Virginia Tech Owl Cluster Expansion | Faster Research Computing

Boosting Computing Power at Virginia Tech

Virginia Tech has completed a significant expansion of its Owl high-performance computing cluster. The Advanced Research Computing group added substantial CPU, memory, and GPU resources to the system. This upgrade responds to a 30 percent increase in system users over the last year and a 62 percent surge in GPU demand during 2025. Researchers now have access to more hardware to handle increasingly complex data workloads.

The expansion includes 76 new CPU nodes, each packed with 96 cores and 768 gigabytes of memory. The university also added eight large-memory nodes featuring three terabytes of memory each, alongside four new GPU nodes equipped with eight NVIDIA B200 GPUs apiece. These additions bring the total Owl cluster size to 17,408 CPU cores, 174 terabytes of memory, and 32 high-end GPUs. This effectively doubles the university's existing CPU capacity for academic research.

Impact on Academic and Applied Research

The upgraded hardware is already supporting critical projects across campus. One such effort involves developing new microscale materials, a project backed by the Defense Advanced Research Projects Agency. Graduate research assistant Jianpeng Chen notes that the NVIDIA B200 GPUs are critical for training large-scale artificial intelligence models. These models allow for faster design and testing of materials in three-dimensional spaces, which was previously a slower process. The team expects this to shorten the gap between theoretical models and real-world innovation.

Other departments are using the increased capacity to run complex simulations. Danesh Tafti, a professor of mechanical engineering, leads a group studying turbine and rocket propulsion and renewable energy systems. His team runs large-scale fluid dynamics simulations that require hundreds of cores simultaneously. Historically, these jobs faced delays of a week or more while waiting for available computing slots. The additional CPU nodes aim to minimize these queues, allowing researchers to iterate their work much more quickly.

Technical Upgrades and Future Investment

All new nodes in the Owl cluster utilize direct-to-node liquid cooling. This design helps the hardware run at higher speeds while reducing energy consumption. By managing thermal output more efficiently, the university lowers both its operational costs and the risk of hardware throttling during intense processing tasks. The architecture supports high-bandwidth memory and the FP4 floating-point format, which lets the GPUs handle specific AI processing tasks up to four times faster than previous generations of hardware.

Virginia Tech funded this $5 million expansion through a fiscal year 2026 investment. The university is already looking toward the next phase, with a planned $6 million investment for fiscal year 2027. This consistent capital injection aims to keep the school's research infrastructure competitive on a global scale. Researchers at the institution continue to receive these resources at no cost, accompanied by support from computational scientists who assist with workflow optimization.