The project funded the acquisition of a high-performance computing cluster named Hive to support data-driven research in multiple fields including astrophysics, data-enabled chemistry, biology and health, and materials science and manufacturing. The Hive cluster consists of 484 Xeon compute nodes and 2.5 petabytes of storage, connected by 100 GB/s EDR Infiniband interconnection network. Some of the nodes are enhanced/specialized to meet different application requirements ? 16 nodes with quad V100 NVIDIA GPUs, 16 with large local disks, 16 with large local Solid State Disks (SSDs), and four with much larger main memory. In addition, a 16 node HPE/Cray NSP-1 system with Fujitsu 48-core A64FX ARM-based processors and 32 GB High Bandwidth Memory is procured. The Hive cluster is used as primary resource for computation and data-enabled research of 35 faculty members and their research groups totaling about 250 users during its span of operation. In addition, the system is made available via the NSF XSEDE network by deploying the Hive gateway using SciGaP?s Apache Airavata framework.

Intellectual merit: The project funded instrumentation that simultaneously impacted multiple disciplines. Research in astrophysics has contributed to our understanding of binary black holes, gravitational waves, and the first generations of stars and galaxies in the early Universe. Computational chemistry research created detailed machine learning models for numerous reactions and materials. The research has provided further insight into electrochemistry, filling in some details of fundamental processes by atomistic simulation. Hive allowed for the first systematic investigation of the range-dependence of intermolecular interactions in molecular crystals, which should help theorists who are designing improved algorithms for rapidly computing accurate lattice energies. Multi-scale datasets generated by materials science researchers on Hive are now allowing us to harvest a new class of materials knowledge systems that are focused on accelerating the rate of materials discovery, development, and deployment in practical advanced technology applications, directly in support of the national Materials Genome Initiative. Finally, research in high performance computing led to the development of parallel algorithms and software tools for learning graphical models, constructing tensor decompositions, and carrying out analytics on hypergraphs.

Broader impacts: The project led to multiple open-source software products that contributed to the high-performance computing software ecosystem that exists in several science and engineering disciplines. For example, the SPARC DFT software helps accelerate the speed of computational chemistry research. Research in materials science supported the continued development of www.pymks.org as an open-source repository for codes capable of establishing process-structure-property linkages needed to accelerate materials innovation efforts (in direct support of the federal Materials Genome Initiative). Researchers also contributed data and software to Materials Commons (https://materialscommons.org/), and continued the development of www.materialhub.org as an open-access sharing repository for material microstructure datasets. During Covid, some capacity of the cluster is redirected in support of scientific work to fight the pandemic. Research studies conducted on Hive of the main protease in the SARS-CoV-2 virus revealed that binding of inhibitors can shift the dominant protonation states, an important consideration for drug design efforts.

The project also created valuable datasets that are useful in the respective scientific communities. For example, the computational chemistry datasets generated are being utilized by theoretical chemistry methods developers and for machine learning applications in helping parameterize, train, or calibrate new methods. The knowledge created using the Hive cluster was critical in establishing the start-up company Multiscale Technologies, Inc., which provides software and hardware solutions aimed at accelerating materials innovation/design cycles for its customers using emergent AI/ML and high-throughput strategies.

In addition to supporting research by Georgia Tech investigators, up to 20% of the capacity of the Hive cluster is set aside for external users, with particular focus on HBCUs and MSIs.

Last Modified: 04/22/2023
Modified by: Srinivas Aluru