Computation of carbon emissions

[LucieLgnr]

Hello,

I'm an analyst in life cycle assessment, and I list all the tools available to calculate the environmental impact of artificial intelligence/cloud computing.

I have some questions about CodeCarbon (I read the documentation but my skills in programing are rather limited) :

• Where regional carbon intensity of electricity come from ? Is it from Electricity map ?
• Is CodeCarbon able to calculate the carbon intensity for each province of Canada (Québec, Ontario, etc) and the United States ?
• If the computing occurs in a private cloud, is it possible to adapt CodeCarbon with the specifications of the hardware ?
• Is the tool able to identify if the IA training is distributed over several servers ?

Thanks,

Lucie

[benoit-cty]

Hello, thanks for your interest and sorry for the delay.

Where regional carbon intensity of electricity come from ? Is it from Electricity map ?
By default it's a static file we build from Our World In Data : https://github.com/mlco2/codecarbon/blob/master/codecarbon/data/private_infra/global_energy_mix.json
But CodeCarbon car use your Electricitymaps API if you provide it.

Is CodeCarbon able to calculate the carbon intensity for each province of Canada (Québec, Ontario, etc) and the United States ?
Only if you use the Electricitymaps API.

If the computing occurs in a private cloud, is it possible to adapt CodeCarbon with the specifications of the hardware ?

CodeCarbon auto-detect the hardware. You also have the ability to specify the PUE of the datacenter il you know it.

Is the tool able to identify if the IA training is distributed over several servers ?

Yes, by running CodeCarbon on all node and using the API to aggregates all the data. We are currently rebuilding this part with authentication. The previous version is on http://dashboard.codecarbon.io/