Penn needed shared AI infrastructure that could support researchers across many disciplines without weeks-long wait times, fragmented storage, or the cost barriers that kept frontier-scale training and inference off campus.
Reported impact
Penn cut wait times from weeks to hours, delivered a 12x speedup in vision-language-action model training, raised experimental throughput 14x, reduced a 40,000-document analysis project from 1.5 years of compute to three days, and now serves about 950 researchers across 200 groups with 120,000 GPU hours per month at roughly $1 per GPU hour.
Penn built a university-wide AI research platform called Betty on NVIDIA DGX SuperPOD so researchers across robotics, biomedicine, genomics, and other fields could use the same high-performance infrastructure. The deployment combines NVIDIA DGX B200 systems, NVIDIA InfiniBand networking, a VAST Data 15x10 array for multi-petabyte storage, and dark-fiber connectivity back to campus through a Flexential colocation site. NVIDIA Mission Control provides cluster-wide observability, while the scheduling layer uses both Slurm for traditional HPC workloads and NVIDIA Run:ai for AI-native training and inference. This architecture lets Penn ingest large research datasets, keep GPUs fed at high throughput, and serve many labs from one governed platform instead of isolated departmental clusters.
Research datasets move into the VAST Data array and are accessed over InfiniBand by DGX B200 nodes inside the DGX SuperPOD. Slurm and Run:ai schedule different classes of jobs onto the shared cluster, while Mission Control monitors health and utilization so operators can optimize experiments and return capacity to the wider Penn research community.
That platform is Betty, built on an NVIDIA DGX SuperPOD with NVIDIA DGX B200 systems.
The cluster runs both Slurm for traditional high-performance computing workloads and NVIDIA Run:ai for AI-native workloads, including inference services.
Today, Betty serves approximately 950 registered researchers across 200 research groups, delivering 120,000 GPU hours and 30,000 jobs per month at a subsidized cost of roughly $1 per GPU hour.
VAST Data 15x10 array provides more than 250 GB/s of read throughput and 60 GB/s of write throughput across multiple petabytes of data.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Aug 27, 2026 Texas A&M University Texas A&M needed to give researchers across seven institutions access to shared GPU infrastructure for AI and simulation, but isolated workstations, slow grant-funded clusters, and long queues were limiting… $1M saved on a single project DDN high-performance storage · NVIDIA Base Command Manager · NVIDIA DGX SuperPOD NVIDIA Sample Aug 3, 2026 StudyFetch StudyFetch needed to keep AI tutoring affordable while scaling live lecture transcription and personalized learning for millions of students, but off-the-shelf cloud transcription created cost and latency ceilings that… 100M+ learning interactions NVIDIA H100 GPUs · NVIDIA L40S GPUs · NVIDIA NIM NVIDIA Sample Jul 30, 2026 Stanford University Stanford needed a campus-wide GPU platform that could support interdisciplinary AI research at production scale without forcing labs to depend on fragmented clusters, complex allocations, or external resources for large… 18 million GPU hours NVIDIA Base Command Manager · NVIDIA DGX SuperPOD · NVIDIA Hopper GPUs 