Green AI Carbon Accounting Audit
Green AI Carbon Accounting Audit
Training Information
- GPU: NVIDIA V100
- Number of GPUs: 3
- GPU runtime: 37 hours
- Training type: pre-training
- Region: europe-west4
- Power Usage Effectiveness (PUE): 1.11
Environmental Impact
- Total GPU energy consumption: 36.963 kWh
- Total carbon emissions: 7.393 kg CO2eq
Calculation
Total Energy Consumption
Using the NVIDIA V100 TDP of 300 W:
energy_kWh = (300 W × 3 GPUs × 37 hours × 1.11 PUE) / 1000
Total energy consumption = 36.963 kWh
Carbon Emissions
Using the europe-west4 grid carbon intensity of 200 gCO2eq/kWh:
co2_kg = (36.963 kWh × 200 gCO2eq/kWh) / 1000
Total carbon emissions = 7.393 kg CO2eq
Carbon Accounting Metadata
The co2_eq_emissions metadata records the estimated carbon emissions associated with this training run.
| Metric | Value |
|---|
| Hardware | NVIDIA V100 |
| Number of GPUs | 3 |
| GPU runtime | 37 hours |
| PUE | 1.11 |
| Location | europe-west4 |
| Training type | pre-training |
| Energy consumption | 36.963 kWh |
| Carbon emissions | 7.393 kg CO2eq |
| Grid carbon intensity | 200 gCO2eq/kWh |
Source Training Run
The environmental accounting is based on the assigned training run configuration and the provided GPU TDP and regional grid-carbon reference values.
The reported emissions value is rounded to three decimal places as required.
