Center for Translational Research in Medical Image Computing

Overview of Services

The primary computing resource at CBICA is an Enterprise Linux-compatible HPC cluster.

The cluster is connected to the University of Pennsylvania Hospital System network (HUPNet), secured by the hospital system firewall.

The cluster is housed in secure datacenter, with 24x7 monitoring, redundant power, cooling and network connections.

The cluster is used for interactive work and software development, as well as batch processing of image data. Image processing jobs in the lab use the SchedMD Slurm queuing system to efficiently manage resources for both parallel and serial computing jobs.

The CUBIC cluster consists of the following resources for batch jobs:

• 169 servers
• 7884 CPU-cores
• 67TB of RAM
• 118 (3x H100, 22x A100, 1x L40, 26x A40, 4x V100, 62x P100)

The nodes have a total of 200TB of local storage for installed software, temporary data, and network filesystem caching.

To support applications that require large amounts of memory, 3 batch servers have 3TB of RAM each.

The cluster has nodes dedicated to interactive work that can be used for software development, visualization of results, and submission of jobs to the batch computing nodes, with the following resources:

•  4 servers
• 160 CPU-cores
• 1.9TB of RAM
• 6 GPUs

Servers communicate directly via a 10Gb/sec network to a 100Gb/s network core, and have access to approximately 2.4PB of highly-available storage.

Several highly-available infrastructure servers manage data backups, a source code revision repository, an internal network information system, the CBICA public website, and internal documentation systems.

A rich software environment is provided, with over 3,000 software packages from the CentOS repositories and about 600 additional specialized packages for image processing, deep learning, and scientific computing.

Contacts