Supercomputing in the real world
Supercomputers are the most powerful machines of their kind in the world, capable of high-performance computing (HPC) to solve the world’s hardest mathematical problems. This isn’t just theoretical work: in the hands of researchers, supercomputers toil around the clock to further the fields of science, technology, medicine, and much more.
The main benefit of supercomputers is their ability to process numerous calculations simultaneously using parallel computing to solve complex problems very quickly. For scientists, this presents the opportunity to virtually test theories and quickly see the results – for example, simulating how a certain material could withstand space travel. The results can then be used to influence the next test, allowing for rapid, iterative improvements.
The world’s first supercomputer, the CDC-6600, was used to analyze cell photographs from early CERN experiments which tracked the movements of electrically charged particles. Today’s fastest supercomputers operate at more than a billion times the speed of the CDC-6600 and can solve calculations that would be impossible to do manually – even if the best mathematicians in the world were given the entirety of human history to do so.
All of this power can sometimes feel intangible because it’s hard to wrap one’s head around some of the problems on which supercomputers are working. Yet supercomputing actually has some standout real-world benefits, which we’re already starting to see improve science around the world.
One of the main use cases for supercomputers being talked about right now is AI training. While high-performance computing is a necessary ingredient for training generative AI models, supercomputers are also being used to push the boundaries of science in ways that can help society and all businesses.
The biggest threats in the finest detail
To model even a regional weather system, modern meteorologists use supercomputers to crunch immense amounts of live weather data to produce accurate predictions for upcoming weather events.
The newest generation of supercomputers can go far beyond the weather forecast, however. As we stare down the climate crisis, researchers and international bodies are using exaflops of processing power to create virtual replicas of the biosphere. This enables the modelling of how the climate itself is changing and can offer invaluable insights into the future of our planet.
The European Commission’s Destination Earth project, for example, aims to produce a digital twin of the entire Earth to allow for climate change modeling to a detail of 5km. This allows for incredibly granular simulations of flooding and other natural disasters arising from climate change on a region-by-region basis. In addition, the project aims to map out how the changing climate will affect biodiversity and food security so that we can prevent the worst from taking place and prepare for the unavoidable.
Finland’s LUMI supercomputer, powered by almost 12,000 AMD Instinct MI250X GPUs and more than a quarter of a million AMD Epyc CPU cores, is a key pillar of Destination Earth, having expended 10% of its total compute capacity to date on the project. It may seem counterintuitive to use a supercomputer, which are notoriously power hungry, to model climate change. AMD and HPE’s work making it as energy efficient as possible, however, has helped negate the potential climate impact: LUMI sits in the top 25 greenest supercomputers list and is powered entirely by hydroelectric energy.
If you ask a physicist to model something like climate change, they’ll ask you how detailed you’d like their model to be. In theory, there’s no limit to this as macro systems such as the Earth’s climate are made up of incalculable numbers of micro-processes. The more we can simulate, the better insight we get into the future of our planet.
To really push beyond the boundaries of current climate science we’ll need more computing power than ever before – such as is found in the Frontier supercomputer. With 1.35 exaflops of compute at its disposal, equal to more than a quadrillion calculations per second, Frontier leans on almost 38,000 Instinct MI250X GPUs and just short of 9,500 AMD Epyc 7713 CPUs.
Since coming online in 2022, it has been used to scrutinize our understanding of the universe from the spread of galaxies, all the way down to the structure and behavior of atoms and neutrinos.
El Capitan, the world’s most powerful supercomputer in the world at 1.742 exaflops, is also being used to probe the very edges of imaginable physics. Brought online in November 2024, El Capitan is largely dedicated to nuclear weapons work, such as virtual weapons tests and stockpile safety simulations. This enables Lawrence Livermore National Laboratory (LLNL) to provide the US government with up-to-date information on the state of the US nuclear stockpile without any nuclear tests necessary.
The supercomputer is another success story for AMD, relying on 11 million of its CPUs and GPUs to set new computational records.
Saving lives through virtual breakthroughs
Alongside climate change, medical research is one of the most pressing fields where supercomputing is already making a real difference.
LUMI is also at the center of important ongoing cancer research. The ComPatAI consortium, which is focused on developing advanced AI models to detect breast and prostate cancer in tissue samples, was one of five recent Finnish research projects chosen to use the supercomputer’s vast resources in 2024.
The consortium, a joint effort between researchers from the Institute of Biomedicine at the University of Turku, University of Eastern Finland, and Department of Pathology at the Fimlab laboratories, has fed nearly a petabyte of anonymized slide images from Fimlab into LUMI to help train a neural network which could help pathologists quickly and accurately diagnose patients.
Frontier has also been used by a research team at Oak Ridge National Laboratory (ORNL) to model how molecular life reacts to specific climate and genetic signals. This may allow us to alter our agricultural patterns to ensure food supplies aren’t catastrophically impacted by climate change, or even develop crops that do better in the new climates that emerge.
The same research could also open the door to tailored drug treatments that don’t put patients at risk of developing chemical dependencies. This points to a major benefit of supercomputers: just like parallel processing, research developed through supercomputers can have many simultaneous benefits for the world.
It’s hard to say which hurdles supercomputers will help us overcome next. The latest generation has been pointed at problems from nuclear fusion to designs for revolutionary new materials. What can be said with confidence is that we’re far from reaching the limits of what supercomputing has to offer – and that in the years to come, it will go further and faster than ever before.
Source link
