About this Research Topic
Traditionally, the speed and capacity of supercomputers have more or less doubled every second year by squeezing twice as many transistors onto the same chip area. This development, characterized by Moore’s law, has for some years been held back by physical barriers, which have caused new and innovative approaches to be developed to meet the rapidly growing need for more computer power. This accelerating need is, to a large degree, fuelled by the data deluge arising in a growingly instrumented world and the advances in data analytics, machine learning (ML), and artificial intelligence (AI) to analyze it. The result is an increasing level of heterogeneity and complexity in high-end servers and supercomputers, not only in terms of the combination of processors like CPUs and GPUs, but also in memory management, data communication, and solution methods. This development is fundamental to reach and go beyond the exascale barrier in supercomputing.
While current hardware development is to a large degree motivated by challenging applications in need of efficient AI/ML computation, for instance by adding processors that are designed bottom up for such algorithms, models of physics-governed systems are of more importance than ever. In particular, it is important to understand how heterogeneous hardware with accelerators designed for specific workloads like deep learning can also speed up more traditional HPC workloads, such as large-scale simulations of systems governed by laws of physics. This is essential to be able to enrich the models with the details needed to identify important physical effects, especially in multi-physics settings, and to move advanced models into practical use by speeding up computation.
This Research Topic is designed to provide researchers and practitioners dealing with HPC implementations of physics-based models with sound advice on how to realize high performance from current and future heterogeneous computing resources for their applications. Such advice may cover programming strategies, software tools, algorithmic designs and real-world applications.
In this Research Topic we highlight how physics-based applications and software frameworks can take advantage of the rapidly increasing computational power that is currently offered to the market or expected in the near future. In particular, we welcome contributions in or connected to the areas listed below:
• Application of HPC for physics-based models on heterogeneous computer systems
• Software frameworks for heterogeneous computer systems (must include examples relevant for physics-based models)
• Use of domain-specific processors for physics-based models
• Application of HPC in hybrid methods combining physics-based and data-driven modeling paradigms
We welcome the following article types: Original Research, Methods, Review, and Technology and Code.
Topic Editor Prof. Mark Parsons is Director of UoE HPCx Ltd and Director and Charitable Trustee of Research Data Scotland. All other Topic Editors declare no competing interests with regards to the Research Topic subject.
While current hardware development is to a large degree motivated by challenging applications in need of efficient AI/ML computation, for instance by adding processors that are designed bottom up for such algorithms, models of physics-governed systems are of more importance than ever. In particular, it is important to understand how heterogeneous hardware with accelerators designed for specific workloads like deep learning can also speed up more traditional HPC workloads, such as large-scale simulations of systems governed by laws of physics. This is essential to be able to enrich the models with the details needed to identify important physical effects, especially in multi-physics settings, and to move advanced models into practical use by speeding up computation.
This Research Topic is designed to provide researchers and practitioners dealing with HPC implementations of physics-based models with sound advice on how to realize high performance from current and future heterogeneous computing resources for their applications. Such advice may cover programming strategies, software tools, algorithmic designs and real-world applications.
In this Research Topic we highlight how physics-based applications and software frameworks can take advantage of the rapidly increasing computational power that is currently offered to the market or expected in the near future. In particular, we welcome contributions in or connected to the areas listed below:
• Application of HPC for physics-based models on heterogeneous computer systems
• Software frameworks for heterogeneous computer systems (must include examples relevant for physics-based models)
• Use of domain-specific processors for physics-based models
• Application of HPC in hybrid methods combining physics-based and data-driven modeling paradigms
We welcome the following article types: Original Research, Methods, Review, and Technology and Code.
Topic Editor Prof. Mark Parsons is Director of UoE HPCx Ltd and Director and Charitable Trustee of Research Data Scotland. All other Topic Editors declare no competing interests with regards to the Research Topic subject.
Keywords: HPC, heterogeneous computing, physics-based models, GPU acceleration, domain-specific processors, software frameworks, applications
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.
