Skip to main content

A controlled reduction in data accuracy saves energy

Published on 2.11.2020
Tampere University
Jari Nurmi Tampereen yliopiston Hervannan kampuksella.
Computers, the Internet of Things (IoT) and high-speed mobile broadband connections use up so much power that the energy required for computing is estimated to exceed global energy production by 2040. An international research project coordinated by Tampere University looks into the possibility of saving energy by reducing the accuracy of data.

The project titled Approximate Computing for Power and Energy Optimisation (APROPOS) is exploring new solutions for reducing the energy consumption of computers. The study focuses on approximate computing (AC), a technique for striking an optimal balance between data accuracy and energy usage.

Achieving a high level of accuracy takes longer and requires more computing power and therefore consumes more electricity. According to Jari Nurmi, professor of computing systems at Tampere University, a number of systems that are used for collecting, transferring, processing or storing data are error tolerant.

“Both time and energy are saved when we compromise on accuracy. With data optimisation, we can save up to 10-50 times more energy as opposed to fully accurate computing,” says Nurmi, who leads the international APROPOS project.

The sufficient level of accuracy is context-dependent

When the final output is interpreted by the human senses, slight inaccuracies or random errors are usually acceptable. For example, compressing a digital photograph reduces the number of pixels but the quality of the image may still be good enough. The required degree of accuracy depends on context.

“The rule of thumb is that any data relating to health and safety must always be as precise as possible. However, with machine learning it is possible to relax accuracy and still deliver the desired results. In addition, data mining systems and other data-driven applications may not necessarily require precise results to be useful,” Nurmi says.

Nurmi mentions autonomous driving as one interesting application area for AC. The APROPOS team will examine the types of data and the degree of accuracy required in the context of self-driving cars.

New talents sought for developing energy saving solutions

The interest in AC has been growing in recent years. Professionals with the ability to deliver architecture-, software- and system-level solutions are in great demand. Through research and learning, the project aims to identify new methods and tools for saving energy.

“The training of new professionals is an important part of the APROPOS project. The field of AC has definite global potential,” Nurmi says.

During the international APROPOS project, 15 doctoral researchers will be trained to apply AC techniques to develop novel energy saving solutions for embedded and high-performance computing systems. The call for applications will be launched at the turn of the year.

The four-year APROPOS project involves 14 organisations from all over Europe. The project has received funding under the EU’s prestigious Marie Skłodowska-Curie Innovative Training Networks programme and has an overall budget of €4 million.

Further information:

Jari Nurmi
tel. +358 40 506 4460
jari.nurmi [at] (jari[dot]nurmi[at]tuni[dot]fi)

Text: Anna Aatinen
Photo: Sari Laapotti


What is an approximation?

An approximation is an estimate that provides results that may contain random errors or small systematic errors. Approximation is a useful technique when there is not enough or too much data available. The effects of approximations on the end result can be tested, for example, with the help of simulations.