The session will provide a broad overview of the three regional strategies and roadmaps towards the convergence of HPC and AI.

The session will provide a deeper overview of the regional strategies by displaying use cases.

• Japan: TBD
• US: TBD
• EU: IA Factories and Giga Factories

Session overview – Use cases

This session aims to present regional and international use cases of interest to explore expectations, design and effective implementation of digital continuum components.

Among the use cases:

• The Square Kilometer Array (SKA) is a workflow-intensive project that would benefit significantly from operating across multiple infrastructures. Documentation is available for this use case

Session overview – Continuum Digital Twin (CDT)

In addition to the use-case, the session will explore the concept of a “Continuum Digital Twin.” (CDT) This digital twin could abstract real systems to support cross-facility workflows without requiring direct interaction with physical infrastructure. By sharing only essential information needed for orchestration, the digital twin safeguards sensitive aspects of the real infrastructure. Real-time data flows from the infrastructure to the digital twin, ensuring an efficient and secure data exchange. A similar idea has been developed in the paper “Digital Twin Continuum: a Key Enabler for Pervasive Cyber-Physical Environments”.

In particular, this session will explore the idea of using the CDT concept for optimizing infrastructure (compute + storage + network) allocation and usage, especially from a sustainability point of view.

See Workflows Community Summit 2024,  Future Trends and Challenges in Scientific Workflows – https://arxiv.org/abs/2410.14943

https://ieeexplore.ieee.org/document/10637565

Session overview

Exascale and Post-Exascale applications are becoming increasingly difficult to build, deploy and maintain under the double pressure of the growing machine complexity and the applications’ needs to combine multiple compute and data processing paradigms (HPC, HPDA & AI). To address these challenges, our community needs to foster HPC dev-ops methodologies and tools to enhance productivity and improve interoperability, functionality and performance portability, as well as reproducibility

Session objectives

Topics of discussion cover all HPC dev-ops related subjects, including (but are not limited to):

• Synergetic user/administrator software build & deployment for HPC
• Package managers for HPC
• Containerization & virtualization at exascale
• Software binary repositories and caches
• CI/CD/CB at exascale
• Everything-as-code for HPC
• Cloudification of the supercomputers
• AI: new software stacks and practices
• Shared HPC software metadata and catalogs (softwareheritage.org)
• Methodologies and tools for reproducibility in HPC
• HPC software sustainability: funding, communities, foundations, …

Session overview

In this session, we want to explore the increasing importance of rapidly evolving AI-driven and AI-coupled HPC/HPDA workflows in computational science and engineering applications.

These workflows typically involve the concurrent, real-time coupled execution of AI and HPC/HPDA tasks in ways that allow the AI systems to steer or inform the HPC/HPDA tasks and vice versa.

Enhancing the entire computing chain—from AI methods to workflow implementation and orchestration—requires the development of common collaborative benchmarks to allow fast progress and evaluation.

The shared benchmark approach has typically demonstrated its usefulness in deep learning with ImageNet or the BLUE benchmark, which allowed the identification of critical technology like ConvNets or Transformers and stirred the international community toward relevant research directions.

Adapting this methodology to drive progress in the integration of AI into HPC/HPDA software and application developments would allow overcoming traditional limitations and lead to significant effective performance enhancement — measured by science discovery for a given amount of computing — on different computing architectures while also testing various capability and capacity metrics such as accuracy and scalability.

These benchmarks should be based on shared insights from different coupling modes for AI and HPC/HPDA workflows and on the identification of execution motifs most commonly found in scientific applications with well-defined data and comparison metrics.

Adopting standardized practices and tools at the international level is essential for creating a consistent and collaborative framework and accelerating progress

Session objectives

• Propose/develop a common conceptual basis for understanding AI-driven HPC/HPDA workflows
• Identify and analyze different modes of coupling AI into HPC/HPDA workflows
• Identify execution motifs most commonly found in scientific applications, allowing to analyze the primary performance challenges underpinning AI-driven HPC/HPDA workflows
• Define shared and collaborative proxy-apps and benchmark exercises
• Develop standardized and specific benchmarks, including data/simulators and validation metrics,
• Address different modes of coupling AI into HPC/HPDA workflows
• Address coexistence and communication between HPC and AI tasks in the same workflow
• Plan a Development Timeline with Scheduled Meetings
• Create a roadmap with scheduled meetings (e.g., ISC25, SC25, InPEx26) to ensure timely progress and collaboration

Example use-cases may include:

• AI-Based Image Reconstruction pipelines
• In-situ AI-driven adaptive HPC simulations
• AI-based steering ensemble of simulations
• AI-based multi-stage HPC/HPDA pipelines
• AI-based robust inference methods

Adaptive execution for training large AI models

Session overview

This session will explore the transformative role of Generative AI in scientific research.  Key themes include AI-powered hypothesis generation, data augmentation and simulation, and scientific writing and communication with LLMs. Discussions will explore synthetic data generation, discuss models like AlphaFold, GNoME, and FourCastNet, reproducibility and reliability, etc.