SC4RC 2026
Computing plays a vital role in many areas of research supporting modelling, simulation and data analysis across disciplines ranging from literature and economics, to medicine and astrophysics. Its use has a significant negative impact on the environment, from raw material extraction and processing in production of computing hardware, through energy use in its operation, to its decommissioning and reuse/recycling. The demand for higher capacity and efficacy of hardware and software tools, as well as increasing dataset size, suggests these environmental impacts will continue to grow, unless synergies between sustainable research practices, resource optimisation and impact mitigation are recognised and exploited.
All research communities have a responsibility to limit the environmental impacts of their activities, and there is a pressing need for a coordinated and interdisciplinary focus on computing impacts. this conference aims to facilitate knowledge exchange between data-intensive disciplines, allowing the sharing of best practice, the development of common resources, and the catalysis of new interdisciplinary collaborations. Key topics include, but are not limited to:
- estimation and monitoring of environmental impacts and energy usage of computing,
- responsible use of computing hardware (sustainable procurement, extending and optimising use phase, responsible disposal),
- resource-efficient software,
- carbon-aware scheduling,
- FAIR and effective data curation, storage and sharing,
- addressing the rebound effect,
- community building and training,
- general and field-specific best practices,
- changing user behaviour and incentivising sustainable computing practices (addressing e.g. external drivers).
Key dates
- Early Bird registration discounts have now ended.
- Bursary requests have closed.
- Abstract submissions are closed.
- Notification of decisions: late March 2026.
- Registrations are closed.
Programme
A draft timetable can be found in the left-hand menu.
Confirmed speakers include:
- Ben van Werkhoven (Leiden University)
- Raghavendra Selvan (University of Copenhagen)
- Jo Walton (University of Sussex)
- Wim Vanderbauwhede (University of Glasgow)
- Michael Sparks (University of Manchester)
- Jeremy Cohen (Imperial College London)
- Julia Steinberger (University of Lausanne)
Proceedings
Although there will be no proceedings for this first instance of SC4RC, during the course of the week we will draft a position paper, with conference participants as authors, which we will submit for publication.
Organising Committee
| Anica Araneta | University of Cambridge, UK |
| Auroop Ganguly | Northeastern University, Boston, USA |
| Grant Hill | University of Sheffield, UK |
| Loic Lannelongue (Co-Chair) | University of Cambridge, UK |
| Rakhi Mahbubani (Co-Chair) | Rudjer Boskovic Institute, Croatia |
| Peter Millington | University of Manchester, UK |
| Ayan Paul | Northeastern University, Boston, USA |
| Karolos Potamianos | University of Warwick, UK |
| Markus Schulz | CERN, Switzerland |
You can contact the OC with your comments and suggestions at info@sc4rc.org.
Local Organising Committee
| Catharine Noble | Markus Schulz |
| Kristina Gunne |
Programme Advisory Committee
| Sebastian Buschjager | Aurélie Bugeau |
| Manuel Cubero-Castan | Caterina Doglioni |
| Kristin Lohwasser | Kristy Pringle |
| Charlotte Rae | Joanna Taylor |
| Simon Portegies Zwart |
Sponsors and endorsers
-
-
Welcome & Housekeeping 503/1-001 - Council ChamberConveners: Kristina Gunne (CERN), Loic Lannelongue (University of Cambridge), Markus Schulz (CERN), Rakhi Mahbubani (Rudjer Boskovic Institute (HR))
-
1
Welcome to SC4RC
Welcome by the SC4RC 2026 Organising Committe
Speakers: Loic Lannelongue (University of Cambridge), Rakhi Mahbubani (Rudjer Boskovic Institute (HR)) -
2
Welcome to CERN
Welcome by the CERN-IT Department Head Simone Campana
Speaker: Simone Campana (CERN) - 3
-
1
-
Discipline Summaries 503/1-001 - Council Chamber
Summary talks about sustainable computing efforts in the different disciplines represented at the conference.
-
4
Scientific Computing Across Disciplines: A Student-Driven Evaluation of Institutional Computing Footprints and External Resource Usage
Experimental research increasingly relies on computing to analyze data, model systems, and apply AI, expanding its role across all scientific disciplines. This shift has driven a surge in demand for computational power, alongside higher energy use and material consumption, compounding the already significant environmental footprint of experimental research.
To address this, a student-driven course was developed to assess research-related environmental impacts. Collaborating with a scientific computing team, students analyzed internal infrastructure and external computing use, including AI tools, to estimate carbon and material footprints and compare them with other research activities.
Results show computing is widespread but unevenly used: most researchers consume about 1,000 CPU hours annually, while a small group accounts for over 20% of total energy use. The computing footprint stems mainly from equipment production (57%), followed by services (25%) and storage (12%).
Speaker: Jeroen Dobbelaere (Institute of Science and Technology Austria) -
5
Green computing in neuroimaging research: Identifying strategies to reduce energy consumption, and training end users through roving workshops
We have measured and provided recommendations for reducing the carbon footprint of functional magnetic resonance imaging (fMRI) research. In a review paper, we provided ten recommendations for green neuroimaging computing, from the stage of analysis planning through to data dissemination. In empirical studies, we used carbon tracking tools to estimate the carbon footprint of widely used fMRI data processing pipelines. On the basis of our results, we have made empirical recommendations for neuroimaging research computing to maximise environmental sustainability without compromising data quality. Over the past year, we have disseminated our findings and recommendations via roving workshops, at neuroimaging departments across the UK and virtually at institutions overseas. These have included demonstrations of green computing tools, including digital carbon trackers and climate aware task schedulers. The work discussed here may provide a model through which others can conduct and disseminate green computing research in their own disciplines.
Speaker: Dr Nicholas Souter (University of Sussex) -
6
High Energy Physics, LHC and Computing for LHC [Online]
TBD
Speakers: Caterina Doglioni (University of Manchester, UK), Caterina Doglioni (The University of Manchester (GB)) -
7
Balancing accuracy and computational cost in computational chemistry
As molecular simulations increase in scale, the environmental cost of computational chemistry is becoming harder to ignore. High‑accuracy quantum methods remain too expensive for routine large‑scale use, while empirical force fields trade fidelity for speed. Machine learning potentials (MLPs) offer a middle ground, yet their computational and energy demands remain significant, especially for long timescales and large systems. This reflects a broader challenge in responsible research computing: balancing accuracy, computational efficiency, and environmental impact. We present a strategy to reduce the computational footprint of MLPs while preserving accuracy and stability. Our active‑learning framework minimizes evaluation cost and training‑set size by using kernel density estimation on node energies to identify out‑of‑distribution configurations, avoiding costly ensemble‑based uncertainty quantification. Benchmarking against reactive force fields and foundation MLPs assesses accuracy, speed, memory use and environmental cost. Applied to a hydrated silicate ionic liquid, our tailored MLP delivers major speedups, lower memory use, and reduced energy per simulated picosecond, enabling more sustainable molecular simulations.
Speaker: Jelle Vekeman (University of Antwerp) -
8
Discussion
-
4
-
4:00 PM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
Interactive Session 503/1-001 - Council Chamber
-
9
Power Measurement Techniques for Research Computing
Software and hardware energy optimisation is one of the key levers to reducing the operational environmental impacts of computing activities. To support this, a granular understanding of how computers use power is essential but rarely straightforward. Confounding factors can arise from (1) differences between platforms (e.g. high-performance computing vs laptops), (2) variability in accessible power measurement data, and (3) challenges in assigning power usage to individual computational processes or applications. Stemming from this, a multitude of approaches have been developed for measuring the power usage of computational work.
This workshop will feature an introductory talk giving an overview of different power measurement approaches, the main ones being (1) based on modelling, (2) using performance counters accessible in-silico, and (3) using external power meters. The benefits and drawbacks of each approach will be compared, and specific examples from each will be highlighted. Following this, the floor will be opened up for discussion. With guidance from facilitators, attendees will be encouraged to share any experience they may have with the power measurement of computational work, discuss pros and cons of different approaches in the context of their own work, and identify blockers or opportunities for adoption of different tools.
From this session, we hope to encourage the judicious adoption of existing power measurement tools and promote more consideration of power usage in computational research. Furthermore, we hope to identify any widespread issues or complaints with currently available tools, with the view of informing the development of improved tools addressing these issues.
Speaker: Dr Jack Coker (University of Cambridge)
-
9
-
Interactive Session 500/1-001 - Main Auditorium
-
10
GreenDIGIT Tools for Environmental Sustainability in Digital Research Infrastructures: Assessment Methodology and Metrics Publication System
The GreenDIGIT project has developed a comprehensive environmental sustainability framework for digital Research Infrastructures (RIs). In this interactive session, we introduce and demonstrate two key tools delivered by the project. First, the Environmental Impact Assessment Methodology offers a structured, lifecycle-based framework for evaluating carbon emissions, energy consumption, waste, and water use. It prioritises improvement actions across three dimensions: impact magnitude, likelihood, and implementation effort. This is complemented by a Self-Assessment Questionnaire that benchmarks RI practices against European policy and regulatory requirements, supporting ex ante planning, structured prioritisation, and ex post monitoring and reporting. Second, the Environmental Impact Metric Publication System (EIMPS) operationalises sustainability assessments at scale by enabling the automated submission, validation, normalisation, and dissemination of environmental metrics across heterogeneous Grid, Cloud, IoT, and Network infrastructures. Built on a Common Information Model (CIM) and a dedicated Metrics Database, EIMPS provides customisable reporting on energy intensity and operational efficiency. Together, these tools equip digital RIs with a coherent, adaptable solution for consistent sustainability evaluation and transparent environmental reporting
Speakers: Damla Rowlandson, Shashikant Ilager (University of Amsterdam), Iida Lehto (Mandat International)
-
10
-
-
11
Accelerating Science, Decelerating Carbon: Sustainable Computing at CERN 503/1-001 - Council Chamber
At CERN, where particle physics meets cloud-native scale, workloads are increasingly powered by GPUs and accelerators for simulation and machine learning. These advances are incredible for science but they raise new questions around efficiency, energy, and carbon impact.
We’ll share how we’re building a metrics-driven framework, rooted in Kubernetes and CNCF tooling, to measure and improve workload efficiency across on-prem and cloud environments. By defining a shared set of metrics for cost and carbon reporting, we’re turning observability data into actionable insights for sustainable computing.
You’ll see how real HEP workloads use this framework to visualize performance, uncover inefficiencies, and guide smarter infrastructure decisions. The goal: empower every team to understand the environmental footprint of their cloud-native applications—and make sustainability a first-class metric in our stacks.
Speakers: Amine Lahouel (CERN), Laura Eve Sarah Llinares -
12
Energy & Emissions Monitoring of Compute Workloads Using CEEMS [Online] 503/1-001 - Council Chamber
With the rapid acceleration of ML/AI research in the last couple of years, the energy consumption of the Information and Communication Technology (ICT) domain has rapidly increased. As a major part of this energy consumption is due to users’ workloads, it is evident that users need to be aware of the energy footprint of their applications. Compute Energy & Emissions Monitoring Stack (CEEMS) has been designed to address this issue. CEEMS can report energy consumption and equivalent emissions of user workloads in real time for SLURM (HPC), Openstack (Cloud) and Kubernetes platforms alike. It leverages the Linux perf subsystem and eBPF to monitor the performance, I/O and network metrics of the applications which can help the end users to identify the bottlenecks in their workflows rapidly and consequently optimize them to reduce the energy and carbon footprint.
