The Bitcoin network consumes approximately 40 to 60 TWh/year, or about 0.15% of global annual power generation, and 0.024% of global total energy output.
Original title: “Hardcore丨BTC’s true power consumption is only 0.15% of global power generation? 》
Written by: Tyler Bain
Translation: Sherrie
In recent years, many people have claimed that Bitcoin and miners who protect the network through SHA-256 proof of work consume too much energy. But what are the data on which these statements are based? Does the source calculation use flawed or reasonable methods and assumptions? How much electricity has the Bitcoin network used in history?
Methods and misunderstandings
Due to the large and globally distributed topology of the Bitcoin network, the amount of electricity and energy consumed by miners cannot be fully verified, but must be estimated . In the energy consumption of the past few years, a large number of reputable sources have participated in an attempt to estimate Bitcoin’s network energy consumption in a calmer and more data-based way:
- University of Cambridge, Judge Business School (JBS)
- The International Energy Agency (IEA)
- Electric Power Research Institute (EPRI)
- Coin Center
- CoinShares
- Marc Bevand
- Hass McCook
- Alex de Vries
- Myself
The estimation methods seem to fall into two main categories: economic methods based on financial assumptions and physical methods based on engineering principles . At the BTC2019 conference, we conducted a comprehensive comparison and comparison of these two estimation methods.
In understanding all these annual usage estimates, it is important to understand that electricity consumption is usually measured in two ways: instantaneous (power, watts, kilowatts, etc.) and the same instantaneous power measurement integrated over time (energy, joules, kilowatt hours ( kWh) etc.).
The problem of network energy estimation based on economy
Economics-based methods for estimating Bitcoin network energy consumption usually assume completely rational market behavior, which can be easily manipulated by incorrect assumptions of some input variables.
In theory, the Bitcoin mining industry is rational, profit-maximizing, and perfectly competitive: the marginal revenue of mining should tend to equal the marginal cost (MR = MC) . This means that in a long enough time, the market should find a balance point at which the energy cost per unit of Bitcoin production should be roughly equal to the market value of Bitcoin when it was minted. This calculation method can be refined as: “How much electricity can Bitcoin network miners bear?”
Usually, these types of estimates rely too much on a single volatility variable: the market transaction price of Bitcoin. Here is a simple example of this estimation:
Let’s try this estimate. Bitcoin generates one block approximately every 10 minutes-6 per hour, or 144 per day. Currently, a bitcoin block contains a coinbase block reward of 6.25 BTC; that is, 37.5 bitcoins per hour, or 900 new bitcoins to miners every day. As of the time of writing, the current market transaction price of Bitcoin is approximately US$10,750, which is equivalent to US$9,675,000 per day that Bitcoin miners can use to generate electricity.
This energy is equivalent to the annual electricity consumption of Bitcoin miners about 35.3 terawatt-hours . Assuming that the price of Bitcoin remains unchanged for one year, the average cost of electricity in the United States remains unchanged.
Although this method relies too much on Bitcoin prices, it also relies heavily on the miners’ hypothetical electricity costs. The calculations and conclusions of such estimates may be very different and may even be manipulated, depending on the assumptions used as inputs: energy cost ($/kWh) and Bitcoin price ($/BTC).
Here we use the average cost of electricity in the United States of US$0.10/kWh. However, in the United States, electricity costs are actually seasonal, from state to state, city to city, and in some cases, community to community. The same disharmony exists in global electricity costs. This does not even include a wide range of industrial, commercial or residential electricity tariffs, adding more sources of error to these economically based estimation techniques. In fact, this calculation’s heavy dependence on energy prices has another flaw: some highly creative miners have almost zero fuel costs when they mine excess energy that would otherwise be wasted, unavailable or cut. .
In my opinion, this quick exercise emphasizes why this economic-based estimation method is an oversimplification method, accompanied by the following questions:
Bitcoin mining, computing power, and network energy consumption are less responsive to sudden price changes than these economic-based estimation methods .
Economics-based models claim that after the Bitcoin block reward halving cycle (that is, every 210,000 blocks or about 4 years), energy usage and network miner rewards will be reduced by half, and the difficulty and proof-of-work-based The data proves this point.
