Tuesday, January 2, 2024

Global AI Electricity Consumption is an Issue, But We'll Get Better at it


Over time, lots of things can be done to lessen AI energy consumption:
*  More efficient AI hardware
*  Cooling advancements including liquid immersion or indirect evaporative cooling
*  Model optimization: quantization, pruning, and knowledge distillation to reduce the complexity of AI models without sacrificing accuracy
*  Edge computing in several forms will help as well, including on-device computations
* Scheduling of AI operations to even out workloads should help as well

Renewable energy will help a bit, at least in terms of carbon footprint. 

No comments:

"Live by the Sword, Die by the Sword"

It is a fair question to wonder how global high-performance computing capacity might change if interests that want slower or little AI data ...