Besides CPU energy usage, CEEMS supports reporting energy usage and performance metrics of workloads on NVIDIA and AMD GPU accelerators. CEEMS has been built around the prominent open-source tools in the observability eco-system like Prometheus and Grafana. It has been designed to be extensible and it allows the Data Center (DC) operators to easily define the energy estimation rules of user workloads based on the underlying hardware. Moreover, it supports eBPF based continuous profiling of the user workloads on SLURM and Kubernetes platforms which proved to be an effective solution in optimizing the workloads. Currently, CEEMS has been deployed on JeanZay HPC [4] platform monitoring more than 2000 nodes and 4000 GPUs that has daily job churn rate of 20k batch jobs.
Speaker: Dr Mahendra PAIPURI (CNRS) -
13
The Green Algorithms Dashboard: Bringing Carbon Visibility to Research Computing 503/1-001 - Council Chamber
The environmental impact of research computing is largely invisible to the people generating it. Research Computing systems (e.g. HPCs) consume significant resources, yet quantifying their energy use and carbon footprint, especially at the level of individual jobs and users, remains a genuine challenge. As a result, researchers rarely have visibility into the carbon cost of their work, making meaningful progress toward sustainable computing practices difficult to achieve.
The Green Algorithms Dashboard bridges this gap. Drawing on data from workload managers like SLURM, it provides fine-grained daily monitoring of resource usage and estimated carbon footprint across different levels, from individual users and research groups to departments and institutions. The platform adapts the underlying methodology of the widely adopted, peer-reviewed Green Algorithms calculator [1]. The accuracy and reliability of the underlying methodology has been independently validated against physical power meters and alternative software-based tools [2].
The dashboard is designed with data privacy as a core principle: all computations are performed and results stored locally on the HPC system, giving teams control over the scope and granularity of what is monitored and shared. A clean, comprehensive UI presents carbon estimates in a way that is intuitive and actionable, empowering researchers and institutions to make more environmentally conscious decisions about computationally intensive work.
This talk or poster will present the technical architecture behind the platform, the design philosophy that shaped it, and early insights from our first pilot deployments. We will also outline the road ahead, including planned support for GPU-intensive workloads, broader scheduler integration, and the open challenges we are working to solve.[1] Lannelongue et al. Green Algorithms: Quantifying the Carbon Footprint of Computation. Advanced Science. 2021, 2100707.
[2] Jay et al. "An experimental comparison of software-based power meters: focus on CPU and GPU," 2023 IEEE/ACM CCGrid. doi:10.1109/CCGrid57682.2023.00020.Speaker: Navirah Kamal (University of Cambridge)
-
11
-
Social: Welcome aperitif 61/1-201 - Pas perdus - Not a meeting room -
-
-
-
14
Faster, Greener, Precise Enough: Challenges and Directions in GPU Auto-Tuning 500/1-001 - Main Auditorium
High-Performance Computing (HPC) drives discovery across science and industry and underpins the rapid advances in AI. At the heart of modern HPC platforms is the Graphics Processing Unit (GPU), which delivers the bulk of compute power but also dominates energy consumption. As GPU architectures increasingly prioritize low-precision arithmetic for AI workloads, HPC applications that depend on higher precision face new programmability challenges alongside new opportunities in mixed-precision computing.
Crucially, the energy efficiency of GPU applications depends not only on compute utilization but also on memory traffic patterns, and the fastest implementation is not always the most energy efficient. Reliable exploration of these trade-offs is further complicated by the limited accuracy and temporal resolution of current power measurement tools. Combined with the vast, discontinuous design spaces inherent to GPU programming, manual optimization is infeasible.
Automatic performance tuning, or auto-tuning, offers a proven approach to this problem, automatically searching for optimal configurations across algorithm, application, and hardware parameters. To address the emerging demands of mixed-precision computing and energy-aware execution, the field is now moving toward constrained and multi-objective optimization to enable systematic exploration of the trade-offs between performance, energy consumption, and numerical accuracy. In this talk, I will highlight key challenges, recent developments, and future directions in GPU auto-tuning.
Speaker: Ben van Werkhoven (Leiden University) -
15
A quick overview of the physical and social science of the planetary crisis 500/1-001 - Main Auditorium
TBC
Speaker: Julia Steinberger (University of Lausanne) -
16
Responses & discussion 500/1-001 - Main Auditorium
-
11:00 AM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
Contributed Talks: (Track A) Project Design 500/1-001 - Main Auditorium
-
17
Estimating the environmental impact of computing for a future accelerator facility
The proposed ISIS-II Neutron and Muon Source underwent a Life Cycle Assessment (LCA) during the early feasibility and design stage. The potential environmental impacts were evaluated across construction, operation, and decommissioning to identify and integrate environmental sustainability practises from its inception.
As with most modern accelerators, computing is - and will be - essential for the design and operation of ISIS-II, yet predicting the computing impact for a facility that is proposed to run from 2040 to 2100 comes with its challenges. This work shares those challenges and explores the assumptions made to attempt to predict the required computing resources of a future accelerator.Speaker: Dr Hannah Wakeling (John Adams Institute, University of Oxford) -
18
Integrating sustainability in the design of the Wide-field Spectroscopic Telescope’s data processing and storage
Astronomers have a unique perspective on the Earth, its fragility, and the absence of a ‘planet B’. Yet the carbon footprint of their activities and instruments remains substantial. The amount of data generated by these increasingly precise technical instruments, and the subsequent computing infrastructure needed to process and store it, poses a key challenge to sustainability.
The Wide-field Spectroscopic Telescope (WST) is a 12-meter spectroscopic facility currently under development and potentially operational in Chile in 2040. One of its key science cases is the so-called ‘time domain astrophysics’, which requires rapid follow-up observations of transient sources in the night sky. The WST will be equipped with a provisional number of ~600 detectors, which will gather data continuously every night over an expected lifetime of 50 years. The expected data volume ranges from 1 to 3 PB per year.
In this talk, we will present the first estimates of the carbon footprint associated with the data reduction pipeline and storage infrastructures. We will highlight the WST's ongoing effort to incorporate sustainability considerations into hardware selection for data processing and storage, and explore how location affects the carbon footprint of the different solutions. We will also discuss how integrating sustainability early in the design process of research infrastructures can effectively mitigate the environmental impact.
Speaker: Laurane Fréour (University of Vienna) -
19
Discussion
-
17
-
Contributed Talks: (Track B) Hardware/Software Optimisation 40/S2-D01 - Salle Dirac
-
20
Understanding and Evaluating DRAM's Operational and Embodied Emissions
Problem:
AI workloads are driving server memory capacity beyond the terabyte range, placing DRAM at the centre of both the hardware supply crisis and the environmental cost of modern computing [3]. Memory already accounts for 40–46% of total server energy consumption , rivalling or exceeding the processor [1, 2, 8, 9], especially for training and inference of large language models [5, 10]. In contrast, many sustainability studies across data-intensive research disciplines (e.g., genomics, astrophysics, climate modelling) still treat the processor as the dominant driver of carbon emissions, primarily because of the lack of a concrete method to account for DRAM’s energy [7]. Further, DRAM's environmental cost begins at the fab, and existing lifecycle assessments for research computing rarely capture DRAM's manufacturing-phase (embodied) emissions [4, 6] alongside its operational footprint [10].Goal:
The goal of this interactive presentation is to give an in-depth look into how modern DRAM consumes energy and how we can start to reason about DRAM’s operational and embodied carbon emissions. I will first present a brief, discipline-agnostic introduction to how DRAM works and how it interacts with the processor, requiring no prior hardware knowledge. We will then look at how DRAM dictates modern computing system performance, and why memory has become a dominant power consumer in datacenters. We will next understand how to optimize applications to reduce the energy and carbon emissions associated with memory.Expected Takeaway:
Audiences should expect to gain an understanding of how DRAM impacts the overall system’s performance, power, cost, and carbon emissions. Attendees will leave with a practical estimation methodology, applicable regardless of research domain, for accounting DRAM's operational and embodied carbon, directly supporting community efforts to build shared sustainability accounting standards. It is important to remember that efficiency gains alone are insufficient if workload growth continues to outpace them (Jevons paradox)[7].Speaker: Mr Aditya Manglik -
21
Mitigate. Innovate. Sustain.
Mitigation starts by treating technical debt and digital waste as sustainability problems. Idle compute, chatty microservices, oversized payloads, unbounded concurrency, always-on resilience, and forgotten data, environments, or duplicate workflows all become permanent excess: more kilowatt-hours, more emissions, more cloud spend.
We uncover where energy is wasted in everyday systems, from CPU boost events with no user value and I/O waits misread as CPU issues to zombie instances and hidden digital waste across storage, pipelines, and tooling.
These findings become a green debt backlog tied to measurable reductions in energy, cost, and operational heat.
The talk reframes performance as value per watt. It explores event-driven designs, lean data contracts, carbon-aware compute placement, and CI/CD checks for efficiency.
Attendees leave with practical habits to mitigate waste, innovate greener systems, and sustain gains long after the conference.
Speaker: Wilco Burggraaf -
22
Reducing machine learning model emissions by integrating physics knowledge
In many science and engineering disciplines, limited representative training data, poor reproducibility and low interpretability hinder the complete integration of AI. The environmental impact of this growing technology is also of particular concern. Physics-informed machine learning (PIML) seeks to answer the aforementioned challenges, with many techniques leveraging physical knowledge to reduce training data requirements. We aim to explore how and when these approaches reduce model emissions.
In this initial work, we embed physical insight into our ML models, assessing performance and emissions jointly on two simple benchmarks, one synthetic and one from an engineering lab experiment. The work demonstrates how a PIML approach can reduce emissions based on reduced training data, yet also highlights the increased model complexity from additional hyperparameters to be optimised, and the tradeoff between these factors to decrease overall emissions.
Speaker: Daisy Bradley (University of Sheffield) -
23
Sustainability-Aware Workload Shifting Beyond Carbon Intensity
Time- and location-shifting of computational workloads is widely proposed to reduce data-centre emissions by exploiting variation in electricity carbon intensity. However, CO$_2$-only optimization can shift burdens to places where impacts are experienced locally, such as water withdrawals in stressed basins, worsened air-pollution exposure, and increased stress on constrained grids. We present Orca, a sustainability-aware workload shifting framework that jointly considers global climate impacts and heterogeneous local criteria. Orca integrates region- and time-dependent signals for carbon, water-stress--weighted water use, air-pollution exposure proxies, and grid-stress indicators, and formulates scheduling as a multi-objective optimization problem. Using Pareto analysis, preference weighting, and optional impact caps, Orca exposes and mitigates trade-offs between emissions reduction and local burdens. A three-region case study shows that CO$_2$-optimal shifting can worsen local outcomes, while Orca produces context-sensitive schedules that better balance global and local sustainability objectives.
Speaker: Geerd-Dietger Hoffmann (Green Coding Solutions, University of Potsdam)
-
20
-
Contributed Talks: (Track A) Data Storage 500/1-001 - Main Auditorium
-
24
Engineering a Scalable, FAIR Data Infrastructure for Resource-Efficient Research
The Department of Aeronautics at Imperial College London is addressing the environmental and accessibility challenges of modern research by deploying a scalable repository architecture. This system integrates a custom InvenioRDM interface with Ceph object storage to manage massive computational datasets in alignment with FAIR principles. By leveraging software-defined storage on commodity hardware, the department avoids carbon-intensive "forklift upgrades," allowing for sustainable, incremental capacity growth.
