I recently created a thread that on Twitter regarding the lower-bound estimates for how much electricity the Bitcoin blockchain consumed using publicly available numbers.
The first part of this post is a slightly modified version of that thread.
The second part of this post, below part 1, includes additional information on Bitcoin Cash, Ethereum, Litecoin, and Monero using the same type of methodology.
The original nested thread started by explaining why a proof-of-work (PoW) maximalist view tries to have it both ways.
You cannot simultaneously say that Bitcoin is – as measured by hashrate – the “most secure public chain” and in the same breath say the miners do not consume enormous quantities of energy to achieve that. The fundamental problem with PoW maximalism is that it wants to have a free energy lunch.
All proof-of-work chains rely on resource consumption to defend their network from malicious attackers. Consequently, a less resource intensive network automatically becomes a less secure network.1 I discussed this in detail a few years ago.
Part 1: Bitcoin
Someone recently asked for me to explain the math behind some of Bitcoin’s electricity consumption, below is simple model using publicly known numbers:
- the current Bitcoin network hashrate is around 50 exahashes/sec
- the most common mining hardware is still the S9 Antminer which churns out ~13 terahashes/sec
Thus the hashrate pointed at the Bitcoin network today is about 50,000,000 terashashes.
Dividing one from the other, this is the equivalent of 3,846,000 S9s… yes over 3 million S9s.
While there is other hardware including some newer, slightly more energy efficient gear online, the S9 is a good approximate.
Because the vast majority of these machines are left on 24/7, the math to estimate how much energy consumption is as follows:
- in practice, the S9 draws about 1,500 watts
- so 1,500 x 24 = 36kWh per machine per day
Note: here’s a good thread explaining this by actual miners.
In a single month, one S9 will use ~1,080 kWh.
Thus if you multiply that by 3,846,000 machines, you reach a number that is the equivalent of an entire country.
- for a single day the math is: ~138.4 million kWh / day
- annually that is: ~50.5 billion kWh / year
For perspective, ~50.5 billion kWh / year would place the Bitcoin network at around the 47th largest on the list of countries by electricity consumption, right between Algeria and Greece.
But, this estimate is probably a lower-bound because it doesn’t include the electricity consumed within the data centers to cool the systems, nor does it include the relatively older ASIC equipment that is still turned on because of local subsidies a farm might receive.
According to a recent Wired article:
In Iceland, the finance minister has warned that cryptocurrency mining – which uses more power than the nation’s entire residential demand – could severely damage its economy.
Recent analysis from a researcher at PwC places the Bitcoin network electricity consumption higher, at more than the level of Austria which is number 39th on that list above. Similarly, a computer science professor from Princeton estimates that Bitcoin mining accounts for almost 1% of the world’s energy consumption.2
Or to look at it in a different perspective: the Bitcoin network is consuming the same level of electricity of a developed country – Austria – a country that generates ~$415 billion per year in economic activity.
Based on a recent analysis from Chainalysis, it found that Bitcoin – which is just one of many proof-of-work coins – handled about $70 million in payments processed for the month of June. Yet its cost-per-transaction (~$50) is higher than at any point prior to November 2017.
You don’t have to be a hippy tree hugger (I’m not) to clearly see that a proof-of-work blockchains (such as Bitcoin and its derivatives) are currently consuming significantly more resources than they create. However this math is hand-waved away on a regular basis by coin lobbyists.
The figure also didn’t include the e-waste generated from millions of single-use ASIC mining machines that are useful for about ~12 months; or the labor costs, or building rents, or transportation, etc. These ASIC-based machines are typically discarded and not recycled.
Part 2: Bitcoin Cash
With Bitcoin Cash the math and examples are almost identical to the Bitcoin example above. Why? Because they both use the same SHA256 proof-of-work hash function and as a result, right now the same exact hardware can be used to mine both (although not simultaneously).3
So what do the numbers look like?
The BCH network hashrate has been hovering around 4 – 4.5 exahashes the past month. So let’s use 4.25 exahashes.
