Tutorials

PortFastRunFast

It has never been easier to port applications to GPUs and run fast as with the AMD MI300A GPU system. The San Diego Supercomputer Center is deploying Cosmos, the first MI300A system sponsored by the US National Science Foundation. There are tremendous benefits to scientific applications exploiting the computational resources of the latest GPU architectures. However, real-world application  codes can be challenging to port to GPUs due to their complexity, being often composed of tens of thousands of lines spread over many components and with many external dependencies. This hands-on tutorial will show how the MI300A APU programming model eases the porting, reducing the time to solution by running as much as possible on the GPU by Exploiting the MI300A APU single memory architecture.

Julia for High-Performance Computing

We propose a 3-hour, hands-on tutorial on the Julia language for high-performance computing applications. Julia empowers HPC users providing a high-productivity and HPC ecosystem powered by just-in-time (JIT) compilation via LLVM. Attendees will access The National Energy Research Scientific Computing Center resources to run end-to-end applications using the Gray-Scott reaction-diffusion solver. We will show Julia's support for CPU/GPU performance portable models (KernelAbstractions, JACC) powered by vendor GPU back end (e.g. NVIDIA, AMD), and distributed memory parallelism using its message passing interface wrapper (MPI.jl). Codes will be publicly available, while we will follow up with support and computing access after the tutorial.

Introduction to Performance Portable HPC Development with Spack, Kokkos, and Trilinos

This 4-day tutorial provides a comprehensive introduction to tools and libraries essential for developing HPC applications. Day 1 covers Spack, a package manager for wrangling the complexities of HPC software libraries across platforms. Days 2-3 delve into Kokkos, a performance portability abstraction layer that enables writing single-source C++ applications targeting CPU and GPU architectures. The final half-day covers Trilinos, a collection of numerical solvers for large-scale scientific and engineering problems. Join us to gain practical experience developing HPC software using powerful open-source tools, which will equip you with the skills to develop and analyze performant applications on modern supercomputing architectures.

AI with the AMD software stack: from zero to hero

Artificial Intelligence (AI) is revolutionizing various industries by boosting efficiency, automating processes, and enhancing decision-making. Additionally, it plays a key role in scientific advancements by tackling complex challenges. However, many users are not familiar with how seamless and accessible the AMD ecosystem is for running various AI models. In this hands-on tutorial, we will explore the AMD AI ecosystem and how effortlessly models can be ported to AMD hardware. We will demonstrate running PyTorch, TensorFlow, and other frameworks on AMD hardware—without modifications to the original model code. We will provide insights on the latest models, such as DeepSeek and discuss general tuning techniques to optimize performance.