The infrastructure features a self-healing, S3-compatible backend designed to eliminate "dark data" through domain-specific metadata curation. To reduce energy consumption associated with unnecessary network egress, the platform supports flexible retrieval, enabling researchers to inspect granular data subsets rather than downloading entire multi-terabyte files. Ultimately, this ecosystem prevents redundant, energy-heavy re-computations by transforming primary data into a permanent, reusable asset. The proposed talk details the lifecycle of this transition from ad-hoc management to an integrated, environmentally conscious research framework.
Speaker: Irufan Ahmed (Imperial College London) -
25
Discussion
-
24
-
1:10 PM
Lunch break 61/1-201 - Pas perdus - Not a meeting room -
-
26
Sustainable Computing in an Era of Rising Hardware Costs and Slowing Per-Core Progress 500/1-001 - Main Auditorium
The cost structure of computing hardware is undergoing a fundamental shift. Prices for accelerators, high-performance CPUs, memory, storage, and networking have risen due to architectural complexity, supply-chain constraints, and demand from AI and large-scale simulation. At the same time, single-core CPU performance gains have slowed, with overall improvements driven mainly by higher core counts. This combination—rising costs and diminishing per-core gains—has become particularly acute for memory and solid-state storage. Together, these trends are reshaping procurement strategies and the environmental footprint of research computing.
This talk examines how these changes alter hardware replacement economics and redefine sustainable computing. Frequent refresh cycles now compete with strategies focused on extending system lifetimes, selective upgrades, and optimizing software for parallel, heterogeneous platforms. Drawing on examples from large-scale scientific computing, it explores trade-offs between cost, performance, energy efficiency, and carbon impact, arguing that sustainability must incorporate economic and architectural drivers alongside energy considerations.
Speaker: Ian Fisk (Simons Foundation) -
27
Software Engineering and Sustainability: Experience Cultivating Better Systems 500/1-001 - Main Auditorium
Sustainability and software engineering are often treated as separate concerns. They do intersect. One often focuses on energy, carbon, and environmental limits. The other focuses on code, systems, longevity, reproducibility, and delivery.
This talk asks if the link runs deeper. Both fields face similar pressures. Demand rises. Resources are finite. Complexity grows. Hidden costs accumulate. I draw on work in network systems, media, broadcasting, open source, mentoring, platform-scale systems, and research computing.
If so, better software engineering is not merely adjacent to sustainability. The two can be symbiotic.
Through projects and examples, the session explores how to build systems that scale better and waste less. It also asks how systems can create more capable participants. Finally, I consider what opportunities lie beyond net zero, and how we build for that world.
Speaker: Michael Philip Sparks (The University of Manchester (GB)) -
4:00 PM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
Poster Lightning Talks 500/1-001 - Main Auditorium
-
28
Speakers: Adrian CIOCAN (MAIF, France, La Rochelle University, France), Amine Lahouel (CERN), Ana Catarina Gouveia Braz (École Polytechnique Fédérale de Lausanne), August Ning (EPFL), Dr Christina Bremer (University of Cambridge), Christophe Farges, Dr Colin C. Venter (University of Huddersfield (GB)), Georgios Sarantakos, Hannah Scott (Imperial College London), JAIME IGLESIAS BLANCO (Spanish National Research Council (CSIC)), Jaime Iglesias Blanco (CSIC), Jeroen Dobbelaere (Institute of Science and Technology Austria), Jyoti Bhogal, Kirsty Pringle, Laura Eve Sarah Llinares, Matteo Bunino (CERN), Matteo Zanotto, Maximilian Horzela (Georg August Universitaet Goettingen (DE)), Michael Philip Sparks (The University of Manchester (GB)), Nicholas Souter (University of Sussex), Pascal Emmenegger, Rubén Rodríguez Álvarez (EPFL), Sadie Bartholomew, Tia Haddad (Kingston University London), Vladimir Bahyl (CERN), Dr Xavier Eric Ouvrard (EPFL EcoCloud), Yishak Tadele Nigatu (University of Trento)
-
28
-
-
29
Developing a sustainable institutional research computing culture
A sustainable research computing culture is vital to ensure research institutions can meet their sustainability aims while supporting ever-increasing demands for computational capacity driven by modern research practices. Currently, challenges in three areas make it difficult for sustainability practices to become fully embedded within universities and for a sustainable, responsible research computing environment to develop.
The first is a lack of genuine buy‑in from staff and students. Without this shared commitment and a supportive community, implementing effective policies and practices becomes significantly more challenging.
Secondly, awareness of IT sustainability issues remains limited across much of the research landscape. Education and training to develop skills in sustainable computing is a key requirement for making low‑impact digital practices routine rather than exceptional.
The third challenge is the lack of integration into organisational strategy. Strong, coordinated advocacy is required to shift perceptions, influence decision‑making, and establish sustainability as “business as usual”.
This talk and poster will outline work underway at Imperial College London - supported by Research England research culture funding - to foster a sustainable institutional research computing culture and create models that can be adopted elsewhere. Building on the above observations, our approach centres on three pillars:
Community: Building community through engagement and events, to develop shared understanding among researchers and technical professionals of sustainable computing practices, particularly as AI and high‑performance computing become increasingly prominent.
Education: This pillar is anchored in the development of a new green and sustainable computing training course. Building on open educational resources, it offers core knowledge and best practices, supports progress toward Green DiSC certification, and provides a resource for cultural change.
Advocacy: Involves working with institutional leaders and sustainability teams to highlight practical improvements, large and small, and includes partnering with an artist to develop visual outputs that inspire broader engagement with sustainable IT.
Speakers: Hannah Scott (Imperial College London), Jeremy Cohen (Imperial College London) -
30
A Comprehensive Data Analysis of Three IBM POWER9 Cooling Systems in the ExaDigit Framework [Online]
The increasing power densities of exascale supercomputers require complex cooling infrastructures. ExaDigiT [1] addresses this need by providing a digital twin for liquid-cooled supercomputers through Functional Mock-up Unit (FMU)-based thermo-fluid models. These FMU simulators are accurate but computationally expensive for rapid what-if scenario exploration and Monte Carlo uncertainty quantification studies. Hence, replacing these simulators with efficient deep learning surrogates is a key step toward energy-aware digital twins. But effective surrogate design requires a prior understanding of the input-output relationships, temporal dynamics, non-linearities, and spatial coupling. This work addresses that gap through a systematic statistical analysis of FMU-generated data for three IBM POWER9 systems: Marconi100, Summit, and Lassen, all within the ExaDigiT framework. We performed six progressive analyses: (i) direct effect, (ii) multivariate effect, (iii) rate of change, (iv) non-linearity, (v) spatial interaction, and (vi) physics consistency.
Our key findings are (i) a structural decoupling between the primary and secondary loops, as the primary loop is actively managed and substantially nonlinear and load-following; (ii) level-driven dynamics (median rate/level ratio = 0.33) with 54% of pathways exhibiting higher-order transient responses and Autocorrelation Function (ACF) persistence beyond 1,000 lags; and (iii) negligible inter-CDU thermal propagation that was confirmed by distance-independent correlations and asymmetry ratio Rasym = 0.95. and (iv) near-zero thermodynamic violations with tolerance-calibrated energy conservation constraints.REFERENCES
[1] Wesley Brewer et al. “A Digital Twin Framework for Liquid-cooled Supercomputers as Demonstrated at Exascale”. In: Proceedings of the International Conference for High Performance Computing, Networking, Storage, and Analysis. SC ’24. Atlanta, GA, USA: IEEE Press, 2024. DOI: 10.1109/SC41406.2024.00029. URL: https: //doi.org/10.1109/SC41406.2024.00029.Speaker: Mr Yishak Tadele Nigatu (University of Trento) -
31
Applied AI for Power Management and Predictive Maintenance in the LHCb HLT2 Data Centre
In the ODISSEE EC project, we are developing applied-AI capabilities for the LHCb HLT2 data centre to improve energy efficiency and operational reliability, with the longer-term aim of integrating these building blocks into an end-to-end digital twin for computing infrastructure.
In this presentation, we will describe the initial work and architecture around two lines of development. For power management, we build models to predict when parts of the farm will be idle from workload signals (e.g., LHC operational status and Monte Carlo job queues) so that nodes can be placed into low-energy states, and we explore approaches to improve cooling-plant efficiency, starting from tuning existing control logic and evaluating data-driven control strategies on historical monitoring data. For predictive maintenance, we apply unsupervised anomaly detection to telemetry from both compute nodes and cooling/environmental equipment, with a human-in-the-loop process where LHCb experts review model findings and help define what constitutes an actionable anomaly. Moreover, we are planning on evaluating applications of the Energy Aware Runtime from Energy Aware Solutions.
We will show the monitoring data sources, data preparation, and first baselines, and discuss how these results can evolve toward a digital twin that links workload and scheduling decisions to heat and cooling behaviour, enabling simulation of PUE and total energy consumption under realistic operating conditions.
Speaker: Matteo Bunino (CERN) -
32
CompilePython.com: Capturing Community Knowledge on Compiling Python for Real-World Software
This contribution introduces CompilePython.com, a community-driven project bootstrapped in February 2026.
Python is widely used in research, but it is often slower than compiled languages such as C++ or Rust. The combination of widespread use and performance leads to a green compute problem. Slower software means more compute time, more energy, and more CO₂.
Compiling Python for real-world software sounds straightforward - until you try it on actual projects.
There are many tools and experiments, but practical knowledge is fragmented. What works? What breaks? What subset of Python compiles reliably in practice? What trade-offs are involved when moving beyond examples to production-style code?
CompilePython.com provides a collaboration space for researchers and practitioners to document real-world experience of compiling Python: successes, failures, working subsets, and engineering constraints. It is not a compiler or a finished product, but a GitHub Pages-based site focused on credible, experience-led guidance.
In 2026 the project will host informal "Lightning Camps" centred on real projects, tool comparison, and sustainability implications grounded in practice.
The contribution outlines the motivation, current stage, and roadmap, and invites contributions from others tackling similar challenges.
References
- CompilePython.com - https://compilepython.com/
Speaker: Michael Philip Sparks (The University of Manchester (GB)) -
33
Cutting Carbon in SKA Science Computing though Lifecycle Management, Load Shaping, and Accelerators
IT electricity needs are growing quickly. It is the responsibility of all actors, including the scientific community, to tackle that growth and its associated carbon emissions. In particular, some scientific projects, as is the case of the SKAO, need that their data centers are located physically adjacent to the data production site; in some cases, the electricity consumption of a large HPC data center can rival that of the local communities. The SEAMS (Sustainable and Energy-Aware Methods for SKA Observatory) project aims to reduce the total carbon emitted to produce a set of scientific products in the field of radio interferometry. In particular, the objective is to optimize the operation of the Science Data Processors of the SKA telescopes. SEAMS follows three main ideas: First, the proposal of a framework, CEO-DC, to orchestrate the lifetime of computing equipment in a data center. CEO-DC considers the relative impact of reducing the carbon emissions of the local electric grid, extending the lifetime of existing hardware in the data center, or upgrading the hardware to new components with higher energy efficiency, considering the carbon released during the lifecycle of that new hardware. Second, exploring, in collaboration with other projects, the modulation of instantaneous computational loads to the availability of carbon-free electricity. Finally, designing new accelerator architectures that bring to the forefront energy efficiency rather than absolute performance, and new scheduling techniques that can select the appropriate HW resource for each computation step to minimize overall energy consumption while guaranteeing the required scientific output.
