Energy Intensity of Internet Traffic

How much energy is required to transfer 1GB of data over the internet from London to Paris?

Introduction

This post is another comparison of LLMs. I asked how much energy is required to transfer 1GB of data over the internet from London to Paris. As background, I am moving my personal archive from Evernote to the open-source Joplin. The Joplin servers are in Paris. I live in London and have about 7GB of data to sync between four machines. I wondered about the carbon footprint of all those bits flying around!

The results of this comparison are consistent with my LLM-efficacy working hypothesis that LLMs perform very well on fact-based questions when the training data is clear and unmuddled. Hence, they are good at answering programming questions because they read the docs, and the docs are consistent. Because they read all the docs, they are probably better than humans. But, like humans, they are less good when the training data is muddled, e.g., “Who won the 2020 US Presidential election?”

For this question, the internet and literature (=training data) contains a wide range of answers. The lack of consensus is reflected in the wide range of answers.

Related posts

1. Python programming problem involving classes
2. Managing Sublime Text Projects
3. Joplin-Evernote migration policy
4. Defining Bullshit
5. Why Buy Insurance?
6. Science and Pseudo-science

---

Question

How much energy does it take to move 1GB of data from London to Paris over the internet?

Commentary

A human curated internet search uncovers Aslan et al. (2018), a paper that has been cited by 220 other works Google Scholar. It is a meta-study, including estimates from 14 other works. The abstract says

This study identifies representative estimates for the average electricity intensity of fixed-line Internet transmission networks over time… Differences in system boundary, assumptions used, and year to which the data apply significantly affect such estimates. Surprisingly, methodology used is not a major source of error, as has been suggested in the past. This article derives criteria to identify accurate estimates over time and provides a new estimate of 0.06 kWh/GB for 2015.

The paper notes considerable variation over time:

Existing estimates for the electricity intensity of Internet data transmission, for 2000 to 2015, vary up to 5 orders of magnitude, ranging from between 136 kilowatt-hours (kWh)/GB in 2000 (Koomey et al. 2004) and 0.004 kWh/GB in 2008 (Baliga et al. 2009). While increased efficiency over time can account for 2 orders of magnitude of this variation.

The paper presents a linear regression extrapolation energy cost from 2015 forward, indicating a rate below \(<0.01\) kWh/GP today.

Response Summary

Here are the estimates offered by each model. Claude volunteered a range, GPT and Gemini provided one when prompted. Gemini answered in Joules, which it converted into kWh when prompted. Only GPT’s answer is in the range implied by Aslan et al. (2018). Gemini’s answer seems to ignore all the tech infrastructure and give an answer just accounting for the photons in the fibre optic cable.

Model	Point Estimate	Range (per GB)
GPT4.o	0.01 kWh / GB	0.005 kWh to 0.02 kWh (95%)
Claude 3	1-2 kWh / GB	0.1 to 10 kWh
Gemini Adv	Essentially zero	2.78e-20 to 2.78e-18 kWh (95%)
Aslan	0.06 kWh / GB	Trends to < 0.01 kWh

Verbatim Responses

GPT 4.o

Estimating the energy required to transfer 1GB of data from London to Paris over the internet involves several factors. These include the energy consumption of network equipment, data centers, and transmission lines. Here’s a simplified approach to provide a rough estimate.

1. Energy per Bit Transferred:
   • Studies suggest that the energy consumption for transferring data over the internet can vary significantly, but a common average estimate is around \(0.01 \text{ kWh/GB}\).
2. Calculation:
   • Given that 1GB of data transfer typically requires \(0.01 \text{ kWh}\), we can use this value directly.

Therefore, to transfer 1GB of data from London to Paris the Energy is approximately

\[ 0.01 \text{ kWh/GB} \times 1 \text{ GB} = 0.01 \text{ kWh} \]

This estimate is highly generalized and actual energy consumption can vary based on the specific network infrastructure, efficiency of data centers, and other factors involved in the transmission process.

Followup question: Can you give a confidence interval range?

To provide a confidence interval for the energy consumption estimate, we need to consider the variability in the estimates from various studies. Typically, energy consumption per GB of data transfer over the internet ranges from (0.005 ) to (0.02 ).

