Introduction to the ROCm™ Ecosystem
Mario Ruiz-Noguera (AMD) · 4 hours
A technical and hands-on introduction to AMD’s open-source ROCm™ software stack and the HIP (Heterogeneous-Compute Interface for Portability) programming model. Topics include GPU vs. CPU architecture, PyTorch on ROCm, GPU kernel development, memory management, performance optimization through shared memory, streams and synchronization, and CUDA-to-HIP portability. Remote compute resources and all required software are provided; no prior GPU programming experience is needed.
Agentic AI Design with AMD Technology
Mario Ruiz-Noguera (AMD) · 4 hours
A hands-on exploration of the evolution from generative AI to autonomous AI agents. Participants will learn the core building blocks of agentic systems — models, orchestration, memory, and tool integration — as well as design principles for trustworthy agent-based applications. The tutorial covers Retrieval-Augmented Generation (RAG), modern agent architectures, and emerging open standards for agent interoperability, with practical labs on AMD’s high-performance computing platforms.
Principles and Practice of Scalable and Distributed AI Training and Inference
Dhabaleswar K. (DK) Panda, Nawras Alnaasan (The Ohio State University) · 4 hours
An end-to-end treatment of the AI model lifecycle — training, fine-tuning, and inference — on modern HPC clusters. The tutorial examines parallelization strategies (data, tensor, pipeline, and expert parallelism) for large-scale models such as GPT, LLaMA, DeepSeek, and ViT, and covers distributed middleware including PyTorch, Megatron-LM, DeepSpeed, and communication backends (NCCL, MPI). Hands-on labs provide practical experience with distributed training and inference on GPU clusters.
High-Performance and Smart Networking Technologies for HPC and AI
Dhabaleswar K. (DK) Panda, Benjamin Michalowicz (The Ohio State University) · 4 hours
A comprehensive overview of the interconnect landscape powering modern HPC and AI systems: InfiniBand, RoCE, Omni-Path, Slingshot, Ultra-Ethernet, NVLink, AMD Infinity Fabric, CXL, and SmartNICs/DPUs. The tutorial examines how these technologies integrate through software stacks (UCX, Libfabrics), GPU-aware communication (GPUDirect RDMA), and AI-specific hardware (Cerebras, Microsoft MAIA, Meta MTIA). Three hands-on exercises on an InfiniBand cluster illustrate real-world communication trade-offs.
Managing HPC Software Complexity with Spack
Caetano Melone, Todd Gamblin (Lawrence Livermore National Laboratory) · Fernando Posada (Oak Ridge National Laboratory) · 8 hours (2 × 4 h)
Spack is the open-source package manager that has become a de facto standard for HPC software deployment, used in production on 5 of the current TOP500 top-10 systems — including the exascale systems El Capitan, Frontier, and Aurora. This full-day tutorial covers Spack’s core capabilities: installing and querying packages, authoring custom package recipes, managing reproducible environments, configuring for specific hardware, and deploying optimized software stacks at HPC facilities. More than 80% of the session is hands-on, using cloud-hosted VMs pre-loaded with Spack binary packages.
Introduction to Quantum Machine Learning on Quantum Computers
Gleydson Fernandes de Jesus (SENAI CIMATEC) · 4 hours
An introduction to quantum machine learning (QML) that bridges quantum computing fundamentals and machine learning concepts. Topics include qubits, quantum gates, quantum circuits, and hybrid quantum-classical learning workflows. Participants will learn how QML models are built, how they differ from classical approaches, and how to execute experiments on cloud-based quantum platforms and simulators. Access to SENAI CIMATEC’s Kuatomu supercomputer is planned for enrolled students.
Development of Parallel Applications with the Task-Oriented Paradigm in the StarPU Runtime
Lucas M. Schnorr (UFRGS) · Vinicius Garcia Pinto (FURG) · Raymond Namyst, Samuel Thibault (INRIA / University of Bordeaux) · 8 hours (2 × 4 h)
An advanced tutorial on the task-based parallel programming paradigm and the StarPU runtime, designed for heterogeneous HPC systems with multicore CPUs and GPU accelerators. Session 1 covers the StarPU model (Sequential Task Flow, codelets, data handles, DAG construction) with hands-on exercises on matrix computations and data partitioning. Session 2 addresses heterogeneous programming with CUDA/HIP and GPU implementations, followed by performance analysis using the StarVZ trace visualization framework to identify bottlenecks beyond traditional metrics.
Simulating Quantum Algorithms with Q-Team
José David Bañuelos Aquino (Q-Team, México) · 4 hours
A practical introduction to quantum computing and algorithm simulation using the Q-Team platform. The tutorial covers the fundamentals of quantum information theory, qubits, quantum gates (Pauli X/Y/Z, Hadamard, phase shift, controlled gates), and quantum circuits, then guides participants through developing and simulating quantum algorithms in Q-Team’s graphical and programming environments. Ideal for students in mechatronics, electronics, IT, and related fields with a background in linear algebra and combinational logic.