Speakers: Dr Ruben Rodriguez Alvarez (EPFL Embedded Systems Lab), Dr Denisa-Andreea Constantinescu (EPFL Embedded Systems Lab), Dr Miguel PEÓN-QUIRÓS (EcoCloud EPFL) -
34
Datacenters and Artificial Intelligence: Piloting the Energy Performance of AI Usages as a Strategic CSR Challenge [Online]
The rapid diffusion of artificial intelligence (AI) is significantly reshaping the energy profile of modern datacenters. Recent international assessments forecast a sharp rise in electricity demand driven by large-scale inference services and generative models [1]. Yet AI systems continue to be evaluated primarily through algorithmic metrics such as accuracy and latency, while their operational energy footprint remains weakly integrated into governance and deployment decisions. This imbalance creates structural tension between digital innovation and corporate sustainability commitments.
Recent studies underline the growing environmental impact of AI infrastructures and the need for sustainable scaling strategies [2, 3, 4]. However, most approaches remain model-centric, focusing on training efficiency or carbon accounting frameworks [5, 6, 7]. Datacenter optimization still relies predominantly on infrastructure-level indicators such as Power Usage Effectiveness (PUE) and cooling efficiency [8, 9], leaving operational AI usages insufficiently governed from a sustainability perspective.
This extended abstract asks: How can AI usages be integrated into datacenter-level CSR governance through actionable, decision-oriented indicators? It proposes (1) a usage-centric reframing of AI energy governance, shifting the analytical unit from models to operational AI usages in production, and (2) a structured set of Key Performance Indicators (KPIs) aligning AI deployment with Corporate Social Responsibility (CSR) objectives and regulatory frameworks such as the European Corporate Sustainability Reporting Directive (CSRD) [10].
Drawing on Green AI principles [5, 6], sustainable AI scaling research [2, 4], and datacenter energy management literature [8, 11], each AI usage is characterized by business purpose, execution frequency, operational criticality, and energy intensity. The proposed KPIs—Energy Intensity of AI Usage, Business Value per kWh, Marginal Energy Cost of Performance, Energy Criticality Ratio, and an AI CSR Controllability Index - transform AI energy management into a structured managerial decision process. In this framework, energy sobriety becomes a measurable optimization strategy rather than a constraint.Speakers: Dr Adrian CIOCAN (MAIF, France, La Rochelle University, France), Dr Angela CIOCAN (CERADE, ESAIP, Angers, France, L3i, La Rochelle University, France), Prof. Vincent COURBOULAY (L3i, La Rochelle University, France) -
35
EcoCloud: An Experimental Research Facility for Sustainable Computing and Computing for Sustainability
The rapid rise of artificial intelligence and data-intensive workloads has led to a dramatic increase in hardware deployment, energy consumption, and overall carbon footprint. Research computing is not immune to this trend, even when its objectives explicitly target sustainability. At the same time, conventional cloud and high-performance computing (HPC) platforms prioritize production stability, which significantly limits the ability to experiment with novel energy- and carbon-aware mechanisms.
Addressing these challenges requires dedicated experimental infrastructures that enable the evaluation of sustainable computing approaches at scale.
EcoCloud, a research experimental facility at EPFL, is designed as a living laboratory for sustainable computing and computing for sustainability. It supports experimental research across the full stack---from hardware and cooling technologies to system software, orchestration, and scheduling. The platform integrates fine-grained energy monitoring and carbon-intensity data at infrastructure, rack, server, and workload levels, enabling comprehensive measurement and analysis. EcoCloud also supports experimental and adaptive schedulers, including carbon-aware and renewable-aware approaches.
The facility offers a wide diversity of hardware, allowing comparative studies across multiple generations and performance profiles, as well as customized, ad hoc configurations incorporating various accelerators, including GPUs and FPGAs. In addition, EcoCloud enables experimentation with advanced cooling techniques, ranging from traditional air cooling to immersion cooling and state-of-the-art microchannel direct liquid cooling. Retrofitting older hardware with liquid cooling is also explored as a means to extend equipment lifespan and reduce environmental impact.
In this talk, we present the design principles, operational insights, and research opportunities emerging from operating EcoCloud as an experimental research facility. We advocate for the broader adoption of experimental research clouds as a key enabler for accelerating innovation in sustainable and responsible research computing, and for facilitating the direct transfer of validated solutions into production data centers to improve their sustainability.Speaker: Xavier Eric Ouvrard (EPFL EcoCloud) -
36
Energy impact of programming languages and their frameworks on business/web applications
Context. The energy consumption of the IT sector has been steadily increasing for several years and even accelerating, rivaling that of some industrialized countries (equivalent to the United Kingdom in 2022). In the software engineering industry, energy consumption constraints are often neglected during application specification, in favor of immediate requirements such as performance, scalability and delivery deadlines.
Objective. Propose an experimental protocol to evaluate, in a reproducible way, the performance and energy footprint of functionally identical business applications developed in several languages/frameworks (e.g., Java, PHP, Python, Rust), while integrating industrial software quality criteria (maintainability/scalability/code quality).
Experimental design. Starting from realistic application design documents, we will develop functionally identical applications in each chosen technology stack, with the goal of having identical inputs/outputs (endpoints, database, logging). Execution environments will be standardized: single hardware, fixed OS and versions, and execution configurations documented using industry standards. Load will be controlled by a traffic generator as used in performance testing, with emphasis on reproducibility of runs (timing and request parameters). The goal is to subject each application to an identical workload and duration. Scenario variants will allow focusing on specific aspects (response time, load spikes). Power consumption measurements will be carried out using software tools based on reliable methods (RAPL) and will be correlated with standard performance metrics.
Analysis plan. Languages/frameworks will be compared according to the energy consumed for an identical workload and execution time. The primary criterion will be the amount of energy used for each test scenario. Secondary criteria will cover latency, CPU/memory usage and stability.
Expected contributions. A test protocol and reproducible results, and a collection of functionally equivalent applications.
Speaker: Christophe Farges -
37
Enhancing Sustainable Research: A holistic Approach to Sustainable Computing (Poster)
At EPFL, the EcoCloud research center is at the forefront of both IT for sustainability and sustainability in IT. Among its flagships initiatives, HeatingBits — funded by EPFL Solutions 4 Sustainability— unites six EPFL laboratories and EcoCloud to develop and experimentally validate an holistic approach to datacenter design and operation optimized for minimal carbon footprint and seamless interaction with local energy systems.
HeatingBits combines (a) energy efficient data centers design based on an DC/DC energy distribution system and on-chip cooling for higher performance and life cycle of the CPUs/GPUs, and (b) carbon- and market-aware optimal control of its assets, such as PVs and battery energy storage systems, considering real-time carbon-intensity estimation of power grid electricity. Heat generated by computation is recovered through an innovative direct liquid cooling technology featuring custom micro-channel cold plates tailored for high-temperature operation, at approximately 75°C, significantly enhancing heat-recovery value while increasing computing performance.
The project implements a dual-season heat-reuse strategy: supplying EPFL’s central heating network in winter and driving a custom Organic Rankine Cycle system for cogeneration in summer, converting heat in electricity. EcoCloud experimental facility—ideally located next to EPFL’s primary research datacenter and directly above the campus heating infrastructure—serves as a living lab for next-generation sustainable computing, with a particular focus on IT cooling technologies like direct liquid and immersion cooling.
Speaker: Dr Xavier Eric Ouvrard (EPFL EcoCloud) -
38
Estimating the Embodied Impacts of the Internet: Promise and Deception
To transmit data, the research computing ecosystem relies on the Internet, an infrastructure that consumes significant resources to maintain and operate. A growing body of work suggests that much of its environmental footprint is “embodied;” that is, arising from hardware manufacturing and material procurement. Yet, when included in assessments, embodied impacts are often treated as constant values or scaled relative to operational impacts, limiting their reliability (1).
To address this gap, we examine the embodied impacts of routers and switches, the network devices that forward data traffic on the Internet. We first review the few publicly available Life Cycle Assessment (LCA) reports on routers to identify the main contributors to their embodied impacts. Then, we evaluate impact scores of assessment tools at both the component and device levels, using derived input parameters from device datasheets and architectural whitepapers.
The LCA studies reveal that embodied impacts are primarily attributable to Integrated Circuits (ICs), Printed Circuit Boards (PCBs), and Power Supply Units (PSUs). However, two compounded problems hinder reliable estimations. First, critical information about these components is often missing when applying impact assessment tools, forcing one to estimate required input data such as the CPU die area. Second, existing router-level tools lack sensitivity to what appear to be critical parameters, e.g., the PCB area and layer count. As a result, these tools produce biased comparisons across network devices that differ in such characteristics, thereby hindering sound decision-making in sustainable procurement or future hardware development.
Speaker: Pascal Emmenegger -
39
Green computing in neuroimaging researchSpeaker: Nicholas Souter (University of Sussex)
- 40
-
41
GreenPhysECS: Exploring the Practical Utility of the Entity-Component-System Model for Early-Career Researchers Building Energy-Aware Parallel Research Software
This contribution will describe GreenPhysECS, an exploratory project examining whether the Entity–Component–System (ECS) architectural model can make the development of parallel research software more accessible to early-career researcher, particularly in physics.
Highly parallel hardware is now standard, yet building scalable scientific software remains challenging for researchers without extensive software engineering experience. While experienced developers can design systems that exploit modern hardware effectively, doing so reliably and reproducibly at the point of first implementation is significantly harder for novices. The ECS model was originally developed to make naturally parallel GPU friendly code easier to write and maintain, and is likely underutilised despite some initial evidence of general utility by AtomECS.
GreenPhysECS provides a containerised C++20 environment with integrated fine-grained energy tracking, allowing different architectural and algorithmic approaches to be evaluated for parallel performance and energy consumption. The project investigates whether adopting an ECS-style architecture from the outset lowers barriers to writing scalable, data-parallel systems.
By focusing on architectural choice at the earliest stages of development, the work explores whether parallelism and energy awareness can become natural properties of research software rather than later optimisations.
The project is currently exploratory, assessing both the practical utility of ECS for early-career researchers and its implications for more sustainable research computing practice. This contribution will describe early results.
References
- GreenPhysECS Research Software Story - https://everse.software/RSQKit/greenphsecs_research_software_story
- GreenPhysECS repository https://github.com/UofM-Green-Compute/GreenPhysECS
- AtomECS - https://arxiv.org/abs/2105.06447
- AtomECS Github - https://github.com/TeamAtomECS/AtomECS
- Entity Component Systems - https://en.wikipedia.org/wiki/Entity_component_system
Speaker: Michael Philip Sparks (The University of Manchester (GB)) -
42
How effective are carbon reporting tools in encouraging sustainable behaviours? Introducing E-SCOUT, a large-scale multi-centre trial
Scientific computing comes with significant environmental impacts, including but not limited to energy and water consumption, carbon emissions, and abiotic resource depletion. A potential pathway to reducing these impacts is the use of carbon reporting tools, such as carbon calculators, that quantify and visualise the impacts and can thus enable researchers to make informed, more environmentally sustainable decisions when using scientific computing. However, the effectiveness of reporting tools remains unclear, in particular to what extent sustainable behaviours adopted sustain over time – a question that could not be answered by earlier short-term eco-feedback research.