This type of model assumes a single global average energy cost ($kWh); the cost of electricity varies greatly from region to region, season, and even to energy source.
This may be an upper limit estimate.
Benefits of physics-based network energy estimation
On the other hand, physics-based network energy estimation methods are often a very strict “digital operation” type that the Bitcoin community is used to.
These methods use independently verified on-chain difficulty, proof-of-work data and heat rate standards published by original equipment manufacturers (OEMs) to more accurately estimate the historical energy input into the Bitcoin mining system. The physical evaluation attempt can best be described as “Bitcoin stoichiometric ratio unit analysis calculation:”
So let’s try this type of estimation using bitcoin proof-of-work data and data released by OEMs. The Bitcoin network difficulty adjusts itself once every 2016 blocks, or about once every two weeks. This difficulty adjustment is to make up for the difference in block production speed, thereby making up for fluctuations in network computing power .
This relationship between difficulty and proof-of-work allows us to derive estimates of network computing power based on block generation rates and related difficulty levels. From the amount of work done at different levels of difficulty in the past ten years, we can roughly estimate the amount of SHA-256 hashes calculated by the Bitcoin network each year. As shown in the figure below, the unit is trillion hashes per year (Th/year ) Or trillion hashes per year. We can also do the same exercise on daily data to generate more subtle calculations.
By 2020, Bitcoin will have about 3934 yota hashes calculated on the network, or about 3934 septillion (“yota” and “septillion” are by far the largest prefix in Science International (SI), (10²⁴)) .
Now that we have an estimate of the annual hashrate, we must compile the efficiency data of the mining machine over the past 11 years to understand how much energy is needed to generate so much work.
Here, it is important to understand the different types of mining equipment that have provided work for the Bitcoin blockchain over the years. Each era and year has distinctly different characteristics of work permit efficiency, which will change the energy consumption of the network over time. The Bitcoin Genesis block consists of work from the CPU (Central Processing Unit), the block is finally adopted GPU (Graphics Processing Unit), then FPGA (Field Programmable Gate Array), and finally ASIC (Application Specific Integrated Circuit) Bitcoin network at an amazing speed Development .
Important note: Efficiency is defined as the useful work (terahash/joules-Th/J) done by the energy consumed to complete the work. However, ASIC original equipment manufacturers usually quote a type of heat rate specification, or the inverse of efficiency, which shows that energy consumption is more than useful work (joules per terahash-J/Th).
As you can see in the logarithmic scale chart below, in the past eight years, the popularity of Bitcoin mining ASICs has been steadily declining every year, which means that the efficiency of network mining has been improving .
Converting these data into annual average heat rates (below) shows a similar sharp decline throughout the history of Bitcoin mining. CPU, GPU, and FPGA benchmarks and published OEM power usage data are used to estimate network average calculations from 2009 to 2012. The miners announced by ASIC in 2020 are visualized above and below to show the continued decline in the heat rate of computing power, but they are discarded from the energy estimate because they are not yet publicly available.
So, now that we have compiled all the necessary data (annual computing power and annual computing power heat rate), let us combine them through engineers’ attempts at the stoichiometry of Bitcoin mining energy:
Simply multiply the work done each year (terahash/year) by the estimated annual heat rate (joules/terahash) of the miners on the system, and you get an estimate of joules/year. We will convert joules/year to kWh/year (1kWh equals 3.6 million joules). The following chart shows the estimated annual energy value.
However, this physics-based estimation method also has some problems:
The number of active miners calculated by efficiency level is unknown. This physics-based model assumes that all miner models on the market have equal participation in the release year.
The model also uses a step function of annual heat rate data as input. This annual data will suddenly change on the first day of each year. As old miners gradually retire and new miners begin to work, the gradual decline in heat rate will be more realistic.
It assumes that the old miners will retire in a year, which is also unlikely because the life cycle of the equipment is currently two years or more.
This may be a lower bound type of estimation.
