Penn’s newest supercomputer is transforming research

The Penn Advanced Research Computing Center (PARCC) is a centralized core facility that supports high-performance computing, AI, and data-driven research across the University. It reflects Penn’s commitment to accelerating impactful, interdisciplinary research through strategic leadership, operational support, and an administrative framework.

A key feature of this ecosystem is “Betty,” Penn’s first University-wide HPC and AI cluster—a group of connected computers that function as a single, far more powerful machine. Launched in May 2025, Betty ranks among the top supercomputers in the world—189th globally—and places in the top ten among academic institutions.

View large image

Named for Frances Elizabeth “Betty” Snyder Holberton, one of the original programmers of Penn’s Electronic Numerical Integrator and Computer (ENIAC)—the world’s first programmable, general-purpose electronic digital computer—Betty was designed to support the most demanding computational workloads across disciplines. The system enables researchers to run complex simulations, analyze massive datasets, and train AI models far more quickly than on a typical desktop computer.

Less than a year after its launch, Betty supports approximately 400 users across 70 research groups, including around 200 active users—some of whom are new to high-performance computing research. Notably, it is reshaping research at Penn, expanding access to powerful computing resources, enabling work that was previously out of reach.

“Betty is accelerating research across the spectrum. We have examples of computing times dropping from months to days and from weeks to hours,” says David Meaney, vice provost for research. “Just as importantly, faculty are now tackling problems that they thought they could never tackle, at speeds they could not imagine.

Penn Today spoke with Michael Borda, associate vice provost for research; Jaime Combariza, PARCC’s inaugural executive director; and Kenneth Chaney, associate director of AI and technology, about the research PARCC has already supported, its full opening to the Penn community, and what’s ahead.

‘Pilot’ phase highlights PARCC’s strengths: Access and democratization

Betty expands and accelerates computational research on campus, enabling it to run faster and more efficiently, says Borda. “By dramatically reducing computing time, we are opening the door to research previously unimaginable,” he says. “It allows researchers to think differently and ask entirely new kinds of questions.”

Previously, high-performance computing at Penn was siloed; three Schools—the Perelman School of Medicine (Penn Medicine), the School of Engineering and Applied Science (Penn Engineering), and the School of Arts & Sciences (SAS)—each had a cluster, but they were small, and researchers needed more power. In addition, the other schools had limited, if any, access to this type of computational power.

“This was one of the challenges we sought to address when we first started this initiative,” says Borda. “One of our stated goals was to democratize access to high-performance computing on campus.”

Now, all Schools not only have access to the computing strength of Betty but also do at a fraction of the cost of an independent cluster—a “win-win,” says Combariza.

View large image

One DGX B200 node—a high-performance AI server—in Betty’s DGX B200 SuperPOD costs hundreds of thousands, so Schools would need to invest millions of dollars to build their own cluster.

“Now they can get started for just dollars,” says Chaney. There is no entry fee for faculty members, and the cost of running an experiment is just $1 per GPU hour and $0.01 per CPU hour.

Combariza and Chaney also stress the next-level user support that PARCC provides.

“We help researchers use the system efficiently—from installing the applications to troubleshooting problems,” says Chaney. He points to two examples: PARCC helped researchers at Penn Carey Law, up against an end-of-week deadline, reduce a monthlong experiment to just eight hours. A Wharton project estimated to take three years to run was completed in a day and a half. “Not only was the experiment now possible, but doing multiple iterations became possible,” he says.

PARCC is now open to the Penn community, and planning for growth

The response to PARCC has exceeded expectations, says Combariza, pointing to the uptick in requests to open accounts from new faculty members. And he anticipates that interest to grow “as they find out how much faster they can run their research.”

PARCC is also fostering interdisciplinary research, says Chaney, with collaborations forming organically across teams that have not previously worked together. He cites new partnerships between Penn Engineering and Penn Medicine, as well as a joint effort between Penn Engineering and SAS—using Betty to tackle complex math problems with a language learning model.

The team is already planning PARCC’s expansion.

“As more faculty members become interested in using the facility, we need to ensure enough resources so they don’t have to wait,” says Combariza.

But it is not just about capacity. “Researchers are pushing the boundaries of what's possible with these systems, tackling new problems and looking for solutions. Our job,” he says, “is to figure out what people require to do that—new software, new hardware, and opportunities to experiment in “sandboxes"—to support their work.”

Borda acknowledges the three schools that piloted PARCC, Penn Information Systems & Computing, and the external partners who worked with the Office of the Vice Provost for Research for helping to make it a reality.

“At the end of the day, this is about empowering researchers,” says Borda. “Removing barriers to high-performance computing enables researchers to pursue ideas and discoveries that simply weren’t possible before.”