MRI: Acquisition of a High Performance Computing Cluster for Next-Generation Computational Science in Southern Colorado

点击翻译按钮获取中文摘要

This project will enable the acquisition, deployment, and maintenance of a high performance computing (HPC) cluster (to be called INCLINE). This instrument will provide much-needed computational resources to the UCCS campus and the Southern Colorado scientific and academic communities. The size and power of the instrument will bridge the growing gap between workstation-level machines and Top 500 supercomputers, allowing researchers to test and leverage code scalability on an HPC platform, to expedite result processing, and to gain expertise on a local HPC environment. The work to be performed will yield insight into biomedical applications, such as microbubble drug delivery and bone fracture, military applications, such as additively manufactured energetics, and civil applications, including improved structural materials. In addition to research applications, INCLINE will be used as an educational platform for teaching the fundamentals of an HPC to undergraduate and graduate students and allow students to investigate classroom problems more deeply with the computational power provided by an HPC. It will also be used to supplement existing outreach programs to spark enthusiasm and interest in HPC in the Southern Colorado community, which is diverse in both population and the range of applications. The instrument’s state-of-the-art hardware is designed to support a broad range of high-performance scientific applications, ranging from compiler design to computational fluid dynamics. The instrument will contain both CPU compute (including standard and high memory), and GPU nodes. The compute nodes will enable standard massively parallel computations such as computational solid mechanics and fluid dynamics. The GPU nodes will allow acceleration on computational physics and machine learning projects and will be used in conjunction with CPU nodes to test optimal load balancing on heterogeneous architectures. All nodes will be connected using InfiniBand high speed interconnects to minimize latency for communication-intense applications, including CFD and computational solid mechanics. A high speed SCRATCH storage file system will minimize I/O latency for applications with large output file sizes and I/O requirements. The estimated peak performance of the instrument is approximately 90 TFLOPS. The Slurm queueing system will be used to manage accounts and allocations across the diverse user base. The robust design of this instrument will allow it to fill the growing need for a local HPC research facility at UCCS and in Southern Colorado and will facilitate the training of the next generation of computational scientists.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.