High-Performance Compute Cluster for Brain Science

Project Summary We request funds for a new high-performance computing (HPC) resource to serve as the cornerstone of GPU-based computing at Brown University. It will become a focal point for established and highly successful multidisciplinary educational programs and research with the Carney Institute for Brain Science. The Faculty and Senior Administration have taken steps to host this critically needed GPU resource. These include (1) hosting and management by Brown’s Center for Computation and Visualization (CCV); (2) partnership with Intel, NVidia, and IBM for training and support; and (3) development of a solid and viable computational science research community. This new resource will expand research capabilities within the multiple centers within the Carney Institute – enhancing the development of computational neuroscience models and broadening the application of machine learning and other modern data analysis pipelines to brain science, and allow unique instructional and training programs at all student levels. Brown has a proven record of collaboration between application scientists within CCV and researchers across scientific disciplines. This combination provides an opportunity to use a state-of-the-art multi-GPU system not simply as “yet another tool” but as a foundation for already planned advanced research in brain science, with examples provided in the main project description of this proposal. The acquisition of this HPC resource will directly impact the faculty listed in this proposal by increasing the speed, quality, and volume of scientific data processed and by enabling the development of computational models spanning all levels of analysis – from the level of biophysics and circuits to the level of systems and computation. Brown’s existing HPC cluster does not yet include large-memory GPU nodes. This has hindered brain science research and student training with state-of-the-art HPC technologies. The deeply interactive faculty from life sciences (neuroscience, cognitive and psychological sciences, biology, and medicine) and physical sciences (applied mathematics, engineering, and computer science) have created a unique scholarly environment of interdisciplinary activity; both students and faculty will benefit immensely from this technology. In response to our growing needs in computing, Brown has built a cluster of ~400 computing nodes, including over 300 GPUs (a mix of NVIDIA Ampere, Volta, and Turing architectures). However, the growth in the scale of the experimental data collected, the computational footprint of modern data analysis pipelines, and the scale of current computational models have severely strained our computing infrastructure. In particular, a common challenge is the high resolution, large dimension, and volume of scientific data that research groups must process, analyze, and model. The lack of large-memory GPU resources on campus and the projected growth in machine learning research needs across brain science, the demand for HPC-based research and education, and the need for on-campus access by outstanding undergraduate and graduate students, make acquiring additional GPU resources essential.