## Overview ENCCS and [KTH](https://www.kth.se/en) come together with the [University of Oregon](https://www.uoregon.edu/) and [ParaTools, Inc](https://www.paratools.com) for the organization of a hybrid workshop. The event brings together code developers, researchers, and research software engineers working in high-performance computing (HPC) and the HPC-AI software ecosystem for science. It provides an excellent opportunity to exchange innovative ideas and best practices for performance monitoring and software management across supercomputers and cloud platforms. This workshop includes: • Introduction to [TAU](https://tau.uoregon.edu), [TAU](https://tau.uoregon.edu) Performance System and the E4S project.• Performance Evaluation on LUMI and Commercial Cloud Platforms.• Instrumentation of MPI and GPU (ROCm and CUDA) Applications.• Performance Data Management with TAUdb ([TAU](https://tau.uoregon.edu) Database)• Analyzing performance data with ParaProf and PerfExplorer.• Containers and Spack Package Managers in HPC with E4S.• E4S (e4s-cl and e4s-alc) Tools on Commercial Cloud HPC Platforms.• AI tools with Jupyter Notebooks, Codium, PyTorch, JAX, NVIDIA NeMo and BioNeMo. The theoretical optimization of complex code is somewhat limited, and great progress in scientific computing has been achieved through performance analysis tools. Indeed, advances in massively parallel systems like the LUMI Supercomputer and Cloud HPC platforms have further boosted our understanding of complex phenomena at unprecedented resolutions. To leverage these resources effectively, robust tools for performance monitoring, debugging, and software management are required. These tools enable the development of key applications for science, industry, and society.knowledge and enabling key applications for science, industry, and society. ## Performance Monitoring with TAU and E4SThe [TAU](https://tau.uoregon.edu) Performance System® is a state-of-the-art, robust suite of tools for performance measurement, analysis, and visualization in high-performance computing (HPC). It is highly portable and supports the full range of HPC systems, software, and applications. TAU provides profiling and tracing capabilities for performance analysis of parallel programs written in Fortran, C, C++, UPC, Java, and Python. [TAU](https://tau.uoregon.edu) (Tuning and Analysis Utilities) can instrument code at the level of functions, methods, basic blocks, or statements, or via event-based sampling, thereby capturing detailed performance data. It supports instrumentation inserted automatically (via the Program Database Toolkit), dynamically (for example with DyninstAPI), during runtime in the Java Virtual Machine, or manually via its API. Its visualization tool, ParaProf, presents performance data graphically in both aggregate and per node/context/thread form, allowing users to rapidly identify performance bottlenecks. TAU can also produce event traces compatible with Vampir, Paraver, or JumpShot trace viewers. The [E4S](https://e4s.io) (Extreme-scale Scientific Software Stack) project is an open-source software ecosystem designed for developing, deploying, and running scientific and HPC/AI applications. It is a community effort to curate and maintain a collection of software packages, enabling both from-source builds and containerized deployments across a broad spectrum of platforms. [E4S](https://e4s.io) provides container images, pre-installed binaries, and source builds of over a hundred HPC and AI software packages, including libraries, runtimes, tools, and frameworks, ensuring performance portability across CPU and GPU architectures. It leverages the Spack package manager as its core delivery mechanism, enabling reproducible, customized, and interoperable builds. The [E4S](https://e4s.io) software distribution is rigorously tested on a wide range of platforms, from laptops and departmental clusters to leadership-class supercomputers, and supports container runtimes such as Docker, Singularity, Shifter, and CharlieCloud. Releases of [E4S](https://e4s.io) are designed to expand