Calculation of Confidence Interval

Using the lower and upper bounds of the energy consumption estimates:

1. Lower Bound: \[ 0.005 \text{ kWh/GB} \times 1 \text{ GB} = 0.005 \text{ kWh} \]

2. Upper Bound: \[ 0.02 \text{ kWh/GB} \times 1 \text{ GB} = 0.02 \text{ kWh} \]

Confidence Interval

So, the 95% confidence interval for the energy required to transfer 1GB of data from London to Paris is approximately:

\[ 0.005 \text{ kWh} \text{ to } 0.02 \text{ kWh} \]

This range accounts for different network efficiencies, types of infrastructure, and other variables that can affect energy consumption.

Claude 3

It’s a great question, but a complex one to answer precisely as there are many factors involved. To give a rough estimate:

Sending 1GB of data over the internet from London to Paris would likely consume on the order of 0.1 to 10 kWh of electricity, with a best guess of around 1-2 kWh.

The main energy costs come from:

1. The energy used by the sender and receiver’s devices (computers, phones, etc.)
2. The energy used by network equipment like routers, switches, and optical repeaters to transmit the data between London and Paris.
3. A portion of the overhead energy required to run the data centers and network infrastructure.

Some key factors affecting the energy consumption:

• The physical distance between London and Paris (~350 km line-of-sight, but longer via fiber optic cable routes)
• The number of hops/nodes the data passes through and the efficiency of the routing
• The energy efficiency of the end-user devices and network equipment
• The energy sources and overhead of the data centers and networks

So in summary, while it’s hard to give an exact number, a rough estimate based on studies of energy intensity of internet data transfer is that sending 1GB from London to Paris would take in the neighborhood of 1-2 kWh, within an uncertainty range of 0.1 to 10 kWh. For comparison, 1 kWh is about the energy used by a typical microwave oven operating for 1 hour. I hope this helps provide a ballpark sense of the energy footprint! Let me know if you have any other questions.

Gemini Advanced

The energy required to move 1 GB of data from London to Paris over the internet is extremely small. Here’s why:

• Data Transfer: Data isn’t physically “moved” like a package. It’s transmitted as light signals through fiber optic cables.
• Energy Consumption: The primary energy consumption comes from the equipment involved in the transmission (routers, switches, etc.), not the data itself.
• Calculations: While precise calculations are complex, a simplified estimate suggests it takes on the order of \(10^{-12}\) Joules¹ (0.000000000001 Joules) to transmit 1 GB of data over the distance between London and Paris.

To put this in perspective, this amount of energy is roughly equivalent to lifting a grain of sand a few millimeters off the ground.

Important Note: This is a simplified estimation. Actual energy consumption can vary depending on numerous factors, including the specific route taken, the efficiency of the equipment, and network traffic.

Followup question: Can you give a confidence interval range?

Providing a precise confidence interval for the energy required to transfer 1 GB of data between London and Paris is challenging due to the numerous variables involved in data transmission. However, based on existing research and considering the efficiency of modern fiber optic networks, we can estimate a possible range:

Lower Bound: 10^-13 Joules (0.0000000000001 Joules) - This assumes highly optimized transmission with minimal energy losses.

Upper Bound: 10^-11 Joules (0.00000000001 Joules) - This accounts for potential inefficiencies, network congestion, and less direct routes.

Confidence Level: 95%

This means we can be 95% confident that the actual energy consumption falls within this range. However, it’s crucial to remember that this is an estimation, and the real value could lie outside these bounds.

Pls convert to kWh

The provided values are in Joules. To convert them to kilowatt-hours (kWh), we’ll divide each value by the number of Joules in a kilowatt-hour (3.6e+6).

The estimated range for energy consumption in kilowatt-hours (kWh) for transferring 1 GB of data between London and Paris is:

• Lower Bound: 2.78e-20 kWh
• Upper Bound: 2.78e-18 kWh

These are just estimates, and the actual energy consumption could be outside this range.

References

Aslan, J., Mayers, K., Koomey, J. G., & France, C. (2018). Electricity intensity of internet data transmission: untangling the estimates. Journal of Industrial Ecology, 22(4), 785–798. https://doi.org/10.1111/jiec.12630

Footnotes

1. Stated as equivalent to 2.7778e-25 kilowatt-hours (kWh) in a follow up question.