Energy Consumption Calculator (AI)
Estimate total energy consumption of AI training/inference from GPU power draw, runtime hours, and data center PUE.
Inputs
Power Usage Effectiveness — ratio of total facility power to IT equipment power. 1.0 = no overhead.
Total Energy Consumption
2,764.80kWh
IT Equipment Energy
2,304.00kWh
Facility Overhead Energy
460.80kWh
Step by step
IT equipment energy
(400W × 8 × 720h) ÷ 1000
= 2304.00 kWh
Total facility energy (with PUE)
2304.00 × 1.2
= 2764.80 kWh
How it works
Total data center energy consumption exceeds the raw power draw of GPUs alone because of cooling, power distribution losses, and other facility overhead — captured by the PUE (Power Usage Effectiveness) metric. This calculator computes GPU energy use directly from TDP × count × hours, then applies the facility's PUE multiplier to estimate total energy drawn from the grid, which feeds directly into carbon emissions estimation.
Formula
total_kwh = (gpu_tdp_watts × gpu_count × hours / 1000) × PUE
- gpu_tdp_watts
- GPU thermal design power in watts
- gpu_count
- Number of GPUs
- hours
- Runtime in hours
- PUE
- Power Usage Effectiveness ratio (1.0 = no overhead)
Frequently Asked Questions
What is a good PUE value?
Modern hyperscale data centers achieve PUE values around 1.1-1.2, while older or less efficient facilities can run 1.5-2.0 or higher.
Does TDP represent actual power draw?
TDP is a manufacturer rating that approximates peak sustained draw; real-world average utilization during training is often 70-95% of TDP, so this is a reasonable upper-bound estimate.
How do I convert this to a carbon footprint?
Feed the totalKwh result into the Carbon Emission Calculator (AI) along with your grid's carbon intensity to get a CO2 estimate.