Context

This page gives some background on our work and why we think every datascientist should start measuring their power usage.

We first provide some element of comparison between CO2 emissions in Some ideas of scale.

We then give some background information on the power used by algorithms and its carbon footprint in AI’s carbon footprint.

The page Selecting hardware references some key elements to select the hardware used (such as the processor or the PUE or even the electricity mix)

We finally give the reader some hints of algorithm to use to optimize its accuracy/efficiency trade-off in The accuracy/efficiency trade-off

• Some ideas of scale
  • Energy consumption
  • CO2 emissions
• AI’s carbon footprint
  • The ICT footprint
  • Large Neural Network Training footprint
• Selecting hardware
  • Server location
  • Power Usage Effectiveness
• The accuracy/efficiency trade-off
  • Red AI vs Green AI
  • Accurracy - efficiency trade-off

Data information Some ideas of scale