Virginia Tech Advanced Research Computing expands Owl supercomputing cluster to accelerate research
Expansions to CPU, memory, and GPU capacity accelerate computational research and large-scale artificial intelligence across the university.
Three men stand in front of a supercomputer that they just brought into operation.
Virginia Tech’s Advanced Research Computing group (ARC) within the Division of Information Technology has added CPU (central processing unit) memory, and GPU (graphics processing unit) resources to the university’s Owl high-performance computing cluster, and researchers are already reaping the benefits. The additional capacity substantially increases access to computing resources, helping meet demand for high-performance computing. The number of ARC users increased by approximately 30 percent year-over-year, and GPU utilization grew by 62 percent in 2025.
The expansion adds 76 CPU nodes each with 96 cores and 768 gigabytes of memory; eight large-memory nodes with three terabytes of memory each; and four GPU nodes equipped with eight NVIDIA B200 GPUs apiece. In total, the expansion brings the total size of the Owl cluster to 17,408 CPU cores, 174 terabytes of memory, and 32 NVIDIA B200 GPUs – doubling CPU capacity and adding groundbreaking new capabilities to expand the boundaries of the artificial intelligence frontier.
“The new resources are intended to help researchers spend less time waiting for computing capacity and more time advancing their work,” said Alberto Cano, associate vice president for research computing. “The expansion gives Virginia Tech additional flexibility to support a broad range of research, from large-scale simulations to artificial intelligence training and inference, while continuing to manage university resources responsibly.”
Owl HPC liquid-cooled compute node.
Additional capabilities impress researchers
The expanded capacity already is helping researchers plan and execute computationally demanding projects. Graduate research assistant Jianpeng Chen is working with Dawei Zhou, associate professor of computer science, on a project that is developing new microscale materials in a project supported by the Defense Advanced Research Projects Agency.
The GPU resources are particularly important for their research. “The addition of NVIDIA B200 GPUs has been transformative, providing both significantly greater computing capacity and faster training capabilities,” Chen said. “These resources are enabling us to successfully train large-scale AI models for microscale materials design, accelerating scientific discovery, and helping translate advanced machine learning techniques into next-generation materials innovation.”
The work combines CPU-based density functional theory simulations with large generative foundation models to study materials across extensive three-dimensional design spaces. Use of the models allows researchers to investigate complex relationships between material structures and their properties.
Other researchers rely on ARC resources for computational fluid dynamics simulations. William S. Cross professor of engineering, Danesh Tafti, and members of his research group use high-performance computing to study problems in turbine and rocket propulsion, harvesting energy from renewable sources, as well as biological and biomedical systems within the department of mechanical engineering. Their work includes large-scale simulations of shocks interacting with turbulent boundary layers in supersonic flows, ocean wave-energy harvesting structures, bat flight dynamics, gas-solid fluidized beds, capillary hemodynamics, and multiphase gas-liquid-solid flows in various systems.
The group currently has seven students who rely on ARC resources. One graduate research assistant said that large jobs requiring hundreds of cores can take several days or more than a week to start running, limiting the ability to test code and iterate on research. Additional CPU capacity is expected to reduce those delays and improve turnaround time for large-scale jobs.
A deeper dive into the expansion
Owl’s CPU resources support research that requires large numbers of cores, substantial memory, or both. The new large-memory nodes will accommodate workloads that cannot be efficiently run on standard systems, while the additional CPU nodes reduce wait times for long-running, high-core-count jobs.
The GPU expansion supports large-scale artificial intelligence and machine learning workloads, including model training and inference. The new NVIDIA B200 GPUs provide 180 gigabytes of high-bandwidth memory and approximately eight terabytes per second of memory bandwidth. They also support FP4, a four-bit floating-point format that can improve processing speed for some AI workloads. FP4 breaks data up into smaller, simpler segments, allowing the chip to process the information up to four times faster than before.
All new nodes use liquid cooling, allowing them to operate at higher speeds while reducing the energy required for cooling. The approach builds on Owl’s existing direct-to-node cooling design, which is intended to improve energy efficiency and reduce thermal throttling.
ARC resources are available at no cost to the university research community. In addition to providing computing systems, storage, and visualization resources, ARC offers consultation and technical support to help researchers select appropriate systems, optimize workflows, and make effective use of available software and hardware.
The expansion was made possible through a $5 million university investment in fiscal year 2026. Virginia Tech plans to continue supporting the growth of research computing with an additional $6 million investment in fiscal year 2027.
Researchers interested in using ARC resources or discussing the computing needs of a project can contact ARC for consultation and support.
Advanced Research Computing, a unit of the Division of Information Technology, provides high-performance computing systems, storage, visualization resources, and consulting services for Virginia Tech researchers. ARC’s computational scientists, systems engineers, and developers support the university community in using advanced computing resources and software for research across disciplines.
