The sixth, and largest, of Princeton’s Open Hackathons spanned four days, June 4 and June 12-14, while also spanning the range of issues confronting the world scientific community. A model for a more robust research computing framework, the event – and the work inspired by it – may also hold the keys to a more sustainable future.

This year’s Hackathon hosted 11 diverse research teams from academic institutions across North America, supported by 24 research computing mentors – from Princeton, accelerated computing pioneer NVIDIA, the National Center for Atmospheric Research (NCAR) and the National Energy Research Scientific Computing Center (NERSC). Coming together in-person and virtually, the teams explored the potential of high-performance computing and A.I. to help reveal mysteries of the universe, the planet, and the human body.

The Hackathon, then and now

Princeton’s Open Hackathon began in 2019 as a kind of “beta” collaboration between academia and industry. Joining forces with NVIDIA and developer community OpenACC, the Princeton Institute for Computational Science & Engineering (PICSciE) hoped that the knowledge-exchange forum would “offer a valuable opportunity for scholars to enhance and advance their research.”

Its focus, at the start, was on how GPU-based supercomputers could provide 3x the processing speed of CPUs, which had been the standard for high-performance computing (HPC). Faster processing, in turn, meant the ability to process exponentially more data, making research insights more detailed and accurate.

Now in its sixth year, the Hackathon has wildly exceeded hopes: Achieving game-changing speed-ups. Contributing to a new generation of research software engineers, in the face of an HPC talent shortage and the rapid growth of AI. And empowering the cross-pollination of ideas between different scientific domains – which often paves the way for unforeseen innovation. As NVIDIA’s Izumi Barker, Open Hackathon program manager, put it, the event “is combining the dream of endless scientific possibilities with the reality of advanced, innovative technologies…it’s a win-win.” 

Addressing our most critical challenges 

“Research computing” may be abstract to a layperson, but it may be our best path to understanding some of humankind’s most pressing issues, from climate change to disease prevalence and more. A look at some of this year’s Hackathon team projects tells the story:

• DESC – the realistic design of fusion reactors, a potentially clean, safe and abundant global energy source
• ParFlow – integrated, large-scale modeling of water cycles – from bedrock groundwater to the top of the plant canopy – used in climate change projections 
• ASPIRE – exploring biomolecules at near-atomic resolution, helping better explain the characteristics of proteins, viruses and more 
• SlimeGo – modeling the behavior of a slime mold in a petri dish to predict biological randomness, with implications from environmentalism to civil engineering

For a better sense of the extraordinary progress the Hackathon has enabled, we’ve highlighted the work of four teams below.

Deconstructing fusion to power the future 

Nuclear fusion, the same process that powers the sun, is a renewable, safe, less costly solution to the world’s energy problems – if we can safely replicate the process. Yet the plasma particle collisions that produce fusion are hotter than 10 million°C – so the challenge is how to create “thermodynamic equilibrium” within a reactor to safely contain the heat.

Team Gkeyll, which came to the Hackathon with an established numerical model for simulating plasma particle movement, was looking to see how they might speed up their code. Their framework, which aims to get plasma particles to conform to the conservation laws of physics, previously had very long simulation time and high computational costs. 

In the past, as explained by Princeton graduate student and team member Dingyun Liu, the team was only able to run one function of their model at a time on NVIDIA® CUDA® (a parallel computing platform and application programming interface). During the Hackathon, they were able to run three parallel functions on CUDA using NVIDIA cuBLAS, a GPU-accelerated library for AI and HPC applications – a pivot the team called “so successful, so quickly, that we got excited!” It led to immediate speedups of 3-4x, and prompted them to apply the same approach to other aspects of their code – where they saw an immediate 15x speedup by extending their GPU use. Liu, who was new to CUDA, called the mentoring at the Hackathon an “immersive…very important step” toward advancing their projections.

A sea change in mapping blue carbon repositories

The challenge team SeagrassML (short for Seagrass Blue Carbon for Machine Learning) grappled with was how to scale a successful seagrass mapping model that used satellites to gather data. It’s important because seagrasses – like other “blue carbon” sources such as mangroves and salt marshes – capture and store CO2, thus removing it from the atmosphere. Developed at the US National Oceanic and Atmospheric Administration (NOAA), the model would ideally be shared with researchers in other regions of the world who don’t have the same infrastructure and training. 

