Stem Outreach
Funding

Welcome to AHPCRC Project TA4-3

Background

High-performance computing (HPC) in the guise of multi-core processors, GPGPUs (general-purpose graphics processing units), FPGAs (field-programmable gate arrays), accelerators, and solid-state devices is propelling the growth of non-traditional HPC, in particular, its use in field-deployable and on-board systems for processing sensor, signal, and image data. Such HPC-enabled systems can execute applications at extraordinary speedups, allowing some applications to migrate from larger fielded systems to smaller on-board systems, and decreasing the communication between them. Although this can translate to increased capability to meet operational timelines, this capability is limited by size, weight, and power (SWaP) constraints - and, often on-board systems have even stricter SWaP constraints than their fielded counterparts.

Goal

This project addresses many of the practical computational aspects of effectively employing emerging HPC technology in field-deployable and on-board systems. With system SWaP constraints in mind, AHPCRC researchers at UTEP are investigating the tradeoffs associated with executing Army-relevant applications, computational kernels, and alternative algorithms and implementations on different multi-core and many-core processors. They are quantifying tradeoffs concerning execution time, parallelism, precision, memory footprint, output fidelity, power consumption, and energy efficiency, with the goal of effectively mapping applications or application segments to processor architectures. Intelligent mapping of this kind can provide valuable assistance in accelerating the infusion of new technologies into on-board and field-deployable HPC systems, assessing the merit of investing time in program development, and distributing workloads in heterogeneous computing environments. The longer term goals of this work include the development of models to facilitate this application-to-architecture mapping, a model to characterize application power consumption, and a methodology to intelligently and dynamically adapt a program's execution to enhance energy efficiency and, thus, extend the effective life of system operation.