Environmental Impact (π΄TODO)ΒΆ
It is important to be able to asses, at least roughly, the environmental impact of an application, if we want to be able to compare with other (potentially non-software based) solution to the problem they address.
GPU-intensive application have an environmental impact through multiple expenses, for instance:
The impact of the energy expense while using the application depends on how long computation run, how long the user uses intensive computations in a typical workflow, how intense the computation is.
The choice of minimal required limits impacts the need to manufacture new hardware.
Tools for measuring instant energy expense depend a lot on the platform. For instance NVidia provides information through the NVML library.
The carbon impact of this energy depends on the userβs location. Several APIs exist to query this information, like WattTime.
The manufacturing impact is harder to evaluate, especially for non-carbon related, which includes notably the challenging extraction of rare metals, which are non-renewable.