Reducing the carbon footprint of global computing
Compute demand and related energy consumption and climate footprint are increasing at a fast rate, especially due to the widespread use of machine learning, AI, gaming, and blockchain. To tackle this issue, researchers from academia and industry leaders across multiple ICT sectors are exploring and discussing opportunities to reduce data-driven energy consumption. This was the purpose of the workshop “Climate Implications of Computing and Communications” hosted by MIT’s Climate and Sustainability Consortium (MCSC), MIT-IBM Watson AI Lab, and the Schwarzman College of Computing. During this event, experts highlighted options for higher energy efficiency which involve both better algorithms and hardware modifications. An example is the innovative use of vertical stacking in chip design to reduce the distance data has to travel, thus allowing lower energy consumption. Other examples include segmenting standby and full processing to enhance energy saving in computers, as well as sparse matrixes to improve efficiency in neural networks.
As the workshop presentations stressed, to reach the optimal results, both the algorithms and the hardware have to be as efficient as possible and sometimes need to be co-designed through a holistic and multidisciplinary approach.
On the other hand, other experts focused on end-user devices, highlighting prolonging the life of the devices as the best action to reduce one’s digital carbon footprint, or considering options to enable customers to choose among technologies with different climate impacts.