Note: this is about one order of magnitude less hashrate than Bitcoin so you can already guesstimate its electricity usage. But let’s do it by hand anyways.
An S9 generates ~13 TH/s and 4.25 exahashes is 4.25 million terahashes.
After dividing: the equivalent of about 327,000 S9s are used.
Again, these machines are also left on 24/7 and consume about 36 kWh per machine per day. So a single S9 will use ~1,080 kWh per month.
- 327,000 S9s churning for one day: ~11.77 million kWh / day
- Annually this is: ~4.30 billion kWh / year
To reuse the comparison above, what country’s total electricity consumption is Bitcoin Cash most similar to?
Around 124th, between Moldova and Cambodia.
How much economic activity does Moldova and Cambodia generate with that electricity consumption? According to several sources, Cambodia has an annual GDP of ~ $22 billion and Moldova has an annual GDP of ~$8 billion.
For comparison, according to Chainalysis, this past May, Bitcoin Cash handled a mere $3.7 million in merchant payments, down from a high of $10.5 million in March a couple months before.
Also, the Bitcoin Cash energy consumption number is likely a lower-bound as well for the reasons discussed above; doesn’t account for the e-waste or the resources consumed to create the mining equipment in the first place.
This illustrates once again that despite the hype and interest in cryptocurrencies such as Bitcoin and Bitcoin Cash, there is still little real commercial “activity” beyond hoarding, speculation, and illicit darknet markets. And in practice, hoarding is indistinguishable from losing a private key so that could be removed too. Will mainstream adoption actually take place like its vocal advocates claim it will?
Part 3: Ethereum
So what about Ethereum?
Its network hashrate has been hovering very closely to 300 TH/s the past month
At the time of this writing, the Ethereum network is still largely dominated by large GPU farms. It is likely that ASICs were privately being used by a handful of small teams with the necessary engineering and manufacturing talent (and capital), but direct-to-consumer ASIC hardware for Ethereum didn’t really show up until this summer.
There are an estimated 10 million GPUs churning up hashes for the Ethereum network, to replace those with ASICs will likely take more than a year… assuming price stability occurs (and coin prices are volatile and anything but stable).
For illustrative purposes, what if the entire network were to magically switch over the most efficient hardware -the Innosilicon A10 – released next month?
Innosilicon currently advertises its top machine can generate 485 megahashes/sec and consumes ~ 850 W.
So what is that math?
The Ethereum network is ~300 TH/s which is around 300,000,000 megahashes /sec.
Quick division: that’s the equivalent of 618,557 A10 machines.
Again, each machine is advertised to consume ~850 W.
- in a single day one A10 consumes: 20.4 kWh
- in a month: ~612 kWh
So what would 618,557 A10 machines consume in a single day?
– about 12.6 million kWh / day
– about 4.6 billion kWh / year
That works out to be between Afghanistan or Macau. However…
Before you say “this is nearly identical to Bitcoin Cash” keep in mind that the Ethereum estimate above is the lowest of lower-bounds because it uses the most efficient mining gear that hasn’t even been released to the consumer.
In reality the total energy consumption for Ethereum is probably twice as high.
Why is Etherum electricity usage likely twice as high as the example above?
Because each of the ~10 million GPUs on the Ethereum network is significantly less efficient per hash than the A10 is. 4 Note: an example of a large Ethereum mine that uses GPUs is the Enigma facility.
For instance, an air-cooled Vega 64 can churn ~41 MH/s at around 135 W which as you see above, is much less efficient per hash than an A10.
If the Ethereum network was comprised by some of the most efficient GPUs (the Vega 64) then the numbers are much different.
Starting with: 300,000,000 MH/s divided by 41 MH/s. There is the equivalent to 7.32 million Vega GPUs generating hashes for the network which is more in line with the ~10 million GPU estimate.
- one Vega 64 running a day consumes ~3.24 kWh
- one Vega 64 running a month: ~77.7 kWh
If 7.32 million Vega equivalent GPUs were used:
- in a day: ~ 23.71 million kWh
- in a year: ~8.65 billion kWh
That would place the Ethereum network at around 100th on the electricity consumption list, between Guatemala and Estonia.