To close this gap in the literature, we are organising a large-scale trial across different research-performing organisations. Titled ‘Environmentally Sustainable Computing User Trial (E-SCOUT)’, it will be set up as a multi-centre, pragmatic, cluster randomised controlled trial to evaluate the effectiveness of a carbon monitoring dashboard on environmentally sustainable behaviour, pro-environmental attitudes, and awareness of the environmental impacts of scientific computing among researchers in research-performing organisations. The carbon monitoring dashboard used for the trial is the Green Algorithms Dashboard. To increase engagement with both the dashboard and topic, participants will also receive access to a community support package; this includes drop-in sessions, access to resources and guidelines on sustainable computing, and exchanges between participants.
To ensure that the research protocol for E-SCOUT has a strong evidence base, the trial includes a pilot phase with a small group of participants. During this phase, the participants are invited to engage with the dashboard and provide feedback on its design and effectiveness. They are also encouraged to evaluate the community support they receive.
In my talk or poster, I will provide an overview of E-SCOUT’s research protocol, present initial findings from the pilot phase, and share sign-up information for the main trial.
Speaker: Dr Christina Bremer (University of Cambridge) -
43
Implementation of background jobs for recovering CPU usage during the draining of worker nodes
On batch systems with many jobs sharing a worker node, the draining of a node in order to terminate it for operational purposes without job abortions leads to idle CPU cores and a loss of compute time.
This is becoming a prominent issue at German university-based Tier-2 centres, in particular. Towards the High-Luminosity LHC, they are undergoing a transformation and CPU will be provided via large HPC centres of the NHR alliance (Nationales Hochleistungrechnen / National High Performance Computing) in order to cope with the increasing resource needs. NHR resources are booked via their batch systems assigning entire nodes for a certain amount of time, one week in the given case. Overlay Batch Systems were chosen as a common approach to fill these resources with jobs from the LHC collaborations. In this way, the NHR nodes operate as virtual worker nodes of the existing Tier-2 centres where single- and multi-core jobs are flexibly assigned to partitionable job slots. Nevertheless, due to the limited lifetime of the virtual nodes, draining becomes a recurrent state during the operations.
In order to make use of the CPU despite the necessity to drain the node, a background job mechanism has been introduced at the NHR site Emmy in Göttingen. Highly parallelisable simulation jobs are provided by the ATLAS collaboration, run during the draining of the virtual worker nodes and only use the otherwise idle CPU slots. Towards the end of the draining, they run on most of the cores resulting in short run times, which allows for a termination of the node from 100% to 0 with minimal loss due to job abortion.
The implementation of this approach, CPU usage measurements and further aspects are presented.Speakers: Maximilian Horzela (Georg August Universitaet Goettingen (DE)), Sebastian Wozniewski (Georg August Universitaet Goettingen (DE)) -
44
Operationalising Sustainable Software Engineering (SSE) Principles Through Runtime Architectural Tactic Selection in Self-Adaptive Microservices [Online]
Sustainable Software Engineering (SSE) seeks to ensure that software systems meet societal and functional demands while minimising environmental impact and resource consumption across their lifecycle. Achieving SSE requires rigorous architectural design, measurable sustainability metrics, and the systematic integration of sustainability principles throughout both development and operation. Crucially, long-term system resilience depends on balancing technical performance with environmental considerations by accounting for environmental conditions and operational circumstances at runtime. However, existing research offers limited guidance on selecting and enacting sustainable architectural changes specifically for Microservices Architectures (MSA) during execution. This paper introduces the Green Computing Tactic Framework (GCTF), a technology-agnostic framework designed to enable autonomous, runtime sustainability-aware Architectural Tactic (SAT) adaptation. While acknowledging organisational, environmental, and regulatory influences, GCTF focuses on fine-grained technical mechanisms that support architectural-level sustainability decisions. The framework formalises context-aware adaptation as a function of: (i) multi-layer runtime state composition and Quality Attribute (QA) assessment, (ii) behavioural labelling that captures the intended self-behaviour of a service, and (iii) categorisation of architectural tactics aligned with the functional roles of services within their workload context. To ensure principled sustainability, SAT selection is further structured through a green strategy layer that encompasses the avoidance, reduction, and neutralisation of energy and carbon consumption. This prioritises the elimination of unnecessary work, minimisation of resource intensity, and environmentally informed execution of unavoidable work. While initial validation activities are underway, this paper primarily contributes a systematically grounded framework intended to enable and structure sustainability-aware architectural adaptation. Future work will report quantitative and qualitative evaluation results. By elevating sustainability-oriented tactics across architecture, data management, infrastructure, and operations, GCTF provides fine-grained technical mechanisms that complement existing DevOps, GreenOps, and FinOps practices, enabling sustainable software systems that adapt responsibly at runtime.
Speaker: Tia Haddad (Kingston University London) -
45
PUE is broken, what is next?
The rapid growth of artificial intelligence workloads and data centers has made energy usage a primary design and operational constraint. Beyond minimizing total energy consumption, the industry increasingly requires metrics that reflect how effectively energy is converted into useful computation. Over time, several data center sustainability key performance indicators have emerged addressing energy, water, carbon and heat reuse. Among these, Power Usage Effectiveness (PUE) has become dominant due to its simplicity.
Defined as the ratio of total data center energy consumption to IT equipment energy consumption, PUE provides a coarse measure of infrastructure efficiency. However, its widespread use has revealed fundamental limitations. In practice, PUE is often computed using instantaneous power rather than energy over time, obscuring seasonal variations and long-term operational behavior. More critically, PUE does not account for computational productivity, carbon intensity, or heat reuse. In extreme cases, inefficient or underutilized servers can even lead to a seemingly improved PUE, creating misleading incentives.
A central challenge lies in how energy is attributed between IT and non-IT components. Server fan power, power supply unit losses, and idle power are typically classified as IT energy, even though they do not directly contribute to computation. This accounting becomes increasingly problematic with emerging cooling technologies. For example, direct liquid cooling shifts pumping energy to non-IT infrastructure, worsening PUE despite improving server efficiency and heat recovery.
This talk analyzes the structural limitations of PUE and presents experimental results from EPFL’s EcoCloud research center. Detailed measurements across two server architectures quantify idle power, fan power, and PSU losses as a function of load, illustrating how accounting choices can distort PUE and mask real efficiency gains. The talk concludes by discussing directions for more meaningful next-generation metrics, including Infrastructure Usage Efficiency, to better support sustainable data center design and operation.
Speakers: Xavier Eric Ouvrard (EPFL EcoCloud), Dr August Ning (EPFL), Xavier Eric Ouvrard -
46
Scaling up CATS, The Climate Aware Task Scheduler, for cluster computing and international reach [Online]
Reducing energy consumption is vital for sustainable computing, but the carbon cost of a given workload also depends on the electricity grid’s carbon intensity at the time of use. Running computational jobs when carbon intensity is lower, typically during periods of higher renewable generation, can therefore reduce environmental impact for a set amount of energy use.
This principle led to the creation of CATS (Climate-Aware Task Scheduler) at the 2023 Hack Day of the Software Sustainability Institute (SSI) Collaborations Workshop. CATS is a lightweight Python package that schedules tasks based on real-time carbon intensity data from the relevant electricity grid.
Supported further by the SSI, we have continued development and released Version 1.0, which integrates with the Unix at command for small-scale local tasks, in mid 2024, and now we are finalising a Version 2.0 for integration with the Slurm workload manager, targeting batch computing and high-performance computing (HPC) environments.
In this talk, we introduce CATS and demonstrate its use locally and on a SLURM-based cluster using Docker. We discuss our aspirations to work with system administrators at volunteer centres to incorporate CATS Version 2 into key UK HPC systems to provide users with the option to intelligently time shift their jobs through use of a 'green' queue. We seek to understand where CATS sits within the landscape of carbon-aware scheduling and hope to engage with creators of other tools in the area, and indeed from potential users and interested centre managers/sysadmins.
At present CATS only interfaces with the UK's National Grid ESO API, but we are scoping for suitable APIs that can enable use of CATS elsewhere across the world, so particularly encourage ideas from the international audience to help with this.
Speaker: Ms Sadie Bartholomew (National Centre for Atmospheric Science and University of Reading) -
47
Scientific Computing Across Disciplines: A Student-Driven Evaluation of Institutional Computing Footprints and External Resource Usage
Experimental research increasingly relies on computing to analyze data, model systems, and apply AI, expanding its role across all scientific disciplines. This shift has driven a surge in demand for computational power, alongside higher energy use and material consumption, compounding the already significant environmental footprint of experimental research.
To address this, a student-driven course was developed to assess research-related environmental impacts. Collaborating with a scientific computing team, students analyzed internal infrastructure and external computing use, including AI tools, to estimate carbon and material footprints and compare them with other research activities.
Results show computing is widespread but unevenly used: most researchers consume about 1,000 CPU hours annually, while a small group accounts for over 20% of total energy use. The computing footprint stems mainly from equipment production (57%), followed by services (25%) and storage (12%).
Speaker: Jeroen Dobbelaere (Institute of Science and Technology Austria) -
48
Sustainable Data Centre Research: The RF2.0 Project at DESY
Computing forms a fundamental pillar of modern research. Efforts to minimise correlated C02 emissions of research infrastructures therefore need to include plans for sustainable computing models, and research into novel operations of data centres need to be conducted. The team on the Horizon EU RF2.0 project at DESY investigated how their data centre operates and tested a variety of operational policies to manage the environmental impact of a data centre within the larger DESY ecosystem. Some of the discussed saving schemes were trialled during an energy saving program at the data centre during the summer of 2025 where an estimated 3500 kWh of energy was saved. The culmination of DESY’s involvement in the RF2.0 project is the building of POCCET (Proof Of Concept Cluster for Emissions Tracking); a demonstrator showcasing a future of computing centres where computing policies are autonomously implemented, in possible response to external stimuli like grid cleanliness, or price fluctuations.
Speaker: Dr Dwayne Spiteri (DESY) -
49
TBDSpeaker: Jyoti Bhogal
-
50
TBDSpeaker: Amine Lahouel (CERN)
-
51
The Green Frontier of Big Data: Environmental Efficiency of Tape Storage at CERN
As the High-Luminosity Large Hadron Collider (HL-LHC) prepares to generate exabytes of data, scaling storage infrastructure must balance performance with environmental responsibility. This poster highlights the critical role of magnetic tape storage as a sustainable, high-density solution for long-term data preservation at CERN.
While disk systems provide the high-throughput necessary for active analysis, they account for the vast majority of storage energy consumption. In contrast, CERN’s archival strategy leverages the "zero-power idle" characteristic of tape, which stores over 1.2 EB of physics data while consuming less than 3% of the total storage power budget.
Key themes presented include:
- Energy Comparison: Data demonstrating the massive power savings of tape versus constantly spinning disk arrays.
- Operational Longevity: A look at the 15+ year lifecycle of tape media, significantly reducing electronic waste compared to 5-year server refresh cycles.
- Reliability Metrics: Analysis of tape’s superior bit-error rates (10,000x more reliable than HDD) for multi-decadal data integrity.
This poster demonstrates how modern tape technology remains an indispensable, eco-friendly backbone for the future of "Big Science" at the exabyte scale.
Speaker: Vladimir Bahyl (CERN) -
52
Towards Sustainable and Accountable Hybrid Cloud Computing via Carbon Intensity Forecasting
As increasingly powerful yet power-hungry AI models are developed and cloud computing adoption grows, the environmental impact of the Information Technology sector continues to rise. Data centers currently consume more than 400TWh of electricity annually and thus contribute to a globally concerning carbon footprint.
To meet the rising electricity demand and to reduce their emissions, many data centers increasingly rely on renewable power generation. Precise carbon intensity estimations are therefore required to enable emissions-aware scheduling techniques that align workload execution with green energy availability.
