This post is adapted from a presentation I did a few weeks ago.
Computing has often been seen in the past as more environmentally friendly than using lots of paper, but that is not necessarily the case.
There are lots of effects that computing has on the environment (as we shall see), but an important one is how it impacts the greenhouse effect and thus causes climate change. Now, there are lots of greenhouse gases - CO2 is only one (and not even the most potent) - but we can estimate the approximate amount of CO2 required to provide a similar environmental effect. This will be our method of quantitising the impact on climate change and is generally known as "Carbon Dioxide Equivalent" (CO2e). We could measure the electrical power used for our quantitisation, but the efficiency of power generation will be far less than 100% - indeed, gas power stations, in which the majority of UK power is generated, have an efficiency of 49.9%, for instance; so for each Joule of energy used another Joule at least is wasted. This means that CO2e is a better measure of the carbon intensity of electrical devices, as we will already have factored in the loss of efficiency. However, it's not perfect, as the carbon intensity of electricity generation varies around the world. When local information is unavailable, we will use a global estimate of 442g per kWhr.
UK energy mix for 2021, showing energy generation is dominated by gas-fired power stations
Efficiency of gas-fired power stations.
When I talk about "devices" here, I include all ICT devices such as computers, mobile-phones, monitors, printers, routers, servers, keyboards and other peripherals, hard-drives and other components, smart devices, gadgets, etc.. Producing and decommissioning devices creates a lot of CO2e emissions: the devices are said to have “embodied carbon”. While internal-combustion-engined-cars and simple electronic units produce more CO2e in usage than in manufacture, the reverse is true with modern electronics. For instance, the embodied carbon of a laptop can be 214kg of CO2e, roughly 56% of its environmental impact if kept for the expected average lifetime of 4.5 years. Meanwhile, producing a smartphone with an expected lifetime of just 3 years creates even more: for example, the iPhone 6 costs about 81kg CO2e to manufacture, about 3kg CO2e to transport to the customer, about 10kg CO2e in electrical use over its life and about 1kg CO2e to safely recycle at the end of its life, which for a phone that only weighs 129g means its total emissions are roughly 731 times its own mass in CO2e! Generally, manufacture and disposal account for 72% of the environmental impact of a smartphone because they aren’t used for a long time.
PDF on laptop environmental impact.
Study on the environmental impact of smartphones.
Specifications of an iPhone 6, showing its mass as 129g.
The energy used to produce consumer devices in 2020 was 381TWhrs, which is enough to boil kettles for 400 million million cups of tea. However, devices use a lot of energy too: according to the Semiconductor Industry Association, by 2040, "the energy required for computing will exceed the estimated world's energy production" (just the energy used by the devices)!
Table 6 of this report shows electricity use of the production of consumer ICT hardware at 381TWhrs.
Electricity usage of producing cups of tea.
Semiconductor Industry Association report.
Components have a lot of materials environmentally damaging to produce, including rare-earth elements...
(References mostly taken from "Environmental impacts of rare earth production" report.)
The ores for rare-earth elements often contain Thorium Dioxide (containing Thorium-232) and Uranium Oxide (containing Uranium-238), both highly radioactive substances, the dust of which is kicked up into the air by the mining process. The process of mining rare-earth elements produces waste with high concentrations of ammonium sulphate and heavy metals, which are discharged into the soil and air. Once mined, rare-earth element ores must be leached with acids and organic solvents, which are present in the waste from the process. The leaching process can involve large amounts of ammonium sulphate solution, which can then leak into the soil and nearby watercourses, causing eutrophication. Production of ammonium sulphate in the first place also causes eutrophication as well as ecotoxicity of freshwater and organisms, particulate matter formation and CO2e emissions. The large amount of ammonium sulphate used for this process and its potential to stay in the soil and groundwater has resulted in concentrations of 3.5-4g per litre found in the local groundwater. Other impacts are difficult to measure because of the large proportion of mining using this method being done illegally. Ore is then crushed and sorted, using floatation chemicals, which are then discharged into open ponds, allowing heavy metals and inorganic phosphorus-containing compounds to leach into the soil and may leak into rivers through rain-erosion. Waste from this process can be 5% radioactive Thorium, with also high concentrations of fluorides, boron, chlorides, sulphates, ammonium, manganese and iron.
