AxBench: A Multiplatform Benchmark Suite for Approximate Computing 论文

2016IEEE Design and Test引用 232
Parallel Computing and Optimization TechniquesLow-power high-performance VLSI designEmbedded Systems Design Techniques

摘要

Approximate computing is claimed to be a powerful knob for alleviating the peak power and energy-efficiency issues. However, providing a consistent benchmark suit with diverse applications amenable to approximate computing is crucial to ensure fair and reproducible comparisons. This article makes an important attempt toward it in the form of the AxBench suite, which contains applications for CPUs, GPUs, and hardware design with necessary annotations to mark the approximable regions and output quality metrics. -Muhammad Shafique, Vienna University of Technology.