MRI-R2: Acquisition of a Heterogeneous Supercomputing Instrument for Transformative Interdisciplinary Research

Virginia Polytechnic Institute And State University, Blacksburg VA

Investigators

Abstract

"This award is funded under the American Recovery and Reinvestment Act of 2009(Public Law 111-5)." Proposal #: 09-60081 PI(s): Feng, Wuchun Hilu; Khidir W.; King, Scott D. Institution: Virginia Polytechnic Institute and State University Title: MRI-R2/Acq:Heterogeneous Supercomputing Instrument for Transformative Interdisciplinary Research Project Proposed : This project, acquiring a versatile heterogeneous supercomputing instrument, called HokieSpeed, in which each compute node consists of CPUs (central processing units) and GPUs (graphics processing units), aims to empower faculty, students, and staff across disciplines to tackle problems previously viewed as intractable or that required heroic efforts and significant domain expertise to solve. The instrument is expected to catalyze new approaches for conducting research via synergistic amalgamation of heterogeneous supercomputing and cyber-enabled tools that enhance ease of use. In particular, it should allow end users to: - Make commonplace the ability to perform in-situ visualization for rapid visual information synthesis and analysis, and - Control their level of immersion in the discovery process from being immersed (a la 'human in the loop' intuitively making real-time decisions via a large-scale gigapixel display), to observing the instrument automatically collect, organize, and analyze data of visual analytics. Recent trends have exposed the CPU as a 'jack of all computing trades, master of none,' thus giving rise to heterogeneous computing instruments with multiple types of brains (e.g., CPUs and GPUs). Though conventional wisdom believed in 2007 that missing genes in 699 microbial genomes was computationally infeasible, the PI led a team of more than twenty interdisciplinary researchers from eight institutions around the world and developed software cybertool instruments to integrate a set of distributed supercomputing with more that twelve thousand CPUs to complete the task in ten weeks. Coupled with the existing cybertools and those available from NVIDIA the instrument should enable completion of the task in a day with only two researchers. HokieSpeed will have 8,192 CPU cores and at least 61,440 GPU cores in 40 square feet and deliver 35 times better peak performance, 70 times better power efficiency, and 14 times better space efficiency than the current system at the institution (circa 2003) while offering versatility by enabling discovery through a large-scale gigapixel display and appropriately varying degrees of interaction between human and machine. Creating new forms of social research, and 'what if' scenarios for evaluation and promising to deliver research advances across multiple disciplines, the results will be generated often in real time allowing immediate in-situ visualization on a gigapixel display. Broader Impacts: Reaching a broad community, the instrument is expected to herald a new age in multi-purpose and interdisciplinary computing instrumentation creating a computational ecosystem that simplifies access and use of the instrument to the masses. Students, including females and underrepresented minorities, will learn about energy-efficient computing, and visualization and interaction at large scale. Among others, the project will also utilize programs that specifically service broader impacts: MAOP (Multicultural/Minority Academic Opportunities Program), CTech2 (Computers and Technology at VaTech), and CRA-W's CREU (Collaborative Research Experience for Undergraduates). Additionally, it will extend an experimental GPU course and offer a workshop version for on-line instruction. Other activities include K-12 outreach, and delivery by CREU of information technology (IT) to rural and economically disadvantaged areas via virtual computing and the VaTech-STEM K-12 Outreach Initiative.