Despite their potential, these approaches are hindered by the general lack of transparency around the power consumption and emissions of cloud workloads. These metrics are rarely disclosed, keeping users uninformed about the sustainability decisions governing their computational resources.
This work proposes an architecture for minimizing the carbon footprint of workloads in hybrid cloud environments. A carbon intensity and power production forecaster was developed after benchmarking state-of-the-art time series models. These forecasts are integrated into a carbon-aware scheduling model that jointly minimizes emissions and maximizes local cluster usage. An accountant component was also implemented to assess the system accountability through provenance-based descriptions of the emission-aware management of workloads.
We evaluated the system by simulating a hybrid cloud infrastructure comprising a private cluster with on-site renewable power generation and public cloud services located across regions with different carbon intensity profiles. Extensive benchmarks validated the forecaster component over historical data from ElectricityMaps and compared various configurations of the scheduler model, where reductions in emissions of up to 27% were observed against a carbon-agnostic baseline. We present an implementation of the provenance-based accountant component in Kubernetes. A testbed over the EGI infrastructure will be developed in the context of the recently approved HE ENSURE project to validate the proposed system in a real setting under realistic operational conditions.Speaker: Matteo Zanotto (University of Trento) -
53
UrbanTwin: Digital Twins at the Service of a Sustainable Future through Integrated Urban Systems
Due to rising temperatures and climate change, urban areas face significant risks and impacts to livability. Cities are responsible for 75% of greenhouse gas emissions, placing them in a pivotal role in minimising the worst outcomes of climate change. They offer unique opportunities to integrate renewable energy sources, waste heat, wastewater, large-scale datacenters, and buildings, providing the ideal setting to implement a transformative multi-sectoral approach to climate action.
UrbanTwin – a joint initiative of the Board of the Swiss Federal Institutes of Technology – responds to the ambitious goals outlined in the Swiss Energy Strategy 2050 by developing a modular decision-support tool for sustainable urban systems planning. The project is a collaborative effort by EPFL, ETH Zurich, EMPA, Eawag, and WSL, combining physics-based simulations, AI-driven modelling, real-time sensing, and co-optimization of data centers and city infrastructure.
UrbanTwin supports decision-makers by providing a unified framework for modeling the interactions among infrastructures for energy, water, ICT, and human behavior, and for planning investment pathways for sustainable city development. In this talk, we will showcase demonstrators for tools, platforms, and methods that are under development for the city of Lausanne and the EPFL campus:
- REHO: an open-source tool for sustainable urban energy infrastructure
planning, optimizing multi-energy integration across buildings and
districts; - SENSEI platform: a low-power, low-cost Edge AI sensing and
compute node for decentralized environmental monitoring, enhancing
data privacy and scalability, crucial for bringing smart sensing to
cities of all sizes; - CEO-DC: a framework for sustainable scaling of data centers and
integration with renewable energy.
Speakers: Dr Catarina G. Braz (École Polytechnique Fédérale de Lausanne), Dr Denisa-Andreea Constantinescu (EPFL Embedded Systems Lab) - REHO: an open-source tool for sustainable urban energy infrastructure
-
54
Wattnet: High-Resolution Carbon and Water Intensity Modelling to Enable Impact-Aware Research Computing.Speakers: JAIME IGLESIAS BLANCO (Spanish National Research Council (CSIC)), Jaime Iglesias Blanco (CSIC)
-
29
-
14
-
-
55
Welcome & overview of the day 503/1-001 - Council Chamber
-
Contributed Talks: (Track A) Community Building and Behavioural Change 503/1-001 - Council Chamber
-
56
Green Research Software Engineering: A new role to Embed Sustainability in Research Software
Research Software Engineers (RSEs) collaborate with researchers to develop and maintain software, helping to embed best practices that improve reliability and reduce inefficiencies in research workflows.
As awareness grows of the environmental impact of computational research, a new specialism - Green RSE- is emerging. Green RSEs integrate environmental sustainability into software development, ensuring environmental considerations are addressed alongside performance and usability. But despite its potential to significantly reduce the carbon footprint of research computing, the role is still loosely defined and lacks formal recognition or clear career pathways.
To address this gap, the Green RSE Special Interest Group has been established in collaboration with the UK Society of Research Software Engineering. This initiative aims to build a community, share best practices, and define the skills and responsibilities of Green RSEs.
Early findings highlight how Green RSEs can support sustainable software practices across disciplines and contribute to reducing the environmental impact of computational research.
Speaker: Dr Kirsty Pringle (University of Edinburgh / Software Sustainability Institute) -
57
Building Communities in Green Computing
One of the main challenges in adopting green software practices is the limited understanding and resources on the topic. Which components have the most emissions? How can researchers minimise their footprint and engage others? We argue that structured community-building is a critical yet underexplored mechanism for accelerating green computing adoption.
This talk highlights recent community-building activities in green computing, including the Environmentally Sustainable Computational Science forum that connects 170+ members on an online platform, the Green Algorithms Initiative that aims to study and improve carbon calculators and impact measurement tools, and the Green DiSC certification scheme providing a sustainability roadmap for digital research groups.
Drawing on these experiences, we reflect on what has worked to grow and sustain engagement, share lessons on lowering participation barriers, and outline priorities for strengthening the green computing ecosystem ahead.
Speaker: Anica Araneta (University of Cambridge) -
58
Integrating environmental sustainability into bioinformatics training and its delivery
At the European Bioinformatics Institute (EMBL-EBI), we support bioinformatics efforts through research, data services, and training. Training plays an essential role in shaping how biodata scientists design analyses, access data, and utilise computational infrastructure. An aim of EMBL-EBI Training is not only to teach bioinformatic tools, but also their responsible use.
In this talk, we present ongoing work to embed green computing principles across our activities. This includes: (1) developing new training content, such as the “Sustainable computing in science” on-demand course published in January 2026; (2) integrating sustainability into already existing courses; and (3) exploring ways to monitor and reduce the impact of our live courses, both computationally and operationally.
We aim to share practical lessons and open discussion on encountered challenges, such as audience prioritisation, addressing gaps in trainer expertise, and integrating sustainability as standard practice rather than an add-on, making it a core component of scientific training.
Speaker: Dr Flaminia Zane (EMBL-EBI)
-
56
-
Contributed Talks: (Track B) Carbon Accounting 500/1-001 - Main Auditorium
-
59
Towards a Federated Accounting Framework for Sustainable Research Computing: A Landscape Review of the UK's Digital Research Infrastructure
Reducing the environmental impact of research computing requires reliable data on the carbon implications of resource use. To understand the current state of computing resource accounting, we conducted a landscape review of hardware-use accounting across the UK's national Digital Research Infrastructure, focussed on data needed for operational decision-making and sustainability reporting. In this talk, we will present our findings on community requirements, gaps in current accounting practices, and possible solutions for improvement. We invite discussion on how these findings can inform broader efforts to reduce the environmental impact of research computing.
Speaker: Deniza Chekrygina -
60
Life cycle assessment of the LHCb data centres : Overall impacts & Cooling systems comparison
As the global energy demand of data centers continues to increase, reducing their environmental impacts has become a major challenge. This presentation provides an environmental assessment of the LHCb data centers using a Life Cycle Assessment (LCA) approach, in line with CERN’s sustainability objectives. The presentation will include the contribution of the different components of the data centers and a comparison of several data center cooling systems currently in development or already deployed in the LHCb experiment. The objective is to assess the environmental impacts associated with these systems and to identify the main contributors to the overall environmental performance of the infrastructure.
Speaker: Roman Dandoy (Universite de Liege (BE)) -
61
What does that really tell us? Interpreting numbers in sustainability reports
It is encouraging to see more and more studies published about the environmental footprint of the ICT sector. Unfortunately, the outcomes of those studies are often misinterpreted. In fact, one can look at the footprint of a product or activity in many different ways which all make sense but serve different purposes. It is very easy to mistake one purpose for another and thus derive completely wrong conclusions, which may lead to harmful—albeit well-intentioned—decision-making.
I believe we can avoid those misunderstandings by clarifying the different methodological choices and their corresponding purpose. This mental framework helps draw correct conclusions from the growing corpus of sustainability studies. In this presentation, I summarize what I currently see as three of the most important methodological choices. Then, I’ll discuss a couple of examples from computer networks (my area of research) to illustrate how easy it is to misinterpret footprint numbers.
Speaker: Romain Jacob (ETH Zurich)
-
59
-
10:30 AM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
Contributed Talks: (Track A) Community Building and Behavioural Change 503/1-001 - Council Chamber
-
62
UKRI's approach to Sustainable Digital Research Infrastructure
UK Research and Innovation is committed to achieving net zero emissions and to supporting environmental sustainability within the research it funds. This includes improving the sustainability of UKRI’s Digital Research Infrastructure (DRI) and use of digital resources. This presentation will outline UKRI’s approach to achieving this, focussing on the action being taken in four key areas: Funding, Procurement, Training and Engagement, and Monitoring and Reporting."
Speaker: Dr Emily Wallis (United Kingdom Research and Innovation) -
63
Green DiSC: open-access community-driven Digital Sustainability Certification scheme
In the face of growing environmental impacts of computing, there is a legitimate request from researchers to identify what they can do about it. Green DiSC is an open access sustainability certification framework enabling researchers, labs and institutions to tackle the environmental impacts of their computing activities. It will be an opportunity to discuss the scheme, demonstrate how it can work, and showcase how it supports more environmentally sustainable computational research. We will also start by reflecting back on the inception of the scheme, and look ahead what’s coming next.
Speaker: Loic Lannelongue (University of Cambridge)
-
62
-
Contributed Talks: (Track B) Carbon Accounting 500/1-001 - Main Auditorium
-
64
An environmental assessment of computing services in higher education [Online] 500/1-001 - Main Auditorium
Scientific progress increasingly depends on powerful computing infrastructure, yet its environmental impacts across the full life cycle are often overlooked. This study evaluates the energy and resource efficiency of computing hardware and software within the Faculty of Science and Engineering (FSE) at the University of Manchester, informing the university's sustainability targets, including Net Zero by 2050.
Using life cycle assessment (LCA), hardware audits, and software energy profiling, the study compares three computing setups: desktop computers, laptops, and high-throughput computing cluster nodes. System boundaries cover component production and operational use across 15-year scenarios.Specialised compute nodes dominate most impact categories, particularly those driven by manufacturing. Across nearly all indicators, embodied impacts exceed operational electricity use, highlighting the growing significance of manufacturing burdens despite grid decarbonisation. Scenario analyses demonstrate that replacement strategies strongly influence outcomes, underlining the importance of demand management and procurement policies in reducing academic computing's environmental footprint.
Speakers: Caterina Doglioni (The University of Manchester (GB)), Tobias Fitschen (The University of Manchester (GB)) -
65
Carbon Accounting for UK Research Computing: proof of concept & beyond. 500/1-001 - Main Auditorium
UK Research and Innovation’s Digital Research Infrastructure (UKRI DRI) is inherently heterogeneous as the computing resources have grown organically and are specialised for various disciplines. Sustainability approaches for UKRI DRI must therefore span this heterogeneous landscape.
The IRIS Carbon Audit SnapshoT (IRISCAST) project estimated the total carbon costs of 6 UKRI DRI services over a 24 hour snapshot period. The IRIS Carbon Mapping Project (IRIS-CMP) then allocated carbon costs at 2 UKRI DRI services to individual service users. UKRI is currently funding the NetDRIVE project to gather evidence, make progress and make recommendation towards a NetZero UKRI DRI.
