Calculator Inputs
Enter hardware, runtime, workload, and facility assumptions below. The page stays in a single stacked layout, while the form uses 3 columns on large screens, 2 on medium, and 1 on mobile.
Example Data Table
The table below shows a sample AI training scenario and the resulting forecast. You can replace these values with your own operating assumptions.
| Sample Input | Value | Sample Output | Value |
|---|---|---|---|
| Workload Type | Training | Facility Energy per Run | 51.79 kWh |
| Accelerators | 8 | Monthly Energy | 932.19 kWh |
| Power per Accelerator | 350 W | Annual Energy | 11,186.28 kWh |
| CPU and Base Server Power | 180 W | Monthly Cost | $121.18 |
| Memory Power | 80 W | Annual Cost | $1,454.22 |
| Storage and Network Power | 70 W | Monthly Emissions | 313.22 kg CO2e |
| Idle Power Ratio | 35% | Energy per 1M Units | 2.0715 kWh |
| Average Utilization | 82% | Infrastructure Overhead Share | 24.24% |
| Runtime per Run | 14 hours | Units per kWh | 482,734 |
| Runs per Month | 18 | Emissions per 1M Units | 0.6960 kg CO2e |
Formula Used
This estimator combines IT power, utilization behavior, and facility overhead to model realistic AI energy consumption.
1) Effective accelerator power
Effective Accelerator Watts = Accelerator Count × Power per Accelerator × (Idle Ratio + (1 − Idle Ratio) × Utilization)
2) Shared system power
Shared System Watts = CPU Power + Memory Power + Storage/Network Power
3) Total IT power
Total IT Power = Effective Accelerator Watts + Shared System Watts
4) IT energy for one run
IT Energy per Run (kWh) = Total IT Power × Runtime Hours ÷ 1000
5) Facility energy for one run
Facility Energy per Run = IT Energy per Run × PUE
6) Monthly and annual energy
Monthly Energy = Facility Energy per Run × Runs per Month
Annual Energy = Monthly Energy × 12
7) Cost estimation
Monthly Cost = Monthly Energy × Electricity Rate
Annual Cost = Monthly Cost × 12
8) Emissions estimation
Adjusted Carbon Intensity = Grid Carbon Intensity × (1 − Renewable Share)
Monthly Emissions = Monthly Energy × Adjusted Carbon Intensity
9) Workload efficiency
Energy per 1M Units = Facility Energy per Run ÷ (Units per Run ÷ 1,000,000)
Units per kWh = Monthly Units ÷ Monthly Energy
How to Use This Calculator
Follow these steps to estimate power, cost, and emissions for AI model training, inference, or mixed workloads.
- Choose the workload type that best matches your AI activity.
- Enter the number of accelerators and their average operating wattage.
- Add server, memory, and storage or networking power values.
- Set the idle ratio and average utilization to reflect real usage.
- Enter runtime per job and the expected number of runs each month.
- Provide the number of processed units per run, such as tokens or requests.
- Enter the datacenter PUE, local electricity rate, carbon intensity, and renewable share.
- Click Estimate Energy Consumption to show the results above the form.
- Review energy, cost, emissions, and efficiency metrics.
- Use the CSV and PDF buttons to export the result summary.
FAQs
These answers stay brief and use plain HTML only.
1) What does this calculator estimate?
It estimates AI workload energy use, electricity cost, emissions, and output efficiency. It supports training, inference, and hybrid planning scenarios.
2) Why is utilization important?
Accelerators rarely draw full power continuously. Utilization lets the estimate reflect average real work instead of ideal peak conditions.
3) What is idle power ratio?
It represents the share of device power still consumed when the accelerator is not fully busy. Many systems keep drawing meaningful power while waiting.
4) What is PUE?
PUE measures facility overhead. A value above 1.0 means extra energy is used for cooling, power delivery, and other datacenter support systems.
5) Can I use tokens or requests as units?
Yes. The processed units field is generic. You can enter tokens, prompts, images, batches, requests, or another consistent workload measure.
6) Are emissions adjusted for renewable energy?
Yes. The calculator reduces the effective carbon intensity by the renewable share you enter, giving a lower emissions estimate when cleaner energy is used.
7) Is this suitable for exact billing?
No. It is a planning estimator. Actual billing can differ because of cloud pricing rules, reserved capacity, storage fees, or burst behavior.
8) How can I improve accuracy?
Use measured average wattage, realistic runtime, observed utilization, and your actual electricity and carbon factors. Better inputs produce better estimates.