Workshop In conjunction with CARLA 2025
Energy Efficiency and Sustainability in AI, HPC, and Quantum Computing
A Workshop on Sustainable Computing Technologies
Date
September, 23rd
Location
The Jamaica Pegasus Hotel
81 Knutsford Blvd, Kingston 5, Jamaica
Co-located with CARLA 2025
VENUE
Grand Jamaica (Negril) Suite
Workshop Agenda
Time
Title
09:00 - 09:10
Welcome and Introduction
09:50 - 10:10
Paper 1: Analyzing Reproducibility Gaps in HPC AI Workloads (Rojas Yepes, P., Barrios Hernandez, C., Carrillo, O., Le Mouël, F.)
10:10 - 10:30
Paper 2: MERIC: Complex solution for datacenter power monitoring (Vysocky, O., Riha, L., Ptacek, P., Faltynkova, M., Jaros, M., Velicka, D.)
10:30 - 10:50
Paper 3: Scalable Memory Management for Hybrid Quantum HPC (Díaz, G.)
11:00 - 11:30
Coffee Break
13:00 - 14:00
Lunch Break
15:10 - 15:40
Industrial Keynote
AMD
16:00 - 16:30
Coffee Break
17:10 - 17:45
Panel Discussion – Energy Efficiency in Latin American HPC
Panelists: John DeSantis, Lubomir Riha, Carla Osthoff, Dieter Kranzlmüller
17:45 - 18:00
Closing Remarks
Keynote Talk Abstracts
Observed power usage between ARM and x86 nodes
John DeSantis – Texas Advanced Computing Center (TACC)
This presentation examines real-world power consumption observations from TACC’s experience deploying and operating ARM-based supercomputer systems. Through direct comparison with traditional x86-based HPC systems, we present measured power usage data, performance per watt metrics, and operational insights gained from managing both architectures in production environments. The talk will cover practical considerations for system administrators, including cooling requirements, power delivery challenges, and the implications for sustainable HPC operations at scale.
A software approach for energy-efficient HPC
Lubomir Riha – IT4Innovations National Supercomputing Center
This talk will address selected aspects of software-based methodologies aimed at improving the energy efficiency of high-performance computing (HPC) infrastructures executing parallel scientific applications. Particular attention will be given to the energy-aware runtime suite MERIC, developed at IT4Innovations, which encompasses system-level monitoring and control, job-level optimization techniques, and advanced 3D visualizations of supercomputer state. Furthermore, the presentation will discuss the European perspective on the development of a holistic, energy-aware software stack.
Energy Efficiency as a Key to Sustainable HPC and AI
Dieter Kranzlmüller – Leibniz Supercomputing Centre (LRZ) and LMU Munich
The rapid advances in high-performance computing (HPC) and artificial intelligence (AI) are driving unprecedented demand for computational power—and with it, energy consumption. As these technologies mature, energy efficiency and sustainability have become defining challenges for the future of digital infrastructure. This talk will explore strategies for reducing the carbon footprint of advanced computing, highlighting the role of innovative cooling concepts, energy-aware system design, and the integration of renewable energy sources. Special attention will be given to hot-water cooling, which was pioneered at the Leibniz Supercomputing Centre (LRZ) in 2012. How can new infrastructures benefit from lessons learned and what can we expect in the future. By combining perspectives from research, policy, and industry, we will discuss pathways toward a sustainable computing ecosystem that balances performance, cost, and environmental responsibility.
Implementing machine learning-based techniques for energy-efficient resource allocation in Supercomputing environments
Carla Osthoff – National Laboratory for Scientific Computing (LNCC), Brazil
Energy consumption in High Performance Processing (HPP) systems has become an increasingly relevant challenge given the increasing computational demand and the execution of large-scale scientific applications. Factors related to resource allocation, such as the number of nodes (#N) and the number of threads per node (#T), directly impact both execution time and energy efficiency, often assessed through the Energy-Delay Product (EDP). In this presentation, we present a methodology based on the Extra Trees Regressor to recommend hardware resource allocations that minimize the EDP in parallel applications, without the need to execute the application or instrument its code.
Next-Gen Data Center for the Age of AI Reasoning
Pedro Mário (NVIDIA) – Solution Architect for Higher-Education and Research in Latin America
In this talk, we explore how NVIDIA’s full-stack platform is revolutionizing energy efficiency in next-generation HPC datacenters through innovations spanning hardware, interconnects, cooling, and AI. At the core are the Blackwell GPU architecture and Grace ARM-based CPU, delivering exceptional performance-per-watt via heterogeneous compute and advanced power management. NVLINK chip-to-chip communication and silicon photonics drastically reduce energy overhead in data movement, while liquid-cooled systems improve thermal efficiency and datacenter sustainability. We also highlight the role of floating-point emulation with low-bit representations, enabling traditional HPC applications to maintain computational accuracy while significantly lowering power consumption. Finally, NVIDIA’s NEMOTRON LLM models bring AI-driven optimization to datacenter operations, enhancing workload scheduling and predictive maintenance for smarter, greener infrastructure.
Quantum Computing and Sustainability: Where is the Real Advantage?
Francesco Nappi – IQM Quantum Computers
Quantum computing is expected to revolutionize many sectors of research and industry. It is also seen as a promising path toward more sustainable computing. In this work, we compare the energy consumption of a high-performance computing center and a quantum computer, and we present use cases and algorithms that might provide real benefits in the quest for sustainable development.