CC* Compute: Augmenting a 2,560-core EPYC2 Computational Cluster with GPUs for AI, Machine Learning, and other GPU-Accelerated HPC Applications

点击翻译按钮获取中文摘要

This project augments an existing computing cluster at the University of Tennessee Chattanooga (UTC) with 36 NVIDIA A100 80GB Graphical Processing Units (GPUs) for 18 of the existing servers. This upgrade provides over a threefold increase in performance for those computers on many workloads, and even higher speed improvement for certain computational areas, such as artificial intelligence problems. GPUs are the best way, at present, to achieve extremely high computational performance cost-effectively on today's servers, workstations, and desktop systems. Adding GPUs to the existing servers is a straightforward upgrade process. The upgrade enables 13 science drivers, or subprojects, spanning a range of domains and specialties, including researchers at UTC and among collaborating institutions nationwide. Undergraduate and graduate students benefit from using the upgraded computing faculty implemented through this project. The 13 science drivers pursued in this project support current funded and unfunded research in addition to teaching activities for both undergraduate and graduate students. Also, external collaborators at the University of Alabama, University of New Mexico, and Worcester Polytechnic Institute will utilize a significant fraction of this scalable computing resource, usually in collaboration with UTC researchers. Expected users include the more than 40 computational science PhD students at UTC, plus postdocs, Masters students, and undergraduate researchers. The science-driver projects complement existing uses of the cluster while emphasizing GPU-accelerated research and creative activities, which are specifically enabled by the GPU upgrade supported by this funding. These science drivers include: Data-driven Methods for Predictive Intention Models for Drivers and Pedestrians; Above Ground Carbon Sequestration Using GIS and Remote Sensing for Chattanooga; Portable Performance Optimizations for Irregular Communication; Identifying Metabolites from the Data of Tandem Mass Spectrometry; and Molecular Dynamics Simulations of Active Filaments.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.