The increasing demands of computational chemistry, molecular simulations, and scientific computing—especially in atomistic simulations—have made energy consumption and its environmental impact critical concerns. Historically, computational cost was dominated by processing resources, but in recent years, electricity usage and its role in global warming have taken center stage. This shift highlights the urgent need to revisit modeling and simulation techniques, evaluate the energy efficiency of programming languages, and optimize hardware to reduce energy consumption per computation.

Sustainability in scientific computing is not just a general issue—it permeates every aspect of computational workflows. From model development and algorithm design to software implementation and hardware optimization, each step is interconnected. Inefficiencies in one area can cascade across the system, amplifying energy costs and reducing overall efficiency. Addressing these interconnected challenges requires collaboration across disciplines to share experiences, identify common obstacles, and develop holistic solutions.

Scientific Context

This workshop aims to bring together professionals from diverse domains to share insights and foster innovation. Each perspective contributes unique challenges and solutions:

• Model Development: How can models be designed to balance accuracy, scalability, and efficiency?
• Algorithm Design: What optimizations are required to make algorithms both energy-efficient and high-performing?
• Programming Languages: How do languages like Fortran, C++, and Python compare in terms of energy efficiency?
• Software Development: What strategies can optimize memory usage, data movement, and runtime in both open-source and commercial tools?
• Hardware Optimization: How can advancements in GPUs, CPUs, and HPC systems be harnessed to improve energy efficiency?
• Cloud Computing Platforms: How can cloud infrastructure (Azure, AWS, Google Cloud) be leveraged for sustainable scientific computing while managing energy consumption and costs?
• Machine Learning Integration: How can ML techniques enhance computational efficiency while addressing their own energy demands?

By fostering dialogue across these areas, the workshop will create a collaborative environment for sharing best practices and driving innovation. This multi-perspective approach aims to develop solutions that not only improve sustainability within individual categories but also ensure synergy across the entire computational pipeline.

Workshop Objectives

1. Facilitate Cross-Disciplinary Optimization Strategies Enable the exchange of energy-efficient optimization techniques among programming language developers (e.g., C++, Fortran, Python) and scientific software creators to enhance computational performance.
2. Integrate Software and Hardware Efficiency Explore innovative methods to align software with advanced hardware capabilities (CPU, GPU, HPC) to maximize energy efficiency while maintaining high performance.
3. Develop Energy-Efficient Algorithms Advance the design and optimization of algorithms for computational chemistry and molecular simulations, prioritizing scalability and power efficiency.
4. Optimize GPU Utilization Identify strategies for leveraging GPU-accelerated hardware, including GPU-resident computations, vector processing, and balancing GPU-CPU usage for improved energy efficiency.
5. Leverage Cloud Computing for Sustainable Science Explore best practices for utilizing cloud platforms (Azure, AWS, Google Cloud) for scientific computing, focusing on energy-aware resource allocation, cost optimization, and hybrid cloud-HPC workflows.
6. Develop Hybrid and Coarse-Grained Models Address the creation and application of hybrid and coarse-grained models to enable simulations on longer timescales or larger system sizes, leveraging the strengths of multiple modeling scales.
7. Integrate Machine Learning for Enhanced Efficiency Explore the integration of machine learning in scientific computing to enhance simulation accuracy and efficiency, including ML-driven force fields, active learning, Gaussian process models, and neural network potentials, while addressing their energy demands and computational costs.
8. Establish Best Practices for Sustainable Computing Promote actionable practices, such as minimizing memory access, optimizing data locality, reducing internode messaging, and enhancing local storage utilization for scalable algorithms.
9. Advocate for Energy-Aware HPC Resource Allocation Address policies for sustainable HPC resource allocation, incentivizing energy-efficient practices and optimizing resource usage across traditional HPC and cloud platforms.
10. Foster Interdisciplinary Collaboration Build bridges among language developers, algorithm designers, hardware engineers, cloud computing specialists, and software developers (both open-source and commercial) to collectively address sustainability challenges.
11. Support Early-Career Researchers Provide networking and mentorship opportunities for early-career researchers to develop their skills, share their work, and connect with established experts in sustainable computing.
12. Other Sustainability Advocates in Scientific Computing Anyone working in computational science, software development, or hardware design with an interest in sustainability and energy-efficient computing, bringing diverse perspectives and innovative ideas to the discussion.

Expected Outcomes

• Actionable Strategies for Energy-Efficient Software Design Develop practical approaches to reduce energy consumption in scientific software by optimizing algorithms for computational chemistry, molecular simulations, and scalable parallel computing.
• Comprehensive Guidelines for GPU-Centric Algorithm Development Provide clear, implementable recommendations for maximizing GPU capabilities, including GPU-resident computation, efficient vector processing, and balancing GPU and CPU utilization to enhance power efficiency.
• Best Practices for Cloud and Hybrid Computing Establish guidelines for sustainable use of cloud platforms (Azure, AWS, Google Cloud) in scientific computing, including cost-energy trade-offs, workload optimization, and integration with traditional HPC infrastructure.
• Framework for Machine Learning in Sustainable Computing Develop strategies for integrating ML techniques (neural networks, active learning, data-driven models) to enhance computational efficiency while managing their energy footprint.
• Best Practices for Sustainable Scientific Computing Publish a detailed best practices guide, covering key topics such as algorithmic optimization, hardware utilization, memory access efficiency, and data movement strategies to improve sustainability in computational science.
• Framework for Assessing Power Efficiency Develop a robust and standardized framework to evaluate and compare the energy efficiency of scientific software, enabling developers to identify and implement meaningful energy-saving improvements.
• Policy Recommendations for Energy-Aware Resource Allocation Advocate for HPC centers and cloud providers to adopt sustainable resource allocation policies, including incentives for energy-efficient usage and strategies for optimizing computational resources.
• Enhanced Industry-Academia Collaborations Facilitate stronger partnerships between industry and academia to foster shared innovation and develop cutting-edge solutions for sustainable computing.
• Career Development Opportunities for Early-Career Researchers Offer networking, mentoring, and collaboration opportunities for early-career scientists to enhance their skills, gain exposure to interdisciplinary approaches, and build professional connections in the field.
• Establishment of Ongoing Interdisciplinary Working Groups Create dedicated working groups to continue collaboration, monitor progress, and refine strategies for sustainable computing beyond the workshop.
• Integration of Hybrid Modeling and Machine Learning Techniques Encourage the adoption of hybrid models and coarse-grained approaches for simulations, as well as the use of machine learning to enhance computational efficiency and reduce resource requirements.

This workshop will create an opportunity for industry and academia to address sustainability challenges in scientific computing collaboratively. By leveraging diverse expertise, it aims to drive innovation and set the foundation for a sustainable future in computational research.