This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

The demand for high-performance computing (HPC) is expanding as scientific and engineering challenges grow increasingly complex. Traditional CPU-based architectures, while versatile, often struggle to efficiently handle specialised workloads such as machine learning, computational fluid dynamics (CFD), and molecular dynamics. To address this, the integration of domain-specific hardware accelerators like NVIDIA GPUs and ARM chips has emerged as a game-changer, enabling unparalleled performance and efficiency for targeted applications. This paper delves into the role of domain-specific hardware accelerators in revolutionising scientific workflows. We focus on how specialised architectures, available on Rescale'™s intelligent cloud HPC platform, empower researchers and engineers to leverage cutting-edge hardware tailored to their workloads. By combining NVIDIA GPUs for compute-intensive tasks with ARM-based architectures for energy-efficient operations, users can achieve optimal performance while addressing cost and sustainability goals. A critical aspect of this discussion involves the optimisation of workflows for hybrid and heterogeneous computing environments. Integrating domain-specific accelerators requires not only hardware availability but also seamless software orchestration to manage data flows, scheduling, and execution. The platform addresses these challenges by offering a unified environment where users can dynamically select the most suitable hardware configurations based on workload requirements. This flexibility is particularly impactful for industries ranging from aerospace to pharmaceuticals, where precision and efficiency are paramount. The paper also explores practical use cases, such as leveraging NVIDIA GPUs for AI-driven simulation post-processing or ARM-based chips for low-power, high-throughput scenarios. These examples illustrate how hardware accelerators can drastically reduce time-to-solution and cost, enabling organisations to push the boundaries of innovation. Additionally, we will discuss the importance of workload profiling and benchmarking to ensure optimal hardware utilisation, drawing on real-world insights. Attendees will gain an understanding of the technical and operational considerations involved in adopting domain-specific hardware, from software compatibility to deployment strategies in cloud-based environments. We will highlight how to simplify these complexities, allowing users to focus on their core scientific and engineering objectives. Through this exploration of domain-specific accelerators, we aim to demonstrate how specialised architectures are not just advancing performance but also reshaping how organisations approach HPC, paving the way for new breakthroughs in research and industry.