This presentation highlights the key findings of the IRISCAST and IRIS-CMP Projects and re-interprets these in the light of the subsequent work of NetDRIVE and the NetDRIVE Working Group on Metrics and Reporting all in the context of the UKRI DRI.Speaker: Alex Owen (University of London (GB)) -
66
How accurate are current tools and models for estimating software energy consumption? [Online] 500/1-001 - Main Auditorium
Evaluating the energy consumption of software is inherently complex, as software itself does not consume energy directly; rather, it is the hardware on which it runs that does. Over the past few years, numerous tools and models have emerged to estimate energy consumption at various levels, such as servers, containers, processes, and transactions. When assessing the energy consumption of software, it is crucial to understand the accuracy of these tools and models. This session presents the findings of experimental evaluations that assess the precision of various tools and models for this purpose. We particularly focus on those that estimate energy consumption at the container, process, and transaction levels.
Speaker: Andreas Brunnert (Munich University of Applied Sciences HM) -
67
-
64
-
Interactive Session 503/1-001 - Council Chamber
-
68
IT in your research – what impacts?
This interactive workshop, “IT in Your Research – What Impacts?”, aims to raise awareness among researchers about the environmental footprint of their research activities. Targeting group of 20–30 participants, the workshop combines theoretical foundations with hands on practice, encouraging attendees to analyze their own research projects or publications. The session begins with an introduction to Life Cycle Assessment (LCA), covering manufacturing, distribution, use, and end of life phases. Participants explore current challenges in environmental data availability, especially for IT hardware, and discuss differences between self hosted and cloud based computing. Through guided group work and the use of dedicated tools — such as Green Algorithms, cloud impact estimators, and GenAI footprint tools — participants assess their research-related IT impacts and identify opportunities to reduce them.
Speakers: Manuel Cubero-Castan (EPFL, VPS), Julia Paolini (EPFL)
-
68
-
12:30 PM
Lunch break - meet outside Council Chamber Restaurant 1
Restaurant 1
CERN
Sit-down lunch in Restaurant 1
-
Interactive Session 503/1-001 - Council Chamber
-
69
What’s Stopping Emissions Monitoring Across Compute Services? 503/1-001 - Council Chamber
There is a consensus in the sustainable computing community that understanding the carbon footprint of computing is vital. Despite this, wide-spread adoption of emissions monitoring has not yet been realised.
In this interactive session, we will share recent work from STFC’s Scientific Computing Department which has focused on measuring and reporting user carbon emissions in different contexts, as well as plans for an upcoming NetDRIVE-supported (https://uknetdrive.org/) community project to develop training materials on this topic. We will invite participants to discuss existing initiatives, share experiences, and help identify key education gaps and barriers with respect to adopting emissions monitoring tools. The aim of this session will be to consolidate ideas to maximise the impact of future training courses and materials for all stakeholders across compute infrastructure.Speaker: Jessica Huntley (STFC)
-
69
-
Session for discussion and collaborative drafting of content for a Conference Summary Paper.
Miro board: https://miro.com/welcomeonboard/TGxjYWRKU0wrcnFZTlVOUy95RmVMcysxcFhaWTFIOGtCNWFxam5iaXpSZHkvR25heFE3aDI4N2Y1VmcvQzg2QjRxTzJuRlNFUVRxWU8yOHJ5WEwwcnVoL3orZ1lERzloMTNCdnIwWVovL2d0ZjZtVkovZVg2akZZcWFpL1lnMUlnbHpza3F6REdEcmNpNEFOMmJXWXBBPT0hdjE=?share_link_id=851621831312
-
Interactive Session 503/1-001 - Council Chamber
-
70
Beyond the Dashboard: What Should Research Computing Carbon Monitoring Actually Do?
Building tools that monitor the environmental cost of computational work is only the first challenge. The harder question is: what should they actually show, and to whom, and in what form, to make a genuine difference?
This interactive session uses the Green Algorithms Dashboard—a daily monitoring tool for computational resource usage and estimated carbon footprint of HPC systems—as a working example. Participants won't just interrogate what the tool currently does, but what research computing carbon monitoring should do.
After a brief introduction of the tool, participants will be grouped and given a scoped, technically grounded problem, working toward a one-page proposal covering the problem, available data, and a path forward. The aim is a community-informed roadmap, openly published after the conference.
If you submit jobs to a cluster, write research code, or simply want research computing to have a smaller footprint, this is your chance to shape what comes next.
Speaker: Navirah Kamal (University of Cambridge)
-
70
-
3:30 PM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
71
Developing a sustainable institutional research computing culture 503/1-001 - Council Chamber
A sustainable research computing culture is vital to ensure research institutions can meet their sustainability aims while supporting ever-increasing demands for computational capacity driven by modern research practices.
This talk, and the associated poster, will outline work underway at Imperial College London to foster a sustainable institutional research computing culture and create models that can be adopted elsewhere. Our approach centres on three pillars:
Community: Building community to develop shared understanding among researchers and technical professionals of sustainable computing practices, particularly as AI and high performance computing become increasingly prominent.
Education: Underpinned by the development of a new green and sustainable computing training course that delivers core knowledge and highlights best practices, providing a resource for cultural change.
Advocacy: Involves working with institutional leaders and sustainability teams to highlight practical improvements, large and small, alongside work with an artist to develop visual outputs that inspire broader engagement with sustainable IT.
Speaker: Jeremy Cohen (Imperial College London) -
72
Responses & discussion 503/1-001 - Council Chamber
-
5:15 PM
Cern tours - Guides will meet you outside Council Chamber
DCVP: Maarten Litmaath
SC: Antonella Del Rosso -
Free Public Webinar [Online]: Environmental Impacts of Big Computing
YouTube Live link: https://www.youtube.com/live/JQbdZ7mO3ZA
Conveners: Jyoti Bhogal, Kirsty Pringle, Loic Lannelongue (University of Cambridge), Raghavendra Selvan (University of Copenhagen)-
73
Free Public Webinar [Online]: Environmental Impacts of Big ComputingSpeaker: Loic Lannelongue (University of Cambridge)
-
74
Panel Discussion with Jyoti Bhogal, Lauranne Fréour, Kirsty Pringle and Raghavendra Selvan [Online]
Chair: Loic Lannelongue
Panel members:
- Jyoti Bhogal
- Lauranne Freour
- Kirsty Pringle
- Raghavendra SelvanSpeakers: Jyoti Bhogal, Kirsty Pringle, Laurane Fréour, Raghavendra Selvan (University of Copenhagen)
-
73
-
55
-
-
75
On the Challenges in Assessing the Sustainability of AI [Online] 500/1-001 - Main Auditorium
Recent developments in deep learning have transformed several application domains, including computer vision and natural language processing, with new and exciting possibilities in Artificial Intelligence (AI). These advancements have been enabled, and accelerated, by large scale computations on massive data which also translate into increased energy- and carbon- costs. In this talk, I will present an overview of the factors behind the growing energy consumption and carbon footprint of AI, present techniques to quantify these costs and practices that could improve the environmental sustainability of AI. I will also point out the gaps in regulations and tools needed to assess the sustainability of AI, and suggest some directions to bridge this divide.
Speaker: Raghavendra Selvan (University of Copenhagen) -
76
Responses & discussion 500/1-001 - Main Auditorium
-
Poster Lightning Talks: Lightning Talks 500/1-001 - Main AuditoriumConveners: Hannah Scott (Imperial College London), Jyoti Bhogal, Sadie Bartholomew
-
77
Lightning TalksSpeakers: Ana Catarina Gouveia Braz (École Polytechnique Fédérale de Lausanne), Hannah Scott (Imperial College London), Jyoti Bhogal, Sadie Bartholomew
-
77
-
10:30 AM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
78
Modelling Scenarios for Carbon-aware Geographic Load Shifting of Compute Workloads [Online] 500/1-001 - Main Auditorium
We present a linear, statistics based analytical model to evaluate the reductions in CO2e emissions resulting from geographic load shifting. This model is optimistic as it ignores issues of grid capacity, demand and curtailment. In other words, real-world reductions will be smaller than the estimates. However, even with these assumptions, the presented scenarios show that the realistic reductions from carbon-aware geographic load shifting are small, of the order of 5%. This is not enough to compensate the growth in emissions from global data centre expansion.
Speaker: Wim Vanderbauwhede (University of Glasgow) -
Contributed Talks: (Track A) Cluster Operations 40/S2-C01 - Salle Marie Sklodowska-Curie
-
79
An Energy-Efficient Data Centre with Heat recovery at Queen Mary University of London [Online]
Queen Mary University of London Data Center went through a major refurbishment with an upgrade of the facility with heat recovery technology, improving the energy efficiency while increasing capacity. The data center refurbishment ensured a long term home for GridPP cluster supporting WLCG workloads, and also for future projects such as HL-LHC, LSST, SKA. The cooling system uses hot aisle containment with in row water cooling and dry air coolers, recovering server waste heat via a water to water heat pump for university district heating, offsetting gas use while benefiting from increasingly low carbon electricity. Since commissioning, added compute and storage have increased thermal loads, validating cooling and heat recovery systems while building operational experience in scaling, monitoring, stability, and incident recovery. Updated energy, heat recovery, and carbon data show that high performance scientific computing can grow capacity while reducing carbon impact through integrated heat recovery and modern water cooled infrastructure.
Speaker: Dr Sudha Ahuja (Queen Mary University of London) -
80
Wattnet: High-Resolution Carbon and Water Intensity Modelling to Enable Impact-Aware Research Computing.
Wattnet is an open-source platform for exploring the environmental footprint of electricity using open data. It provides real-time, historical, and internally generated 72-hour forecasts of carbon and water intensity for 60 European grid zones at 15-minute resolution, enabling data-driven, sustainable computing practices.
Speakers: JAIME IGLESIAS BLANCO (Spanish National Research Council (CSIC)), Mr Jaime Iglesias Blanco (CSIC) -
81
Adaptive Infrastructure and Efficiency Strategies at a GridKa WLCG Tier-1 Center
As resource demands for High Energy Physics (HEP) and other data-intensive scientific realms reach unprecedented levels, the environmental impact of large-scale computing has has increasingly moved into the focus. Facilities now face the dual challenge of increasing resource pledges while continuously reducing their carbon footprint and total energy consumption.
At the GridKa WLCG Tier-1 Center, we have implemented several measures to address this. A detailed energy monitoring system enables energy-efficient operations. Combined with energy-aware procurement processes, which prioritize performance-per-watt over the entire hardware lifecycle, we have successfully reduced total power consumption even as our resource pledges increase.
A further key ingredient of our strategy is extensive R&D and the early adoption of future technologies. Using our COBalD/TARDIS meta-scheduler for transparent, on-demand site extensions, we have enabled the migration of German University WLCG Tier-2 resources from legacy "basement" environments (PUE~1.x) to highly efficient NHR HPC centers (PUE ~ 1.0x). COBalD/TARDIS facilitates also a "follow-the-renewables" strategy, allowing Tier-1 workloads to overflow into carbon-neutral resources (e.g., LANCIUM Compute) without disrupting daily operations. Additionally, the early adoption of high-efficiency hardware, such as ARM CPUs providing a 20% efficiency increase, has delivered a significant positive impact.
