Low Power Mesh Algorithms for Image Problems

TL;DR

This paper introduces power-efficient mesh algorithms for image processing that activate fewer processors at a time, significantly reducing peak power consumption while maintaining performance.

Contribution

It presents novel algorithms for mesh-connected computers that minimize power usage by activating only a subset of processors, applicable to image problems.

Findings

Significant reduction in peak power consumption.

Minimal impact on processing time.

More efficient power utilization with increased power availability.

Abstract

We analyze a physically motivated fine-grained mesh-connected computer model, assuming that a word of information takes a fixed area and that it takes unit time and unit energy to move a word unit distance. This is a representation of computing on a chip with myriad tiny processors arranged as a mesh. While most mesh algorithms assume all processors are active at all times, we give algorithms that have only a few processors on at any one time, which reduces the power required. We apply this approach to basic problems involving images, showing that there can be dramatic reductions in the peak power with only small, if any, changes in the time required. We also show that these algorithms give a more efficient way to utilize power when more power is available.