According to Dr. Megan Coffer, team leader and research scientist for both NOAA and Global Science & Technology, Inc., the challenge was to train their model to generalize across diverse satellite images, rather than training it on a single image at a time – “to have one person do that for 20 scenes could take months.” The Hackathon was an opportunity to learn how machine learning could make it easier to look at more scenes and generalize across large amounts of commercial satellite data.

And while the work is ongoing, the event did not disappoint. “We set off with this question, not knowing if there would ever be a solution,” Coffer said. “To see a glimpse of a solution was really exciting. It was the motivation we needed to keep making progress.” A highlight of the event for her? “Some of the people I interacted with [there], I would not have interacted with otherwise…all these experts in machine learning, and our mentors, who were wonderful.”

The SeagrassML team reflected the changing climate in another way as well: as the first female-led team at the Hackathon. The fact was not lost on its cofounders, Coffer and her NOAA coworker Dr. Rebecca Trinh, who were pleased to be setting an example. “There have been many situations in science where I’m the only female,” said Coffer; “it’s exciting to be given the opportunity to show that women can lead these things and know as much as others in the room.”

Breaking the ice (code): optimizing freshwater ice mapping 

Like SeagrassML, the AICE team had been working on a machine learning application with implications for better climate modeling and monitoring. Specifically, they were looking to optimize their mapping of ice coverage on rivers and lakes – delineating ice from open water in the face of imaging challenges like cloud cover, and lengthy computational times needed to train their model.

As hoped, the Hackathon accelerated their work significantly, enabling them to update their code more frequently, process larger datasets and get faster response times. By porting from CPUs to GPUs, they achieved an approximately 16x speedup; by also transitioning between Python programming libraries – from TensorFlow to PyTorch – they reached an overall 35x speedup.

Hannah Ross, a computer scientist at the National Energy Research Scientific Computing Center (NERSC), was one of the team’s two mentors. Having been a mentor five times before, she was particularly impressed in her first time at the Princeton event. “The team was amazing, probably one of the best, if not the best, team I’ve mentored so far,” Ross said, stressing that her role was largely support for a more efficient workflow on their established model. At the same time, both she and the team gained a better understanding of machine learning from her NVIDIA co-mentor. “I do think that, in the scientific community, coding is massively undervalued, so it’s great to be able to support some of the people doing this really important work…and also learn a bunch that I can bring back to my job.”

Maximizing N: the n-body problem in astrophysics

Astronomers have come a long way from just using telescopes to study the cosmos; today, they can simulate entire universes. The more computational power you have, the more detailed the simulations can be, and the more detailed the simulations, the more you can test your understanding of the actual universe.

That’s what drew the Peytonites team to this year’s Hackathon. Their goal, coming in, was to see how using GPUs could push the boundaries when applied to the “n-body problem” – using a variable number of particles (“n”) to simulate how celestial bodies interact gravitationally. It was also important to them to continue coding in Python (the programming language astronomers predominantly use), because it’s easy to use and interact with; but since it might take hours to simulate a thousand interacting particles with existing Python code, they needed a way to speed it up significantly. As it turned out, that way was a code they wrote and optimized themselves.

Over just three days at the Hackathon, the team was able to simulate a star cluster with over a million stars using the CUDA platform and their optimized code – “a huge step up,” according to graduate student and team member Robel Geda. What’s more, in past simulations, a “particle” would represent a group of stars, whereas now, with CUDA’s greater computational power, they were able to resolve each individual star. Overall, “we ended up learning more than we had anticipated,” Geda said, including code management and new packaging paradigms (for sharing the code). In everything they learned, he gives exceptional credit to the team’s two mentors: “having someone that you know has seen all the pitfalls and can guide you, who you can reach out and contact, is a really awesome superpower.”

A sustainable model for acceleration and advancement

Thanks in part to the Open Hackathons and the research they’ve enabled, utilization of Princeton’s 1,000 GPUs is at its highest – as is the curiosity, commitment and collaborative spirit of the research computing community. The kind of thought leadership for which Princeton is well known is alive and well in the Hackathon – and will continue to thrive as long as talented, dedicated researchers and software engineers are drawn to it. For those involved – whether for the first time, or as a returning participant or mentor – the Hackathon experience can be transformative.

If you’re excited by what you see, you’re not alone, and we invite you to apply, mentor or participate in whatever way is most meaningful to your work and your community. Here, great minds don’t only think alike – they think together. Learn more at the 2025 Open Hackathon event page.