Future plans, alongside our partners in the recently founded SUSFECIT project, aim to facilitate the transition from a static compute model to a dynamic, energy-aware infrastructure. Through techniques like Power Efficiency via Clock Frequency Tuning, we are moving toward a "Breathing Compute Center" that scales in sync with renewable energy availability and grid demands, further supported by our recently installed rooftop solar plant at the GridKa data center.
This talk provides a comprehensive look at how GridKa is successfully bridging the gap between massive-scale data processing and environmental sustainability.
Speaker: Lars Sowa (KIT - Karlsruhe Institute of Technology (DE)) -
82
Discussion
-
79
-
Contributed Talks: (Track B) Carbon Quantification Tools 500/1-001 - Main Auditorium
-
83
GreenMetaData: Standardising Environmental Impact Reporting for Computational Research [Online]
There is a growing need for standardised approaches to report the environmental impacts of computational research. Funders now require such reporting, researchers are increasingly motivated to disclose the impact of their work, and journals lack clear guidelines to support this practice. The Green Algorithms project addresses this need by providing open-source tools to estimate the carbon footprint of computational workflows.
Building on this, the GreenMetaData (https://github.com/GreenAlgorithms/GreenMetaData) file format represents a first step towards standardising how environmental impact is reported. It is a machine-readable, transparent, and extensible format that enables researchers to document carbon emissions and other sustainability metrics, along with the methodology used. A companion web-based tool (currently under development) supports users in generating these metadata files.
In this talk, we present the GreenMetaData initiative and engage with the research community to refine the format, improve usability, and encourage broader adoption for more sustainable computational science.
Speaker: Jyoti Bhogal -
84
The One Token Model: A Multi-Layer Framework for the Granular Estimation of AI Inference Energy
The integration of Large Language Models (LLMs) into research workflows introduces a largely opaque layer of carbon intensity. Existing approaches to estimating AI energy consumption rely on time-based heuristics or static hardware profiling, which fail to capture the non-deterministic nature of generative inference. Variations in prompt design, quantization, and decoding strategies can lead to significant fluctuations in energy use, limiting the effectiveness of current sustainability assessments.
This paper introduces the One Token Model (OTM), a unified framework that redefines energy measurement through output-normalized attribution, expressed as Joules per token. OTM integrates telemetry across three layers: infrastructure dynamics, model architecture, and inference behavior.
We validate OTM through a real-time monitoring system that quantifies the marginal energy cost of individual inference requests. By enabling fine-grained, comparable measurements across systems, OTM supports energy-aware optimization and promotes more sustainable, transparent research computing practices.
Speaker: Mr Mathieu Francois (Co-Founder & CEO, Antarctica) -
85
Towards a better understanding of environmental impacts of IT : evaluation of carbon emissions generated by organizations’ IT services
Just like other fields, IT has an impact on the environment. To effectively reduce the impact, it must first be assessed so that realistic and efficient actions can be proposed. That is the reason why we propose a methodology that enables organizations that want to work on sustainability in IT to do so independently in a reproductible way. Using an LCA approach that takes into account IT equipment and its electricity consumption, MITSI enables organizations to assess the carbon footprint of the IT services they provide and to share the results with the users of those services. By using MITSI, organizations will improve their knowledge on factors to consider and skills required to assess a carbon footprint for an IT service. They will have a better understanding of their infrastructure and will be able to propose measures to reduce their environmental impact.
Speaker: Julia Paolini (EPFL) -
86
Using co-design to improve computing carbon calculators
Computing carbon calculators enable researchers to quantify and visualise the environmental impacts of their computations, thereby identifying high‑impact changes and exploring mitigation strategies. However, they currently have several limitations, including their limited performativity (i.e., they lead to limited behavioural changes) and their “black-boxed” design in which calculation methodologies and assumptions are hidden from users. Their interface is also generally designed by the team building them with little or no input from the community. To address these limitations, we conducted two online co-design workshops with experts in design and environmental assessments and users of the Green Algorithms Calculator. Using this calculator as an example, the workshops each included two co-design sessions with a focus on “de-black-boxing”, output presentation, and calls to action. This talk provides an overview of the workshops and initial results to help inform the design, development, and use of computing carbon calculators within the SC4RC community.
Speaker: Dr Christina Bremer (University of Cambridge)
-
83
-
1:25 PM
Lunch break 61/1-201 - Pas perdus - Not a meeting room -
-
Contributed Talks: (Track A) Cluster Operations 500/1-001 - Main Auditorium
-
87
Optimizing Energy Efficiency at the WLCG PIC Tier-1: From Predictive Job Scheduling to Real-Time CPU Control
To mitigate the rising energy demands and CO₂ emissions of large-scale scientific computing, this study evaluated energy-aware resource management at the WLCG PIC Tier-1 site. We initially used machine learning to optimize node drainage by routing short jobs during peak energy prices. However, this complex scheduling approach proved impractical, requiring constant retraining for fluctuating workloads.
Consequently, we identified real-time CPU frequency scaling as a superior alternative. Unlike complex scheduling, adjusting CPU frequency is straightforward, delivers immediate power reductions, and ensures controllable performance degradation. Node-level experiments validated this approach, quantifying the exact relationship between frequency, compute performance, and power draw. By estimating facility-wide savings based on these results, this research delivers a practical, scalable framework for automated, center-wide energy optimization at PIC and other WLCG sites.
Speaker: Jose Flix Molina (CIEMAT - Centro de Investigaciones Energéticas Medioambientales y Tec. (ES)) -
88
SUSFECIT: Objectives and Concepts
Scientific computing contributes significantly to the CO2 footprint created by research conducted in ErUM (Research of Universe and Matter). In order to advance the environmental sustainability in this area the BMFTR—funded SUSFECIT (Sustainable Federated Compute unfrastructures) research network has been established with the goal to contribute to developing a strategy and interlinked software ecosystems to reduce CO2 footprint and to increase the energy efficiency of distributed computing resources. The basic idea is to exploit the dispatchability of compute jobs in space and time and use the partition of a federated computing infrastructure at a place, which at a certain time, is (dominantly) powered by renewable energies such as wind and solar power plants. To realise this basic concept, three interlinked ecosystems are to be developed and optimised: (i) to forecast the available energy mix, power costs and required compute power; (ii) to orchestrate jobs on federated and locally distributed compute clusters, taking the forecasts into account; and (iii) to account for CPU and GPU resources used, in relation to elapsed time, power consumption and CO₂ footprint. A digital twin for the above set of ecosystems shall also be developed in order to optimise e.g. operation parameters. The presentation will discuss the basic concept, the content of the three ecosystems and exploratory work, which has been conducted by partners in the research network (DESY, KIT, Universities in Aachen, Bonn, Göttingen, Freiburg and Öko-Institute).
Speaker: Dr Dwayne Spiteri (DESY) -
89
Enhancing Sustainable Research: A holistic Approach to Sustainable Computing
At EPFL, the EcoCloud research center is at the forefront of both IT for sustainability and sustainability in IT. Among its flagships initiatives, HeatingBits — funded by EPFL Solutions 4 Sustainability— unites six EPFL laboratories and EcoCloud to develop and experimentally validate an holistic approach to datacenter design and operation optimized for minimal carbon footprint and seamless interaction with local energy systems.
HeatingBits combines (a) energy efficient data centers design based on an DC/DC energy distribution system and on-chip cooling for higher performance and life cycle of the CPUs/GPUs, and (b) carbon- and market-aware optimal control of its assets, such as PVs and battery energy storage systems, considering real-time carbon-intensity estimation of power grid electricity. Heat generated by computation is recovered through an innovative direct liquid cooling technology featuring custom micro-channel cold plates tailored for high-temperature operation, at approximately 75°C, significantly enhancing heat-recovery value while increasing computing performance.
The project implements a dual-season heat-reuse strategy: supplying EPFL’s central heating network in winter and driving a custom Organic Rankine Cycle system for cogeneration in summer, converting heat in electricity. EcoCloud experimental facility—ideally located next to EPFL’s primary research datacenter and directly above the campus heating infrastructure—serves as a living lab for next-generation sustainable computing, with a particular focus on IT cooling technologies like direct liquid and immersion cooling.
Speakers: Georgios Sarantakos (EPFL), Xavier Eric Ouvrard (EPFL EcoCloud)
-
87
-
-
4:00 PM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
90
Environmentally Sustainable HeAlth REsearch (SHARE) Considerations of ethics, justice, responsibility and reflexivity. 500/1-001 - Main Auditorium
High performance computing health research is implicated in a range of harmful environmental impacts such as energy, water and other resource consumption and waste generation. Numerous tools have been developed - mainly in High income countries- to enable and encourage researchers to assess and reduce these impacts, including guidelines, carbon calculators, and laboratory certification systems. As these tools begin to be rolled out across the wider more global research ecosystem, it is important to explore potential and emerging ethical, social and practical issues. These include tensions with local values when tools are used in different geographical contexts, issues of inequity and injustice and the risks of compliance-based approaches in this field. In this talk, we will present initial findings of a multi-country project aiming at assessing the ethical, social and practical implications of using tools in health research sites in five countries across four continents (UK, India, Ghana, Kenya and Brazil). We will present our study design, including how we developed an approach to compare large qualitative data sets across countries, as well as preliminary themes around meanings of environment, environmental citizenship, communities of practice, and the importance of context.
Speaker: Gabby Samuel
-
75
-
-
91
The Political Economy of Compute: Provocations and Pathways 503/1-001 - Council Chamber
This talk reflects on the work of the Sustainable AI Futures project. Debates about AI and climate are often highly polarised, with both advocates and critics relying on methodologically weak statistics that can give a misleading sense of precision. As a transdisciplinary issue with broad social relevance, AI and climate demand careful attention to the limits of available evidence, alongside a stronger role for technical specialists as educators and communicators. Recent efforts to estimate the net impact of AI on the climate can be understood as a response to this polarisation. However, these approaches are often constrained by the uncritical adoption of assumptions and conceptual frameworks shaped by US-centric big tech. At the same time, an AI backlash has gathered pace, and alternative futures are being articulated that seek closer alignment with the needs of people and planet, though these remain underexamined. AI-related emissions also need to be more clearly situated within total global emissions, and supported by more robust and explicit theories of change. This talk offers a set of provocations, proposes gaps and challenges, and even draws on science fiction and art, in order to open up a broader conversation about sustainable computing in research and beyond.
Speaker: Jo Walton (University of Sussex) -
92
Responses & discussion 503/1-001 - Council Chamber
-
10:15 AM
Coffee break 61/1-201 - Pas perdus - Not a meeting room -
-
Interactive Session 503/1-001 - Council Chamber
-
93
Your planet needs you! Scoping a community of Sustainable Computing Ambassadors. 503/1-001 - Council Chamber
Sustainability work is often driven by individuals acting informally as advocates within their institutions, often without formal recognition, support, or career reward.
This interactive workshop will establish the foundations for a Sustainable Computing Ambassador network. Through facilitated discussions and breakout sessions, participants will map existing sustainable computing activities and co-design adaptable template job descriptions for a range of professional contexts.
Drawing on models such as Data Champions programmes, the workshop will explore how structured ambassador roles can transform informal, individual efforts into coordinated, systemic change. Participants will also consider the skills and support needed to sustain these roles. Following the workshop, the Green RSE Special Interest Group will refine the outputs through ongoing community engagement, making the resulting templates openly available to support organisations in scaling sustainable computing practices.
Speakers: Kirsty Pringle, Loïc Lannelongue (University of Cambridge)
-
93
-
94
-
95
-
1:00 PM
Science Gateway tour Science Gateway
Science Gateway
-
91