In terms of economic activity: Guatemala’s GDP is around $75 billion and Estonia’s GDP is around $26 billion.
What is Ethereum’s economic activity?
Unlike Bitcoin and Bitcoin Cash, the stated goal of Ethereum was basically to be a ‘censorship-resistant’ world computer. Although it can transmit funds (ETH), its design goals were different than building an e-cash payments platform which is what Bitcoin was originally built for.
So while merchants can and do accept ETH (and its derivatives) for payment, perhaps a more accurate measure of its activity is how many Dapp users there are.
There are a couple sites that estimate Daily Active Users:
- State of the Dapps currently estimates that there are 8.93k users and 8.25K ETH moving through Dapps
- DappRadar estimates a similar number, around 8.37k users and 8.57K ETH moving through Dapps
Based on the fact that the most popular Dapps are decentralized exchanges (DEXs) and MLM schemes, it is unlikely that the Ethereum network is generating economic activity equivalent to either Guatemala or Estonia.5
For more on the revenue Ethereum miners have earned and an estimate for how much CO2 has been produced, Dominic Williams has crunched some numbers. See also this footnote.6
Part 4: Litecoin
If you have been reading my blog over the past few years, you’ll probably have seen some of my Litecoin mining guides from 2013 and 2014.
If you haven’t, the math to model Litecoin’s electricity usage is very similar to both Bitcoin and Bitcoin Cash. From a mining perspective, the biggest difference between Litecoin and the other two is that Litecoin uses a hash function called scrypt, which was intended to make Litecoin more “ASIC-resistant”.
Spoiler alert: that “resistance” didn’t last long.
Rather than diving into the history of that philosophical battle, as of today, the Litecoin network is composed primarily of ASIC mining gear from several different vendors.
One of the most popular pieces of equipment is the L3+ from Bitmain. It’s basically the same thing as the L3 but with twice the hashrate and twice the power consumption.
So let’s do some numbers.
Over the past month, the Litecoin network hashrate has hovered around 300 TH/s, or 300 million MH/s.
Based on reviews, the L3+ consumes ~800 W and generates ~500 MH/s.
So some quick division, there are about 600,000 L3+ machines generating hashes for the Litecoin network today.
As an aggregate:
- A single L3+ will consume 19.2 kWh per day
- So 600,000 will consume 11.5 million kWh per day
- An annually: 4.2 billion kWh per year
Coincidentally this is roughly the same amount as Bitcoin Cash does as well.
So it would be placed around 124th, between Moldova and Cambodia.
Again, this is likely a lower-bound as well because it assumes the L3+ is the most widely used ASIC for Litecoin but we know there are other, less efficient ones being used as well.
What about activity?
While there are a few vocal merchants and a small army of “true believers” on social media, anecdotally I don’t think I’ve spoken to someone in the past year who has used Litecoin for any good or service (besides converting from one coin to another).
We can see that — apart from the bubble at the end of last year — the daily transaction volume has remained roughly constant each day for the past 18 months. Before you flame me with a troll account, consider that LitePay collapsed before it could launch, partly because Litecoin still lacks a strong merchant-adopting ecosystem.
In other words, despite some support by merchant payment processors, its current usage is likely as marginal as Bitcoin and Bitcoin Cash.
Part 5: Monero
The math around Monero is most similar to Ethereum in that it is largely dominated by GPUs.
In fact, earlier this year, a large number of Monero developers convinced its boisterous userbase to fork the network to prevent ASICs from being used. This resulted in four Monero forks and basically all of them are dominated by high-end GPUs.
For the purposes of this article, we are looking at the fork that has the highest hashrate, XMR. Over the past month its hashrate has hovered around 475 MH/s.
Only 475 MH/s? That may sound like a very diminutive hashrate, but it is all relative to what most CPU and GPU hashrate performance is measured in Monero and not other coins.