Compare different network energy estimates
In the aforementioned calculation attempts, where do these annual energy consumption estimates fall? Interestingly, our two calculations, even using completely different methods and all the shortcomings discussed above—economic-based estimates (35.3 TWh) and physics-based estimates (40.17 TWh)—are very similar in value. . They also fall within the scope of various other popular estimates made by notable individuals, entities and institutions, as shown in the table below. All these estimators are quite similar, which makes the various estimators and the various methods and different assumptions used credible.
It is worth noting that Bitcoin network computing power (EH/s) seems to be decoupled from the general annual energy (TWh/year) estimation trend. If it is based on physical estimates, it may be due to the decline in the heat rate of SHA-256 ASIC mining equipment, if it is based on economic estimates, or due to halving and price stagnation.
The chart above shows a snapshot of annual energy estimates (TWh/year) at the time of publication, but some of these sources (University of Cambridge [C-BECI] and Alex de Vries [D-BECI]) are actually daily charts a few years ago These annual estimates are published on. This goes back to the previous discussion of energy and energy: logic should prevent plotting annual energy estimates on the solar axis.
In any case, I think it is necessary to compare these published estimates with our use of more continuous time series data going back to the end of 2017 (previous market highs). Economic and physical calculations, Cambridge estimates, and Digiconomist’s results are quite similar in time. These different estimation techniques have once again added some peer review and validity.
Our above estimation method seems to be in good agreement with other various daily interval annual energy estimates, so they are averaged together to create a composite Bitcoin Energy Index (CBEI), as shown in TWh/year below . Each estimate has different assumptions, different levels and sources of inaccuracy, so their combination may be more accurate. This composite estimate (CBEI) recently retested the 60 TWh threshold of Bitcoin’s annual total network energy consumption.
How does this comprehensive energy index compare to the computing power of the Bitcoin network over time? Around the beginning of 2019, CBEI showed a similar decoupling phenomenon. Computing power and energy continued to rise, energy consumption remained relatively stable, and ASIC heat rate and Bitcoin mining incentives had shrunk.
Interestingly, snapshot Bitcoin consumption estimates are usually extrapolated throughout the year, expressed in energy values of TWh/year, and there is no supporting time data or evidence. Daily network power estimates are more popular than all these annual energy consumption estimates drawn on a daily chart. The error of the chart is the amazing chart error that makes people misunderstand the data a lot: the annual energy estimate drawn on the daily axis. Therefore, I took the liberty to convert these daily interval estimates into daily power estimates graphs to correct the above-mentioned graph errors that caused data misunderstandings.
I proposed a comprehensive Bitcoin Power Index (CBPI) compilation from D-BECI and minimum, C-BECI maximum, minimum and estimate, as well as our estimates based on economics and physics above .
This CBPI combination estimates the instantaneous electricity consumption of Bitcoin, expressed in watts, which is a unit of electricity. The peak of CBPI recently reached nearly 7.58 GW, or 1.21 GW of about 6 DeLorean time machines.
CBPI in the environment
Such a large energy value is difficult to understand, especially in the case of each year, so let us take some quick comparisons to put these estimates in perspective:
- Banking system consumes 650TWh/ year
- Gold mining 200 TWh/ year
- PC and console games 75 TWh/ year
- Bitcoin Mining (CBEI) 60 TWh/ year
- Banknotes and coins 11 TWh/ year
- American Christmas lights 7TWh/ year
Based on our above estimates, the Bitcoin network consumes approximately 40-60 TWh/year, or about 0.15% of global annual power generation (26,700 TWh), and only about 0.024% of global total energy production (14, 421, 151 ktoe) . (Ktoe is also a unit of energy: 1,000 tons of oil equivalent, 11.36 MWh.)
Therefore, the energy consumption of Bitcoin today is only a major problem that many people think: it is only a small part of the growing human energy consumption. A century ago, Nikola Tesla proposed an interesting solution to this problem. As recently as September 2020, a study stated that nearly 76% of the Bitcoin network is powered by clean energy. In addition, please remember that once Einstein discovered the equivalence of mass and energy, mankind used the energy contained in the atom, and the energy to promote human progress has become very abundant.
