TR-2021-05

Going Beyond Efficiency to Eliminate computingís Absolute Negative Environmental Impact

Andrew A Chien. 11 March, 2021.
Communicated by Andrew Chien.

Abstract

For over 75 years, computing energy efficiency, cost, and power has increased steadily, enabling extraordinary applications and the digital transformation of society. However, due to Jevonís paradox1 that characterized the idea of elastic demand, improved efficiency and cost has driven an even more rapid increase in computing use. The net effect is that collective negative environmental impact has grown rapidly and in 2021 datacenters will account for 0.33 gigatons of CO2 emissions , high-speed wired and wireless networks 0.5 gigatons of CO2 emissions, and computing devices for >50 million metric tons of e-waste each year . These totals continue to grow at a rapid pace. Further, computingís latest advance, AI/Machine Learning, is accelerating the growth of these negative environmental impacts. MLís proliferating applications combined with exponential growth in model size (and training cost) has reached extraordinary levels, and we are only in the early days of the ML revolution . We must acknowledge the negative impacts of environmental impacts of computing, and strive to eliminate them in order to empower its enormous potential for positive impact.

Original Document

The original document is available in PDF (uploaded 11 March, 2021 by Andrew Chien).