Birds of a Feather Archive

Watts, FLOPs, and Impact: Rethinking Energy in the Exascale Era


Authors: Sameh Abdulah (King Abdullah University of Science and Technology (KAUST)), Bilel Hadri (King Abdullah University of Science and Technology (KAUST)), Natalie Bates (Lawrence Berkeley National Laboratory (LBNL)), Steven Martin (HPE), Siddhartha Jana (Intel Corporation, Energy Efficient High Performance Computing Working Group (EE HPC WG)), Mathis Bode (Forschungszentrum Jülich), Tapasya Patki (Lawrence Livermore National Laboratory (LLNL))

Abstract: Energy consumption has become a key constraint in high performance computing (HPC), shaping system design, procurement, software, and scientific workflows in the exascale era. As large-scale simulations and AI-driven workloads increase power demand, the community must reconsider how performance is measured, optimized, and reported beyond FLOPs alone. This BoF session will convene researchers, facility operators, vendors, and application scientists to discuss energy measurement, performance-per-watt metrics, efficient algorithms and runtimes, workload scheduling, and carbon-aware computing. The session will identify practical gaps, share lessons from production systems, and define community priorities for sustainable, energy-efficient HPC at scale.

Long Description: Energy consumption is now a key defining constraint for HPC. In the exascale era, progress can no longer be judged by peak floating-point performance alone. Power envelopes, cooling capacity, operating cost, reliability, and sustainability commitments increasingly shape system design, procurement decisions, software strategy, and day-to-day operations. At the same time, large-scale simulations, data-intensive workflows, and AI-driven methods are expanding demand for both computation and energy. As a result, the community must rethink how performance, productivity, and impact are measured and optimized in practice. This shift demands coordinated innovation across hardware, software, facilities, metrics, and scientific workflows to sustain meaningful progress in the future.

Recent trends in TOP500 and Green500 highlight this urgency. In June 2026, five systems reached exascale-class HPL performance, led by LineShine at 2.198 Exaflop/s, followed by El Capitan, Frontier, Aurora, and JUPITER Booster. During benchmark operation, these systems collectively drew approximately 151 MW. Aggregate TOP500 performance continued to rise. These data underscore a central reality: exascale is both a performance milestone and an energy and infrastructure challenge.

This BoF session: "Watts, FLOPs, and Impact: Rethinking Energy in the Exascale Era", will provide a focused, vendor-neutral, community-driven forum on energy-aware HPC. The discussion will bring together facility operators, researchers, system architects, vendors, runtime and tools developers, and application scientists to examine how energy considerations are reshaping the full computing stack. The goal is to move beyond headline FLOPs and toward practical, comparable methods for evaluating scientific output relative to energy use.

A major theme is measurement. Today, energy data are available through hardware counters, node sensors, platform telemetry frameworks, and facility-level instrumentation, yet these sources vary in fidelity, granularity, and accessibility across systems. This inconsistency complicates reproducibility and cross-platform comparisons and limits informed decision-making by users, centers, and funders. Participants will discuss what to measure, where to measure it, when to sample, and how to report it so that power and energy data are transparent, reproducible, and actionable.

A second theme is methodology. The session will examine efficiency metrics beyond peak benchmarks, including workload-level performance per watt, scientific output per joule, and time-to-solution under power constraints. We will discuss how to compare results across heterogeneous architectures without obscuring workload differences, and how to report uncertainty, normalization choices, and measurement assumptions.

A third theme is action across software and operations. Topics include energy-efficient algorithms, mixed-precision methods, adaptive runtimes, communication and memory optimization, and scheduler policies that account for power limits, utilization, and thermal behavior. The BoF will also address energy-aware and carbon-aware practices, including workload steering based on energy availability, carbon intensity, and facility priorities.

The format is intentionally interactive: short framing presentations followed by moderated discussion, open Q\&A, and audience-input activities. No more than half of the session time will be spent on prepared presentations. Expected outcomes include a prioritized list of gaps in tools and standards, draft recommendations for reporting and comparability, shared lessons from production systems, and a concrete set of community next steps, including opportunities for cross-institutional collaboration and follow-on coordination after SC26.


Back to Birds of a Feather Archive Listing