High-Performance Computing in the AI Era: Challenges and Future Prospects
By Bio-IT World Staff
March 5, 2025 | In an era where artificial intelligence (AI) dominates discussions in technology, high-performance computing (HPC) remains the backbone of scientific discovery and enterprise research. Despite its critical role, HPC often operates in the background, much like electricity—only noticed when it fails. On the latest Trends from the Trenches Podcast, Dirk Petersen, Director of the Supercomputing Center at Oregon State University, shared his insights on the evolving role of HPC and its integration into AI-driven research.
The Expanding Role of HPC
Petersen, speaking with host Stan Gloss, highlighted a growing trend: the increasing number of researchers who can no longer conduct their work on personal laptops. While HPC was once the domain of a small group of dedicated specialists, it has now become an essential tool for a much broader scientific community.
“Research computing used to be a certain department on the fringes,” Petersen noted. “But it has become much more strategic, much more important.” Without it, he argues, no science happens.
The demand for HPC resources is growing across disciplines, from genomics and climate modeling to materials science and astrophysics. Researchers now rely on supercomputers to process vast datasets, run simulations, and develop machine learning models that drive scientific breakthroughs. The transition from personal computing to large-scale HPC infrastructure has reshaped how researchers approach problem-solving, enabling discoveries that would have been impossible just a decade ago.
Accessibility and Usability Challenges
Despite its necessity, accessing HPC resources remains a challenge. Unlike widely available cloud services, HPC is often perceived as exclusive and difficult to use. Researchers typically require specialized knowledge to navigate these systems, often using SSH terminals and command-line interfaces—barriers that limit widespread adoption.
Petersen pointed out that in enterprise environments, users expect a streamlined experience similar to opening an application like Microsoft Excel or PowerPoint. Many researchers lack the technical expertise required to interact with HPC environments, leading to inefficiencies and underutilization of available resources.
We still don’t have a universal “Compute” button that lets anyone access HPC resources as easily as cloud services, he explained, but the situation is improving. “We’re not quite there with this button, yet. But we’re close. I think we’re getting closer and closer.”
To address these challenges, institutions are developing science gateways—web-based platforms that simplify access to HPC resources by providing graphical user interfaces and pre-configured workflows. These tools lower the entry barrier for new users, allowing researchers to focus on their work without struggling with technical complexities.
The Convergence of AI and HPC
One of the most significant developments in computing is the intersection of AI and HPC. Petersen emphasized that AI is fundamentally just another HPC workload. “For us HPC people, AI is essentially an HPC cluster with GPUs,” he remarked. “We do input, compute, and output, right? And we have done that for 30 years. Now, we do a different input and little bit different compute and output. But essentially, it's the same workflow.”
However, as AI researchers demand faster and more flexible computing environments, existing HPC architectures must adapt to new computational needs, particularly in data-heavy workflows.
The rise of AI-driven research has led to an explosion in data generation, requiring HPC systems to process petabytes of information efficiently. This shift is driving innovation in storage architectures, networking, and accelerators such as GPUs and TPUs. Additionally, AI workloads often require dynamic scaling, leading to increased interest in hybrid computing models that blend on-premises supercomputing with cloud-based resources.
The growth curve of GPUs is “still in the neighborhood of exponential,” Petersen said. Historically for CPUs, the next generation of processors might be 20% faster than your current setup. “But with AI, it’s different,” he said. “There’s a new infrastructure coming out every year, every two years,” prompting massive upgrades to both equipment and power demands. Petersen compared it to the California Gold Rush. “There’s this unlimited thing and we need to be part of that!... I feel a lot of people are getting in their covered wagons and heading west.”
The Future of Research Computing
Looking ahead, Petersen sees a future where HPC resources are as accessible as any other enterprise tool, and non-specialists can easily leverage supercomputing power. However, significant work remains in bridging the gap between cutting-edge AI development and traditional HPC environments.
The push toward democratizing HPC is also driving conversations around funding and resource allocation. As demand grows, institutions must rethink how they distribute computing power fairly among researchers while ensuring long-term sustainability. Open-source software and shared computing frameworks are playing a crucial role in addressing these challenges by fostering collaboration and maximizing resource efficiency.
As Petersen prepares to speak at the Bio-IT World Conference and Expo, April 2-4 in Boston, he aims to shed light on how research institutions and enterprises can better integrate HPC into their AI and data science workflows.
As the director of Oregon State’s new supercomputing complex—built with a gift of $50 million to the OSU Foundation from Jen-Hsun Huang, founder and CEO of NVIDIA, and his wife Lori, both of whom are Oregon State University graduates—Petersen is looking forward to having access to the latest chip types and exploring new architectures.
“I’m super excited about the fact that we will be having this machine. I can see that people on campus are just vibrant. There’s something new and there’s something big coming and people have this glow in their eyes!”