The next stage for the ore is to be “Cracked” by roasting it with either Sulphuric acid or Hydrochloric acid, producing emissions of toxic hydrogen fluoride and Sulphur Dioxide. Further leaching and precipitation then takes place, involving lots of Sulphuric acid, Hydrochloric acid and Sodium Hydroxide, which are very polluting to produce. Much of the waste from this process often exceeds environmental limits for radioactive waste because of its concentration of Thorium-232. The rare-earth elements are then separated in a complex process with hundreds of substages, demanding large amounts of Hydrochloric acid and small amounts of complex organic chemicals, which can leak into the surrounding countryside. Further subsequent precipitation then takes place, using oxalic acid and ammonium bicarbonate, both of which are environmentally costly to produce. The rare-earth elements are then calcinated to rare-earth oxides through a process with only marginal environmental impact, though the general use of fossil-fuels in this process is still damaging.
The final stage of production is metal refining, either through molten salt electrolysis or metallothermic reactions. The greatest environmental impact of these is the waste tungsten or molybdenum cathodes and the high electricity consumption (at 8-12kWh per 1kg of rare-earth metals/oxides produced – about 6.5kg of CO2e).
2021 Carbon intensity of power generated in China, where the rare-earth materials are mined.
Dysprosium is the more environmentally costly of the rare-earth elements to produce, especially impacting watercourses through eutrophication.
As we have seen, energy is used to create the devices that make up the Internet. Electricity is also used in the datacentres processing the data used for the internet - this is covered later in the Datacentres section. Electricity is used when generating websites dynamically on the fly (this can be reduced with using static pages where possible and caching etc.), such that predicted 2020 energy usage of webservers was 299TWhrs (132,158 million tonnes CO2e)!
See table 1 in section 2.1 for energy usage by web datacentres.
Energy is used to transmit data over the Internet, with longer distances using more energy in general. In some cases, data can be cached nationally to save being retransmitted, but since this requires more datacentres, it is debatable whether this saves electricity (though it will decrease delays). Estimated energy usage is difficult to calculate but in a 2017 meta-analysis for 2015 it was estimated to be 0.06kWhrs per gigabyte, with internet traffic in 2017 estimated at 9066 Exabytes and in 2015 at 5365 Exabytes, giving an energy figure of about 543.96 TWhrs (2017)/321.9 TWhrs (2015) or about 240,430.32 million tonnes of CO2e (2017)/142,279.8 million tonnes of CO2e (2015)! As can be expected, the more data sent over the internet uses more energy. The average size of websites has been increasing over the years as has general user traffic, so the energy usage has shot up in recent years!
Meta-analysis on the Electricity Intensity of Internet Data Transmission.
Datasheet with Internet Traffic data among other things.
Electricity is used by the program generating the output based on the data received over the network; in the case of the web, this is the web browser generating the webpage from the HTML, CSS and JavaScript code it received over the network. This can be reduced by cutting down on the use of JavaScript and reducing the complexity of the page in the HTML and CSS code so the browser doesn’t have to process as much. Electricity is used in displaying the output (such as a webpage) on a screen. Really, this falls under the device emissions so we won’t dwell on it, but some types of screen (old CRT monitors and the latest OLED screens) use less electricity for darker colours, though this makes no difference on the more common LCD and TFT screens. For E-paper/E-ink screens, electricity is only used in updating the screen, so on devices with those, electricity can be saved by reducing the number of elements that can change onscreen and reducing wasted space onscreen so fewer "pages" are needed.
Impact of Dark-Mode on battery drain,
Blog post on the limitations of the impact of darker colours on energy efficiency
It’s hard to put numbers to everything as estimates vary by a lot, but the overall 2020 total can be estimated at about 1,675TWhrs (though the EASL Datasheet figure has 1,988), which would be about 740 Teratonnes of CO2e, roughly 42000 car journeys or driving 316 million miles (greater than the distance to the moon)!
Average emissions of CO2 for UK cars (221.4g/mile).
There are somewhat simplified models to measure the carbon intensity of visiting a particular website, giving a rough idea of the CO2e generated by each visit. The website WebsiteCarbon.com features one such model for calculating various websites' carbon intensities. Because the estimates are fairly simplified, though, you shouldn’t rely on the figures for more than a basic guide. For instance, the site uses 442g/kWhr as an estimate of the carbon intensity of a website powered by fossil fuel energy and 50g/kWhr for those powered by renewable energy, however, the actual figures vary country to country. The straight binary choice of renewable vs non-renewable also doesn’t give an accurate picture as very few datacentres are powered by entirely renewable sources, with most using offsetting, with this not being mentioned on the websites. In addition, while the datacentre may have been told the power is renewable, many power companies simply buy a tradable certificate from renewable electricity generation facilities, which provides them with a source of income but does not involve buying any electricity itself.
Model used by WebsiteCarbon.com
Nature article on renewable energy certificates.
Why "100% renewable energy" pledges are not enough - Financial Times.
