Skip to Main Content U.S. Department of Energy
PNNL: High Performance Computing

Test, Assessment and Verification Effort (TAV)
for the DARPA PERFECT Program

DARPA's PERFECT (Power Efficiency Revolution for Embedded Computing) program, initiated in 2012, is spearheading research and development of a multitude of technologies to achieve 75 GFLOPS/W on general-purpose embedded computing. Within PERFECT, multiple teams are independently developing technologies - covering all areas of the computational spectrum - to contribute to this goal. The PERFECT's Testing And Verification (TAV) team, led by the Pacific Northwest National Laboratory (PNNL), is charged with assessing the contributions of each team.

The TAV strategy quantitatively assesses PERFECT technologies individually and with respect to the programs overall performance and power goals. Our methodology uses a combination of benchmarking (code kernels and full applications), modeling, and simulation. The TAV approach is built upon three "pillars" a set of Baseline Architectures that reflects today's state-of-the-art embedded systems, a Benchmark Suite, which contains exemplar applications and kernels of interest to PERFECT, and a Proxy Architecture that represents a flexible and extensible generic target architecture that will be refined over time.

This webpage is the repository for the Benchmark suite effort. The suite is comprised of kernels and applications representing domains of interest to PERFECT. The suite was formed through a survey of applications used by all teams as well as consultations with numerous application domain experts. Each kernel in its own right is computationally important and provides reasonable coverage of a domain of interest. Furthermore, each kernel is algorithmically interesting.

Application Domains Kernels
PERFECT Application 1
Discrete Wavelet Transform
2D Convolution
Histogram Equalization
Space Time Adaptive Processing
System Solver
Inner Product
Outer Product
Synthetic Aperture Radar
Interpolation 1
Interpolation 2
Back Projection (Non-Fourier SAR)
Wide Area Motion Imaging
Image Registration
Change Detection
Required Kernels

The suite contains kernels from the following application domains: Wide Area Motion Imaging (WAMI), image processing, Synthetic Aperture Radar (SAR), Space Time Adaptive Processing (STAP), and general signal processing. Each of the suite.s areas is composed of an algorithmic specification that describes the domain and the algorithms being used and a reference implementation which target their serial implementation. To download the current version of the benchmark suite, please follow the link provided below. For any comments, questions or suggestions, send an email to Joseph Manzano or Nathan Tallent.

Current Version

You can access it through github:

For citation purposes, please use the plain text or bib file, linked to this page

Site Links

Our Research

Past Research