Worked Example

The complete methodology for calculating operational emissions is: 

((((Compute Infrastructure + Network Storage + Intra-Region Data Transfer + Intra-Region Non-compute Networking) * PUE) + (Inter-Region Data Transfer + External Data Transfer + Inter-Region Non-compute Networking + External Non-compute Networking)) * Carbon Intensity * Power Transmission Losses)   

Where:  

Compute Infrastructure = (Processors + Accelerators + Memory + Onboard Storage + Motherboard) * 1.04  

Network Storage (Wh) = ( (0.0029 W * GB of Network Storage) * Hours of Utilization) * 1.04   

To summarize the operational methodology, a worked example for an AWS c6gd.medium instance, operating in a UK data center, is outlined here.   

In this example, the instance has been in use for 18 months. This means the instance has been utilized for 13,140 hours.  

Tailpipe calculates operational emissions hourly, based on hourly CPU utilization data. For this example, the CPU utilization will be 25% for every hour.  

First, Tailpipe finds the instance’s CPU name: Graviton2. This has a TDP of 150. Tailpipe then refers to its CPU utilization power draw dataset. This shows that a CPU with a TDP of 150 at 25% utilization will draw 87 W per hour.  

Second, Tailpipe finds the RAM quantity and type for this instance: 2 GB of DDR4. Tailpipe then refers to its Average Memory Power Draw dataset. This shows that at 25% utilization, the RAM component will draw 0.0598 W/GB.  

Tailpipe then assesses the impact of the other four factors (GPU, storage – SSD, HDD, or network storage – motherboard, and network interface cards), which do not rely on CPU utilization.  

Each of these hourly values is then multiplied by the number of hours of utilization; in this case, 13,140. 

The total is then multiplied by the Power Supply Efficiency Factor of 1.04:  

120416.274 * 1.04 = 125232.925 Wh

The total is then converted to kilowatt hours: 125.233 kWh     

Next, Tailpipe adds the impact of networking – data transfers across the AWS network and internet. For this example, 100 TB of data is transferred across each type of networking: intra-region, inter-region, external, and non-instance. 

The figures for intra-region networking (0.06 Wh) and intra-region non-compute networking (0.06 Wh) are then added together and converted to kilowatt hours: 0.00012 kWh 

This figure is added to the figures for the instance (293.558 kWh) and network storage (0 kWh) and multiplied by the Power Usage Effectiveness factor (1.22) of the data center:  

(0.00012 + 125.233 + 0) * 1.22 = 152.784 kWh  

Next, the networking figures that are external to the data center (and therefore not affected by its Power Usage Effectiveness) are added together and converted to kWh:  

0.00006 + 0.00058 + 0.00006 + 0.00058 = 0.00128 kWh  

These two figures are added together: 152.785 kWh  

This total figure is then multiplied by the carbon intensity of the local grid mix. The UK’s grid mix is usually in the moderate range, between 100 and 200 khCO₂e/kWh. For this example, a grid mix of 150 gCO₂e/kWh is used to represent 18 months of UK grid mix emissions.  

152.785 * 150 = 22917.75 gCO₂e  

This figure is then multiplied by the Power Transmissions Losses factor of 1.08, to account for the 8% of electricity wasted at source in the UK.  

22917.75 * 1.08 = 24751.17 gCO₂e  

Finally, this figure is converted into kilograms: 24.75 kgCO₂e  

This means that a customer’s use of the c6gd.medium instance for 18 months in a data center in the UK produced 24.75 kgCO₂e.   

That’s the equivalent of driving 90.5 miles (145.6 km) in a petrol car.