Despite the problems with the models, they can be a good indicator of website inefficiencies. Large pictures, embedded videos, imported JavaScript frameworks all consume large amounts of energy in transmission. However, one of the biggest culprits is streamed video, with most of the emissions coming from "OnDemand" services like Netflix and Amazon Prime Video. The following graph shows a breakdown of where the worst emissions come from, however, the headline figure of 306 Million tonnes of CO2e produced by internet video traffic in 2018 shouldn’t be glossed over and since this was before the pandemic, figures will have shot up since then!
/shift-project-video-streaming-carbon-emissions.png
Video-streaming electricity usage report.
Paper "On Global Electricity Usage of Communication Technology: Trends to 2030".
Datacentres are huge draws of electricity. From powering the servers to the air-conditioning needed to keep them cool, to the lighting for numerous rooms, to the embodied carbon of the thousands of computers. While datacentres have increased in number, they have also been better in recent years at electrical efficiency, so their energy demands have increased more slowly than would be expected.
/datacentre-energy-demands-table.png
There are other problems with datacentres, of course, as they require huge amounts of water for cooling despite often being built in places with limited water supplies.
News report on datacentre water-supply usage.
Also, in terms of conserving resources, they get through a lot of computer components as it's cheaper to replace something than to repair it, especially at scale. This is costly both in terms of embodied carbon and in rare-earth elements. Because of reliability and redundancy requirements, equipment is often replaced halfway through its expected lifespan.
Datacentre equipment decommissioning.
Global data centre electricity use in 2021 was 220-320 TWh, or around 0.9-1.3% of global final electricity demand. This excludes energy used for cryptocurrency mining, which was 100-140 TWh in 2021.
Datacentre energy usage report.
Datacentres also use lots of land and it’s cheaper to build on undeveloped land than so-called "brownfield sites", so areas of nature are bulldozed to build them.
This has been covered by many people before and thus I won't dwell on it much here, but to reiterate: Cryptocurrency mining used 100-140 TWhrs of electricity in 2021 and Bitcoin in 2020 used 68.7TWhrs of electricity. The estimated energy consumption of a single blockchain transaction is 750kWh (around 350kg CO2e). The annual estimation of the carbon emissions of all cryptocurrency activities has been put to at least 69 Million tonnes of CO2e, though this uses an underestimate of the carbon intensity of the required electricity production, so the true figure is likely to be far higher! Energy consumption of Blockchain technologies is estimated to be that of a medium sized country, though experts’ opinions differ on which country (Ireland, the Netherlands, etc.).
So, enough doom and gloom: what are some practical things we can do about these issues? Well, as shown before, most of the environmental damage from a device is in its manufacture and disposal, therefore this must be the focus. You’ll have heard “Reduce Reuse Recycle” before, but Repair should be there too. In decreasing order of importance:
As we have seen, using the internet involves using a huge amount of electricity to transfer data. Sadly, computer users have little direct control over network energy usage. However, we can request and send less data ourselves at least. Video data comprises a lot of data, so reducing its transmission makes sense. Therefore, turning off your camera on a Zoom call can greatly reduce data transmitted during the call. However, this must be balanced with the social difficulty of communication without seeing each other’s faces. Nonetheless, if there is no need to see faces (such as during a screenshare presentation), this could be a good idea. Note that video-calling still produces far fewer carbon emissions than driving a (even electric) car to meet someone, though.
Similarly, you can reduce the resolution of online video that you stream – it won’t look as good, but less data will be transmitted. You could also think about whether you need to watch an online video at all – if you watch a programme as it is being broadcast (on TV, rather than on catchup services), since this involves receiving something that would be sent anyway, you can save most of the energy that you would have used streaming it later!
If you build a website, use statically served pages to reduce energy usage at the host. Also, reduce the use of JavaScript, images and videos - do you really need a background image for the page, for instance?
Streaming an album over the internet 27 times can use more energy than the manufacturing and production of its CD equivalent, so it's better to buy music as MP3 downloads rather than stream it.
Cloud datacentre energy usage can be reduced indirectly through reducing the amount of data you store there. Think about whether you need to store all your photos with Google Photos, for example.
If you play a lot of videogames, try to play games that don't heavily tax your video-card as this will use a lot more electricity, especially if your computer has to spin up its fans. If you write software, target your software to the oldest devices you can, but with compatibility for newer devices, so that users don’t have to replace their devices to use your software.
Professor Wim Vanderbauwhede of Glasgow University's writings on Frugal computing,
Frugal computing from a consumer perspective,
and Frugal computing from a developer perspective.
The aforementioned "Cool Products Don't Cost The Earth" report.
The aforementioned"Environmental impacts of rare earth production" report.
PDF with the median emissions of energy-generation types (Table A.III.2)
PDF with (page 32, table A1.3) data of carbon intensity per country in 2020.
