Image
profs Subhasish Mitra and Jelena Vuckvic

Mitra and Vučković on reducing AI’s energy consumption

Summary

Energy costs are a worthy investment for companies as many have see their carbon emissions increase nearly 50%.

Oct
2024
Professors Subhasish Mitra and Jelena Vučković, along with other experts, are included in ‘Fixing AI’s energy crisis.’ The Nature article discusses the significant energy consumption associated with training and operating AI models, which raises environmental concerns and potential hindrances to progress in machine learning. 
 
The high energy demand largely stems from the architecture of current chips, where most energy is spent on data movement rather than processing. Researchers are exploring new chip designs to enhance energy efficiency, such as in-memory computing and analog computing techniques, which could significantly reduce energy costs.
 
Innovations like optically-based data transmission via silicon waveguides are also under exploration, delivering faster and more energy-efficient connections. Moreover, future developments could involve 3D stacked chips that bring memory closer to processing units, further decreasing energy consumption.
 
Read the full article, ‘Fixing AI’s energy crisis,’ nature.
Published : Oct 21st, 2024 at 12:41 pm
Updated : Oct 21st, 2024 at 12:57 pm