support for evolving hardware architectures (for example, new GPU platforms) and to incorporate widely used AI/ML frameworks and scientific applications. ## What you will learn After attending this workshop, you will: • Understand core features of performance evaluation tools• Understand performance evaluation tools and techniques for CPU and GPU platforms• Understand how to optimize applications on AMD GPUs on LUMI and NVIDIA GPUs on Commercial Cloud Platforms• Use a comprehensive software stack for HPC and AI applications ## Prerequisites• PhD students, postdocs, industry engineers• Basic familiarity with general physics and plasma physic• Some previous practical experience running some plasma code• Basic familiarity with Unix shell• Basic familiarity with HPC environment is helpful but not mandatory ## Tentative Agenda Friday, 24 October 2025 (**14:00-18:00**): HPC Onboarding Sessions (Online)~2 hours online session to help participants log in the LUMI cluster~2 hours online session to help participants log in the Cloud HPC cluster Tuesday, 28 October 2025 (**10:00-12:00**): HPC Onboarding Session (Online)~2 hours online Q & A session to clear unanswered HPC cluster questions. Wednesday, 29 October 2025: Working with [TAU](https://tau.uoregon.edu) Performance System®| Time | Topic || :------------: | :---------------------------------------------------------------------: || 08:30-09:00 | Reception || 09:00-09:15 | - Welcome, Logistics and Introductions || 09:15-10:00 | - Introduction to [TAU](https://tau.uoregon.edu) || 10:00-10:30 | - [TAU](https://tau.uoregon.edu) Performance System Setup on LUMI and Cloud HPC Platforms || 10:30-11:00 | Break || 11:00-12:00 | - Instrumentation: MPI, GPU Runtimes, tau_exec
- Hands-On: AWS Examples using [TAU](https://tau.uoregon.edu) and ParaProf
- Performance Application Programming Interface (PAPI) || 12:00-14:00 | Lunch || 14:00-15:30 | - Hands-On: [TAU](https://tau.uoregon.edu) and GPU (ROCm and CUDA) Runtimes Examples
- Perfetto.dev Trace Visualizer || 15:30-15:45 | Break || 15:45-16:45 | - TAUdb ([TAU's](https://tau.uoregon.edu) Performance Database)
- PerfExplorer ([TAU's](https://tau.uoregon.edu) performance data mining environment) || 16:45-17:00 | Q & A and Concluding Remarks | THursday, 30 October 2025: Working with the [E4S](https://e4s.io) Project | Time | Topic ||:------------:|:---------------------------------------------------------------------: || 08:30-09:00 | Reception || 09:00-09:15 | - Welcome, Logistics and Introductions || 09:15-10:00 | - Introduction to the [E4S](https://e4s.io) Project || 10:00-10:30 | - E4S System Setup on LUMI and Cloud HPC Platforms
- Simple Docker and Singularity Examples || 10:30-11:00 | Break || 11:00-12:00 | - Containers in HPC
- Spack Package Manager
- Hands-On: MPI Examples on LUMI and Cloud HPC Platforms
- Introduction to [E4S](https://e4s.io) Tools || 12:00-14:00 | Lunch || 14:00-15:30 | - AI Tools
- Jupyter Notebooks
- Codium
- PyTorch, JAX
- NVIDIA NeMo
- NVIDIA BioNeMo || 15:30-15:45 | Break || 15:45-16:45 | - E4S (e4s-cl and e4s-alc) Tools on Commercial Cloud HPC Platforms
- Hands-On: HPC and AI Tools || 16:45-17:00 | Q & A and Concluding Remarks | ## Format/Venue The event is going to be held at the Stockholm RISE offices on KTH campus [Drottning Kristinas väg 61](https://maps.app.goo.gl/9rphvAS7ugRUC46U8) in room **Sundblassalen**. The closest metro station (marked T) is **Tekniska Högskolan**. However, we will also broadcast the seminars and hands-on sessions live for online participants who will be able to participate actively in the workshop through live Q&A and chat. ## DisclaimerDue to EuroCC2 regulations, we cannot except generic or private email addresses. Please use your official university or company email address. This training is for users that live and work in the European Union or a country associated with Horizon 2020. You can read more about the countries associated with Horizon2020 here https://ec.europa.eu/info/research-and-innovation/statistics/framework-programme-facts-and-figures/horizon-2020-country-profiles_e