MRI: Acquisition of Cutting-Edge GPU and MPI Nodes for the Interdisciplinary Pitt Center for Research Computing

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

Computational science and engineering spans research and education across many disciplines, and state-of-the-art cyberinfrastructure resources are needed to tackle large problems and enable innovative strategies in data-enabled science and engineering. This project will greatly expand the interdisciplinary University of Pittsburgh Center for Research Computing (CRC), the core facility for scientific computing and research at Pitt. CRC supports the work of over 800 users in 59 departments across the entire university. The new expanded hardware will advance both undergraduate and graduate courses and educational experience across a similarly broad range of departments and courses. These expanded resources will enable Pitt to expand a synergy between research and education at all levels, reaching beyond the university to faculty, staff, and students at Howard University, other historically black colleges and universities (HBCUs), and many undergraduate faculty and students both regionally and nationwide. The new resources will also expand scientific computing to students in the Pittsburgh Public School district, including nearby Pittsburgh Science and Technology Academy and Pittsburgh Public Allderdice, both urban schools with diverse student populations. The funded resources will consist of 16 state-of-the-art graphics processing unit (GPU) computing nodes including NVIDIA Ampere A100 GPU accelerators. Each GPU node will be ~2x faster than previous generation GPUs and 14-50x faster on scientific computing software than current CPU nodes, and will enable increased machine learning productivity. An additional 36 state-of-the-art MPI nodes containing AMD “Milan” cores and high system memory will enable complex computational simulations. The availability of the new resource will dramatically expand the access and opportunity to GPU and message passing interface (MPI) computing, offering significant speed improvements for an immense range of scientific computing, from machine learning and big data, to quantum chemistry, protein molecular dynamics, energy conversion, nanoparticle catalysis design, weather/wind forecasting, astronomical data analysis, atomistic tunneling electron microscopy measurements, and computer vision. Beyond simple acceleration, the resources will enable transformative research with vastly more accurate weather grids, new machine learning surrogates for quantum chemical calculations of molecular and materials energies and properties, rare-event sampling in protein folding and binding, fMRI neuroscience, and next generation digital astronomy. The resources will immediately benefit over 30 NSF-funded research groups, leveraging over $18 million in research and training grants. The resources will support research in all areas including Chemistry, Computational Biology, Chemical Engineering, Materials Science, Psychology, Astrophysics, Weather Forecasting, Computer Science, and research centers focusing on energy, sustainability, and other key areas of science and engineering. Workshops and courses associated with the new resources will focus on adapting existing software and developing new software for MPI and GPU-computing including a wide range of machine learning methods enabled by the transformation in numeric processing with these expanded resources. These resources will be shared with collaborators at Howard University, other HBCUs, and with regional and national undergraduate schools to broaden participation in computational science. 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.