Betty
Betty is the University of Pennsylvania’s first university-wide high-performance computing (HPC) and artificial intelligence (AI) cluster, marking a transformative step in the campus’s research infrastructure. Designed to support the most demanding computational workloads across disciplines, Betty unifies HPC and AI resources into a centrally managed system available to all Penn schools and centers. With cutting-edge CPU and GPU capabilities, including a DGX SuperPOD and thousands of Zen 4 cores, Betty provides a scalable foundation for innovation in science, medicine, engineering, and the humanities.
DGX B200 SuperPOD
The primary AI system was provided by NVIDIA and built against their reference specifications for a SuperPOD. A dedicated NDR400 infiniband fabric (“Compute Fabric”) allows for scaling of a single experiment to the whole SuperPOD. The system has (31) 8-way GPU nodes.
| Component | Specification |
|---|---|
| GPUs | 8× NVIDIA Blackwell B200 GPUs with NVLink and NVSwitch |
| GPU Memory | 180 GB HBM3e per GPU (1.4 TB total per DGX B200) |
| Peak/Sustained Performance | 9.459 PFLOPs / TBD |
| CPU | (2) 2.1 GHz Intel Xeon Platinum 8570 processors, 112 cores |
| Total CPU Cores | 3474 cores |
| Total Accelerators/GPU Cores | 36,704 |
| CPU Memory | 2TB DDR5 |
| Storage (local) | 30.72 TB local NVMe storage |
| Networking | 200 Gbps Ethernet 2× ConnectX-7 NDR400 Storage Fabric Connection 8x ConnectX-7 NDR400 Compute Fabric Connection |
CPU Standard Memory
The primary classical HPC system was provided by Dell EMC and powered by AMD processors. Interconnected to the storage and other compute nodes through a shared NDR200 link from NVIDIA. These 64-compute nodes and storage are connected using an Infiniband NDR200 network from NVIDIA with 1:1 topology.
| Component | Specification |
|---|---|
| CPU | (2) AMD EPYC 9374 |
| Total Cores | 64 cores |
| Base / Boost Clock | 3.85 GHz / up to 4.3 GHz |
| Peak/Sustained Performance | 252,313 GFLOPs / TBD |
| Memory | 384 GB DDR5 ECC Registered |
| Storage (local) | 800 GB local NVMe SSD |
| Networking | 25 Gbps Ethernet 1× ConnectX-7 NDR200 Storage Fabric Connection |
CPU Large Memory
These compute nodes provide extra memory for applications that have additional in memory requirements. There are (10) of these compute nodes.
| Component | Specification |
|---|---|
| CPU | 2× AMD EPYC 9374F |
| Total Cores | 64 cores |
| Base / Boost Clock | 3.85 GHz / up to 4.3 GHz |
| Memory | 1152 GB DDR5 ECC Registered |
| Storage (local) | 800 GB local NVMe SSD |
| Networking | 25 Gbps Ethernet 1× ConnectX-7 NDR200 Storage Fabric Connection |
The History Behind “Betty”
Photo of Frances Elizabeth “Betty” Snyder Holberton
Betty Holberton working on the ENIAC
Frances Elizabeth “Betty” Snyder Holberton was a pioneering mathematician and one of the original six programmers of the ENIAC, the world’s first general-purpose computer, developed during World War II. Betty played a key role in in translating complex mathematical problems into machine-executable instructions, laying foundational work for modern programming and software engineering.
Naming the cluster after Holberton honors her legacy as a trailblazer in computing history and highlights the critical, often under recognized, contributions of women in early computer science.