Author: Gayle Gaddis
Now in its fifth year at Princeton, the Open Hackathon (formerly the GPU Hackathon) marks a rare opportunity for researchers at the University and beyond to tap into a vast computing community — and with it, the future of research.
The four-day, hybrid event, which finished June 23, paired seven research teams from Princeton, the University of San Diego, Environment and Climate Change Canada, and the University of Maryland College Park with expert mentors to accelerate their understanding of high-performance computing and the revolutionary applications it makes possible.
A Quick Hackathon History
You may think it’s about information security, but it’s not. It all began at Oak Ridge National Laboratory (ORNL) in 2013. The lab had just unveiled its first GPU-based supercomputer to researchers, for whom it was both a marvel and a mystery. While running code on GPUs allowed them to see research results three times faster than with traditional central processing units (CPUs), the users were completely unfamiliar with the architecture. They would have to be trained on a new programming model. In putting that call out to industry and the developer community (to whom “hacking” simply meant programming), the first GPU Hackathon was born.
Since 2018, it has been a partnership between the Princeton Institute for Computational Science & Engineering (PICSciE); OIT Research Computing and academic departments; NVIDIA, a pioneer in accelerated computing; and OpenACC, a nonprofit dedicated to the research and developer community. As Professor Jeroen Tromp, PICSciE Institute director, put it, “The Princeton Open Hackathon offers a valuable opportunity for scholars to enhance and advance their research. By transitioning from traditional CPU-based computing to utilizing GPUs, codes can experience significant acceleration, enabling the exploration of new areas of research.”
Today, the hackathon is still a collaborative, mentor-led environment for accelerating and optimizing scientific applications, but has now become the Open Hackathon, broadening its purview to include not just GPUs, but multicore CPUs and other accelerating technologies.
Industry and Academia in a Symbiotic Relationship
As one of ORNL’s partners in the very first hackathon, OpenACC has long been sharing knowledge that advances scientific breakthroughs by helping researchers and developers expand their accelerated and parallel computing skills. Including hackathons, OpenACC hosts about 70 different learning opportunities yearly, from GPU bootcamps, to summits, to mentor training and certification. It’s what made them an ideal partner for Princeton and its academic mission. Together with NVIDIA, who has provided leading edge tools and talent since the beginning, they form a kind of dream team, who’ve come together to — in the words of Julia Levites, strategic lead for the Open Hackathons — “include everyone who wants or needs help in accelerating their codes or their science.”
This year, that science ranged from physics and astrophysics to climate and ocean modeling. For instance, Princeton’s MagNet team came to the hackathon to build a better mousetrap, effectively, in the field of power electronics. Using neural networks and AI techniques to model the performance of a certain magnetic component — an inductor — they hope to simplify the design of electronic devices. Mentored by experts from NVIDIA, from whom they learned “countless GPU computing techniques,” and running their training model on Princeton’s GPU computing cluster for the first time, the team was surprised by the much higher performance speeds they saw. It “allowed us to experiment more than we ever have in the past” according to teammates Wonju Lee and Edward Deleu. “Our experience at the Hackathon completely exceeded all of our expectations.” The MagNet team walked away from the Hackathon with their code running more than five times faster.
The Mentoring Ripple Effect
That diversity of perspective carries through to the mentors who make the hackathon model possible. Drawn from universities, supercomputing centers, national laboratories and more — experts in a specific programming language or scientific domain or machine learning — the pool of mentors has grown to more than 830 over the last 10 years. What motivates them may vary somewhat: They may want to augment their own skillsets in accelerated and parallel computing. Or explore realms of science that their day job doesn’t touch on. Or mentor for professional development reasons.
What doesn’t vary, however, is the shared love of science. As Izumi Barker, hackathon program manager, put it, “when you can take your passion and see it serving society — advancing science and making a real-world impact — what’s better than that?”
That passion, as it turns out, is contagious. Mentors often will have been program participants themselves, or recommended by a colleague who had been a mentor. Kamesh Arumugam, now senior developer technology engineer at NVIDIA, first became involved as a PhD student, encouraged by his advisor, an active mentor for hackathons at NASA. Five years later, the excitement is still there. “It’s not like a normal workday; it’s exposure working with different domains I would never have seen otherwise.” As for how the hackathon has grown? “We go into it with another motivation…hoping for a long-term engagement with the [research] team.”
Curt Hillegas, Princeton’s associate CIO of Research Computing, added “Open Hackathons are a unique and compelling opportunity for researchers to learn directly from the experts in these forefront technologies, and to build a community around translating science into codes that leverage these powerful resources.”
A Community at The Heart of Scientific Discovery
Ultimately, the hackathons are about more than coding skills. Before COVID provided the impetus for a hybrid program model, teams would all work in the same large open space, and present on their progress to all the participants — turning the room into one massive, collaborative resource pool.
It’s an ideal formula for community, based on knowledge exchange and shared opportunity — opportunity that many researchers might not have otherwise. According to Chong Chong He, from the University of Maryland College Park team, “Everyone had access to supercomputers for free, for a sufficient amount of time, during the Hackathon. That part is particularly great, because not everyone has access to those resources.” It was access that allowed his team, whose research focuses on the gravitational attraction between galaxies, to radically reduce the number of hours running code. Calculating the interaction between millions of stars on CPUs could take a million hours, or scaled down, 1,000 hours. Optimizing their code on GPUs cut that number down to 50 hours.
Today, though some elements may now be virtual, the hackathons are more than ever a cradle of community. Their impact extends outside any individual program, to the entire global research community. As Levites put it, “The goal is not just to accelerate the science; it’s to train these researchers and make sure they build relationships with people who aren’t in the room.”