PASC25: Computing a more sustainable future

“The conference fully met our expectations,” said Laura Grigori, professor of applied and computational mathematics at EPFL and the Paul Scherrer Institute, who co-chaired the event alongside Peter Vincent, professor of computational fluid dynamics at Imperial College London. “We were really impressed by the response we got from the community. Almost every talk and minisymposium addressed how HPC can support sustainability—whether in more efficient drug design, renewable energy, reducing CO₂ emissions, or improving our understanding of climate change.”

Over 150 talks, 49 posters, and 18 published papers tackled the topic from multiple angles. While many focused on how supercomputers contribute to research aimed at helping achieve the United Nations’ Sustainable Development Goals (SDGs), the environmental footprint of supercomputing itself was also under scrutiny. “The positive impact of research made possible by supercomputers is one side of the scale,” Peter Vincent noted. “The other is the environmental cost of the supercomputers themselves. With the scale of data centres growing rapidly—especially due to AI—we must understand their power consumption and CO₂ emissions across their entire life cycle.”

In the opening keynote, Bronson Messer, Director of Science at Oak Ridge National Laboratory, showcased scientific problems that can only be tackled using supercomputers. For example, designing the landing process for a spacecraft descending onto the surface of Mars, where the atmosphere is 100 times less dense than Earth’s, cannot be tested in conventional labs—it requires complex simulations. Likewise, measuring turbulence and noise from large aircraft propellers is beyond the capacity of even the largest wind tunnels. Messer underscored how high-fidelity simulations are essential for scientific and engineering progress.

A focus on carbon emissions from supercomputing centres

Several speakers echoed the call for a better balance. In a minisymposium on sustainable computing for big data infrastructures, Florina Ciorba, professor of HPC at the University of Basel, presented figures illustrating the estimated carbon footprint of simulations run on supercomputers. One of her team’s codes for simulations in astrophysics and cosmology, SPH-EXA, was benchmarked across various machines.

A so-called hero run on LUMI, Europe’s supercomputing system in Finland, used the full machine for 12 hours and was estimated to have emitted CO₂ equivalent to 2.5 flights from New York to Melbourne. LUMI is carbon-neutral thanks to its use of hydroelectric energy. On Frontier at the Oak Ridge Leadership Computing Facility in Tennessee, a semi-carbon-neutral system, the same run would have produced emissions estimated at 7.2 flights. On CSCS’s new “Alps” supercomputer in Lugano, the group could perform four times as many simulations in the same period thanks to the machine’s efficient Grace Hopper 200 Superchips—while still keeping CO₂ output down to the equivalent of 2.3 flights, due to the Swiss grid’s reliance on low-carbon energy sources. In addition, CSCS’s uses a sustainable lakewater-based cooling system.

Sustainability as a comprehensive concept

Other sessions addressed sustainability from the standpoint of software usability, accessibility, and visualisation of meaningful results for policymakers and the public. Paola Cinella, professor at Sorbonne University, for example, explored the challenge of making aviation more sustainable. Air traffic is projected to rise by up to 80 percent in the coming years, with long-haul flights contributing the lion’s share of emissions. “Small improvements won’t cut it—we need radical aircraft redesign,” she said. High-fidelity simulations of novel engine designs like open fan rotors are critical but come with a hefty carbon cost themselves. Here, machine learning offers a potential solution by accelerating simulations and allowing researchers to construct high-fidelity surrogate models that preserve accuracy at a much lower computational cost.

The role of AI in sustainability was further explored in a minisymposium presentation by Yaroslava Yingling, professor at North Carolina State University. Her group combined AI and molecular modelling to tackle phosphorus pollution in North Carolina’s Neuse River, which had caused massive fish deaths. Their model identified optimal locations to deploy newly designed nanomaterials for water cleaning—minimising both environmental impact and cost.

The conference’s public lecture, delivered by Alberto Figueroa, professor of surgery and biomechanical engineering at the University of Michigan, shifted the focus to personalised medicine. He presented how simulations help assist clinicians in treating coronary artery disease in ways tailored to individual patients. Figueroa also discussed the challenges of developing tools for clinical environments, where scientific rigour must meet safety and usability in patient care.

Sustainability was also the theme of this year’s best paper, presented by Katherine Arneson from the University of Illinois Urbana-Champaign. She pointed out that scientific software is often powerful, but not always usable. “It tends to overlook the needs of domain scientists,” she said—a reminder that human usability is part of sustainability too.

A healthy debate on the future of supercomputing

The conference’s themes came together in the final panel discussion on the last day, where experts debated the role of HPC in achieving the SDGs. Sally Ellingson from the University of Kentucky’s Markey Cancer Center emphasised that AI is already helping build digital twins in medicine, but “for truly personalised treatments, there’s still a long way to go.” James Hetherington of University College London recalled how supercomputers helped make life-saving decisions during the COVID pandemic and stressed the need for secure access to patient data.

The panel also explored the idea of “climate-aware” scheduling. Jobs that need more resources could be timed to coincide with cleaner energy availability—for instance, solar power in California during the day, unlike gas-dominated power during the night. Hetherington added that researchers have an obligation to write efficient code to limit resources and environmental cost.