New testbed Rowdy now available
The Rosen Center for Advanced Computing (RCAC) is excited to announce the addition of a new testbed named Rowdy to its computational resources. Rowdy serves as a technology testbed designed for the evaluation and benchmarking of cutting-edge computing, networking, and storage architectures, including accelerators and CPUs.
Rowdy aims to advance computational research by allowing invited users to test and explore new technologies in a controlled environment. Access to Rowdy is currently by invitation only. Researchers with specific use-cases are encouraged to reach out to rcac-help@purdue.edu to request access. Given the experimental nature of the cluster, users should note that there are minimal expectations regarding reliability and availability.
In the future, Rowdy may be used for AI model training or GPU computing with large datasets, providing researchers with valuable insights and opportunities to innovate in these rapidly growing fields.
The system is named in honor of "Rowdy," an inflatable mascot that briefly resided at Purdue, reflecting the spirit of experimentation and fun that the cluster embodies.
At this stage, Rowdy consists of (4) Nvidia Grace Hopper nodes and (2) Grace/Grace nodes, along with an HPE ARM Ampere login node. What is unique about this platform is the NVLink-C2C memory coherency, allowing CPU and GPU threads to concurrently and transparently access both CPU and GPU resident memory locations. Additionally, code running on the GPU has the potential to access around 480GB of system memory, as well as local 96GB of GPU memory.
Rowdy nodes also run Rocky Linux 9 and use Slurm (Simple Linux Utility for Resource Management) as the batch scheduler for resource and job management. The application of operating system patches occurs as security needs are presented. All of the nodes allow for unlimited stack usage, as well as unlimited core dump size. However, disk space and server quotas can still pose as limiting factors to keep into consideration.
RCAC operates a significant shared cluster computing infrastructure developed over several years through focused acquisitions using funds from grants, faculty startup packages, and institutional sources. These were named "community clusters". Currently, they now serve as the foundation of Purdue's research cyberinfrastructure.
"Rowdy is intended to help us pave the way toward future implementations of our production hardware,” says Chris Orr, senior research solutions engineer for RCAC.
“Hardware and concepts tested here will allow us to ‘fail fast’ or ‘realize success’ on a smaller scale. As time goes on, we’ll see new storage and AI accelerator concepts being implemented here.”
To learn more about Rowdy and other RCAC computational resources, contact rcac-help@purdue.edu.