MRI-R2: Acquisition of a GPU-Based Cluster for Plasma, High-Energy Density and Computational Science at UCLA

University Of California-Los Angeles, Los Angeles CA

Investigators

Abstract

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5). Specialized computer processors, designed primarily for video games, called graphics processing units (GPUs) have the potential to revolutionize scientific computing, by providing large computing resources at low cost and low power. A cluster based on GPUs will be constructed at UCLA focusing on plasma simulations using the particle-in-cell (PIC) method. The network that links this cluster together is carefully designed so that the simulation tools of interest will run very efficiently on it. The cluster itself will consist of 96 dual quad-core (Intel i7) nodes each connected to two Nvidia Tesla GPUs (192 GPUs total) and will be housed within the Institute for Digital Research and Education (IDRE). The cluster will have three main purposes. First, the cluster will serve as a testbed for developing new parallel particle-in-cell (PIC) algorithms and codes to run on systems that have many GPUs networked together. The codes should run efficiently on future computers housed at large national, supercomputer centers. Second, the cluster will be an instrument to enable breakthroughs in new particle accelerator technology, fusion energy, space plasmas, astrophysics, and basic plasma science. Third, the cluster will help to advance research and education in broad and diverse areas of Computational Science at UCLA. The plasma physics subjects to be studied could have a broad impact on a wider range of fields that have a significant societal impact, fields such as new compact accelerator technology, fusion energy, and the reliability of communication satellites. The cluster will also have broad educational impact in that students will be able tackle problems at the cutting edge of a variety of fields, that previously would have required resources typical of a national laboratory. The Principal Investigator on this project is also the Director of the UCLA Institute for Digital Research and Education. Two of the co-PI's are IDRE staff members. Through IDRE, the cluster and the activities centered around it researchers in other science domains will learn how to compute on GPU's and the GPU cluster will also be used to train students across a wide array of computational based research topics.