FPGA-Based Multiprocessor Architecture for LQCD Computations

79377S The Department of Energy (DOE) requires the development of mesh-communication-optimized cluster interconnects, scalable to thousands of nodes at modest cost, to perform theoretical calculations of Lattice Quantum Chromo Dynamics (LQCD). Supercomputer architectures has been the solution of choice for meeting the computational demands, but their costs are prohibitive. Clusters would be an appropriate platform for these calculations because of their low cost per compute node, but only if the cluster interconnects were of high bandwidth and low latency. This project will develop an entirely novel approach, which calls for the development of infrastructures to run the LQCD algorithm on field programmable gate arrays (FPGAs). By taking advantage of increased clock rates, built-in mathematical functions, inclusion of high-speed I/O, and integrated processors, FPGAs could provide a fast, reliable, low-cost solution. Phase I will demonstrate the feasibility of accelerating the LQCD algorithm on FPGA-based compute nodes. Phase II will develop a prototype compute node. Commercial Applications and Other Benefits as described by the awardee: The first beneficiaries of this technology should be scientists in national laboratories and universities that are working on QCD-related problems and other various large-scale computation problems. The next set of customers likely would be DoD aerospace contractors and commercial companies that run large-scale simulations.