SC24 Intel Theater Sessions – Accelerating HPC and AI with Open Software

At SC24, Intel hosted a variety of compelling theater talks by ecosystem partners showcasing their use of open standards-based software and modern code – featuring oneAPI and SYCL* – to accelerate the most demanding workloads across CPUs and GPUs from Intel and other hardware vendors.

University of Bristol – The Role of Open Standard Programming Models for HPC

HPC applications need to address the “Three Ps”: Performance, Portability, and Productivity. Dr. Tom Deakin, Senior Lecturer in Advanced Computer Systems, University of Bristol (UK), highlights the opportunities the variety of heterogeneous architectures brings to applications, and how application performance and portability can be rigorously measured and compared across diverse architectures. He shares strategies for writing performance portable applications and presents the roles that ISO languages C++ and Fortran, as well as parallel programming models and abstractions such as OpenMP, SYCL, and Kokkos, play in the ever-changing heterogeneous landscape. Developers will gain an increased understanding of the need to adopt open standard parallel programming models and embrace heterogeneity in their programs in order to attain the best performance.

Fujitsu – FUJITSU-MONAKA: Powering AI Workloads with oneDNN, oneDAL, and OpenBLAS Optimization

Fujitsu aims to deliver a high-performance, energy-efficient Arm-based 2nm CPU for data centers named FUJITSU-MONAKA. This chip will support an open source AI software stack through enablement and tuning of various AI-based frameworks using multiple platform-agnostic software stacks. Dr. Priyanka Sharma, Director of Software Engineering at Fujitsu Research of India and Head of the MONAKA Software R&D Unit, discusses how Fujitsu is collaborating with the Unified Acceleration (UXL) Foundation to accelerate AI workloads on Arm CPUs by utilizing oneDNN and oneDAL libraries, as well as optimizing OpenBLAS math kernels to achieve energy efficiency and sustainability in data centers.

TU Munich and University of Tennessee – How the oneAPI Library Ginkgo Brings Fluid Flow Simulations to Intel® Data Center GPUs

The increasing diversity in the hardware landscape presents a challenge for simulation software to be portable across different systems and distinct processor technologies. In this talk, Dr. Hartwig Anzt, Professor, TU Munich and the University of Tennessee, demonstrates how Ginkgo, a math library that is a popular numerical backend for nekRS and OpenFOAM fluid flow simulations, tackles the portability challenge and relies on oneAPI and SYCL to enable platform portability across a wide range of architectures. He shares how fluid flow simulations can run on Intel® Data Center GPUs by relying on Ginkgo as a oneAPI-ready software product.

Argonne National Laboratory – Porting the ExaStar Multi-Physics Applications Flash-X and Thornado on the Aurora Supercomputer

The Aurora supercomputer, hosted at the Argonne Leadership Computing Facility (ALCF), has pioneered Intel® Data Center GPU Max Series adoption. Mathialakan Thavappiragasam, Postdoctoral Appointee, Argonne National Laboratory, discusses the challenges of porting scientific codes to new heterogeneous systems, with a focus on offloading Fortran- and OpenMP-based multi-physics code bases for simulation of stellar explosions, and demonstrates optimization strategies that achieved over 23x performance improvements.

University of Utah – The Role of SYCL in the “Machine-Learning Assisted Gigantic Image Cancer Margin Scanner” Project

MAGIC-SCAN (“Machine-Learning Assisted Gigantic Image Cancer Margin Scanner”) is an ambitious and innovative project integrating advanced microscopy, automation, and machine learning to tackle the long-standing issue of incomplete tumor removal in cancer surgery. It aims to be the world’s fastest high-resolution tissue scanner, enabling surgeons to locate residual cancer cells within the operating room. In this presentation, Dr. Valerio Pascucci, Professor, Scientific Computing and Imaging Institute, and Kahlert School of Computing, University of Utah, and doctoral student Alper Sahistan, discuss how they address this challenge by using data management and processing techniques developed over almost two decades of high-performance computing research – including acceleration of ZFP data compression on GPUs through SYCL to manage the images – and novel strategies for lightning-fast image processing and ML inference at the edge for effective deployment in the operating room of a rural hospital.

Leibniz Supercomputing Centre (LRZ) – Application Offloading with oneAPI

LRZ’s SuperMUC-NG, one the world’s largest CPU-based supercomputing systems, recently added Intel® Data Center GPU Max Series processors. Dr. Gerald Mathias, Group Lead Application Support, presents examples of how his team ported and optimized DPEcho and GROMACS codes to the new GPU architectures using oneAPI with SYCL.* DPEcho is a C++ reimplementation of the Echo astrophysical magneto-hydrodynamics code. GROMACS is a molecular dynamics package used for simulating biochemical molecules like proteins, lipids, and nucleic acids. Dr. Mathias discusses the porting of these codes and relative performance on GPU systems from multiple vendors.

See the full SC24 playlist on the Intel Software YouTube channel to view recordings of these select SC24 theater sessions and more examples of open, standards-based oneAPI for HPC and AI.

Learn more about open, standards-based, multiarchitecture programming at intel.com/oneAPI.