram, an d Michael F. Ringenburg (Microsoft Corporation)\n---------------------\nA GPU FFT Wrapper to Co-optimize Floating-Point Precision and Library Select ion via Predictive Error Modeling\n\nApproximate and low-precision computi ng are essential for modern applications, and effectively leveraging avail able precision options can deliver substantial gains in performance and en ergy efficiency. \nWe focus on the Fast Fourier Transform (FFT), a represe ntative function used in scientific comput...\n\n\nJulius Lehner, Eishi Ar ima, and Martin Schulz (Technical University of Munich)\n----------------- ----\nPMBS25: The 16th International Workshop on Performance Modeling, Ben chmarking, and Simulation of High-Performance Computer Systems\n\nThe PMBS 25 workshop is concerned with the comparison of high-performance computing systems through performance modeling, benchmarking, or through the use of tools such as simulators. We are particularly interested in research that reports the ability to measure and make tradeoffs in software/hardwar...\ n\n\nSteven A. Wright (University of York, England); Simon Hammond (Nation al Nuclear Security Administration (NNSA)); and Sascha Hunold (Technical U niversity of Vienna)\n---------------------\nA Peak Performance Model for All-to-all on Hierarchical Systems and Its Applications\n\nAn accurate mea sure of communication performance is a key component of optimizing large s cale high performance computing applications. This paper presents a model for the peak performance of all-to-all communication, in the context of sy stems composed of a hierarchy of interconnect bandwidths; a com...\n\n\nRo hini Uma-Vaideswaran (Georgia Institute of Technology), Joshua Romero (NVI DIA Corporation), Daniel Dotson (Georgia Institute of Technology), David A ppelhans (NVIDIA Corporation), and P. K. Yeung (Georgia Institute of Techn ology)\n---------------------\nModelling Load Imbalance In Shared Memory M ulticore Systems\n\nMemory bandwidth has become the primary limiting facto r of performance in many modern HPC applications, and it poses a limit to scalability because the achievable memory\nbandwidth only grows linearly w ith a small number of CPU cores. When the number of cores concurrently usi ng the memory system exce...\n\n\nJohannes Langguth (Simula Research Labor atory, University of Bergen); James Trotter (Simula Research Laboratory); and Xing Cai (University of Oslo, Simula Research Laboratory)\n----------- ----------\nMorning Break - PMBS25\n---------------------\nPerfAnalyzer: R evealing Performance Trends using Version Oriented Visual Analysis of Scie ntific Software\n\nUnderstanding the behavior of scientific software is es sential in maintaining the integrity and transparency of computational res earch. Tracking the changes in computational parameters (input-output para meters, configuration parameters for hardware and software) across differe nt versions of software...\n\n\nSayef Azad Sakin (University of Utah, Los Alamos National Laboratory (LANL)) and James Ahrens (Los Alamos National L aboratory (LANL))\n---------------------\nImplications of Full-System Mode ling for Superconducting Architectures\n\nAs Moore's Law slows, supercondu cting electronics offer ultra-low-power, high-speed computation potential. This paper presents the first full-system superconducting architecture mo deling in gem5, evaluating superconducting cores, caches, and interconnect s under realistic workloads. We extend gem5 wi...\n\n\nKunal Pai, Mahyar S amani, Anusheel Nand, and Jason Lowe-Power (University of California, Davi s)\n---------------------\nExperiences of Porting Structured and Unstructu red Stencil Applications to FPGA using SYCL\n\nIn this paper we explore a stencil application written in SYCL on both CPU and FPGA architectures.\nW e prepare two versions of the application, using a structured grid and an unstructured grid, and then optimise these implementations for CPU and FPG A architectures, with a focus on maintaining portabi...\n\n\nZadok Storkey , Steven Wright, and Ian Gray (University of York, England)\n------------- --------\nDetermining Levels of Detail for Simulators of Parallel and Dist ributed Computing Systems via Automated Calibration\n\nThere are two sourc es of inaccuracy when simulating parallel and distributed computing system s: (i) a simulator implemented at an insufficient level of detail; and (ii ) incorrectly calibrated simulation parameter values. Increasing the simu lator's level of detail can improve accuracy, but at the co...\n\n\nJesse McDonald and Yick-Ching Wong (University of Hawaii at Manoa), Kshitij Meht a and Frederic Suter (Oak Ridge National Laboratory (ORNL)), Rafael Ferrei ra Da Silva (Oak Ridge National Laboratory), Loic Pottier (Lawrence Liverm ore National Laboratory (LLNL)), Ewa Deelman (University of Southern Calif ornia), and Henri Casanova (University of Hawaii at Manoa)\n-------------- -------\nBeyond Guess and Check: Quantifying the Fidelity of Proxy Applica tions\n\nProxy applications are targeted submodels of larger parent applic ations, designed to represent key characteristics such as the programming model, memory usage, or communication behaviors. \n\nProxy applications ar e valuable for system design and optimization, offering a more manageable and privacy-pre...\n\n\nSi Chen (Emory University); Simon Garcia de Gonzal o, Omar Aaziz, and Jeanine Cook (Sandia National Laboratories); and Avani Wildani (Cloudflare)\n---------------------\nCGSim: A Simulation Framework for Large Scale Distributed Computing Environment\n\nLarge-scale distribu ted computing infrastructures like the Worldwide LHC Computing Grid (WLCG) require comprehensive simulation tools for performance evaluation and res ource optimization. Existing simulators suffer from limited scalability, h ardwired algorithms, lack of real-time monitoring, and ina...\n\n\nSairam Sri Vatsavai (Brookhaven National Laboratory); Raees Khan Ahmed (universit y of pittsburgh); Kuan-Chieh Hsu, Ozgur Kilic, Yihui (Ray) Ren, David Park , and Paul Nilsson (Brookhaven National Laboratory); Tania Korchuganova (U niversity of Pittsburgh); Sankha Dutta (Brookhaven National Laboratory); J oseph Boudreau (University of Pittsburgh); Tasnuva Chowdhury (Brookhaven N ational Laboratory); Shengyu Feng (Carnegie Mellon University); Fatih Furk an Akman (University of Massachusetts); Adolfy Hoisie (Brookhaven National Laboratory); Scott Klasky (Oak Ridge National Laboratory (ORNL)); Tadashi Maeno (Brookhaven National Laboratory); Verena Ingrid Martinez Outschoorn (University of Massachusetts); Norbert Podhorszki and Frédéric Suter (Oak Ridge National Laboratory (ORNL)); John Rembrandt (Remy) Steele (Universi ty of Massachusetts); Wei Yang (SLAC National Accelerator Laboratory); Yim ing Yang (Carnegie Mellon University); and Shinjae Yoo and Alexei Klimento v (Brookhaven National Laboratory)\n\nRecording: Livestreamed, Recorded\n\ nRegistration Category: Technical Program Reg Pass, Workshop Reg Pass\n\nS ession Chairs: Steven A. Wright (University of York, England); Simon Hammo nd (National Nuclear Security Administration (NNSA)); and Sascha Hunold (T echnical University of Vienna) END:VEVENT END:VCALENDAR