For example, MoneroBenchmarks lists hundreds of different system configurations with the corresponding hashrate. Similarly there are other independent testing systems that provide public information on hashrates.
Let’s take that same Vega 64 used above from Ethereum. For Monero, based on tweaking it generates around 2000 hashes/sec and consumes around 160 W.
So the math is as follows:
- 475,000,000 hashes/sec is the current average hashrate
- A single Vega 64 will generate about 2000 hashes/sec
- The equivalent of 237,500 Vega 64s are being used
- Each Vega 64 consumes about 3.84 kWh per day
- So 237,500 Vega 64s consume 912,000 kWh per day
- And in a year: 332 million kWh
The 332 million kWh / year figure is a lower-bound because like the Ethereum Vega 64 example above: it doesn’t include the whole mining system, all of these systems still need a CPU with its own RAM, hard drive, and so forth.
As a result, the real electricity consumption figure is much closer to Haiti than Seychelles, perhaps even higher. Note: Haiti has a ~$8.4 billion economy and the GDP of Seychelles is ~$1.5 billion.
So what about Monero’s economic activity? Many Monero advocates like to market it as a privacy-focused coin. Some of its “core” developers publicly claimed it would be the best coin to use for interacting with darknet markets. Whatever the case may be, compared to the four above, currently it is probably the least used for commercial activity as revealed by its relative flat transactional volume this past year.
Above were examples of how much electricity is consumed by just five proof-of-work coins. And there are hundreds of other PoW coins actively online using disproportionate amounts of electricity relative to what they process in payments or commerce.
This article did not dive into the additional resources (e.g., air conditioning) used to cool mining equipment. Or the subsidies that are provided to various mining farms over the years. It also doesn’t take into account the electricity used by thousands of validating nodes that each of the networks use to propagate blocks each day.
It also did not include the huge amount of semiconductors (e.g. DRAM, CPUs, GPUs, ASICs, network chips, motherboards, etc.) that millions of mining machines use and quickly depreciate within two years, almost all of which becomes e-waste.7 For ASIC-based systems, the only thing that is typically reused is the PSU, but these ultimately fail as well due to constant full-throttle usage.
In summation, as of this writing in late August 2018:
- Bitcoin’s blockchain likely uses the same electricity footprint as Austria, but probably higher
- Bitcoin Cash’s blockchain is at least somewhere between Moldova and Cambodia, but probably higher
- Ethereum’s blockchain is at least somewhere between Guatemala and Estonia, but probably higher
- Litecoin’s blockchain is at least somewhere between Moldova and Cambodia, but probably higher
- One of Monero’s blockchains is at least somewhere between Haiti and Seychelles, but probably higher
Altogether, these five networks alone likely consume electricity and other resources at an equivalent scale as The Netherlands especially once you begin to account for the huge e-waste generated by the discarded single-use ASICs, the components of which each required electricity and other resources to manufacture. Perhaps even higher when costs of land, labor, on-going maintenance, transportation and other inputs are accounted for.
The Netherlands has the 18th largest economy in the world, generating $825 billion per annum.
I know many coin supporters say that is not a fair comparison but it is. The history of development and industrialization since the 18th century is a story about how humanity is increasingly more productive and efficient per unit of energy.
Proof-of-work coins are currently doing just the opposite. Instead of being more productive (e.g., creating more outputs with the same level of inputs), as coin prices increase, this incentivizes miners to use more not less resources. This is known as the Red Queen Effect.89
For years, proof-of-work advocates and lobbying organizations like Coin Center have been claiming that the energy consumption will go down and/or be replaced by renewable energy sources.
But this simply cannot happen by design: as the value of a PoW coin increases, miners will invest more capital in order to win those coins. This continues to happen empirically and it is why over time, the aggregate electricity consumption for each PoW coin has increased over time, not decreased. As a side-effect, cryptocurrency mining manufacturers are now doing IPOs.10
Reporters, if you plan to write future stories on this topic, always begin by looking at the network hashrate of the specific PoW coin you are looking at and dividing it by the most common piece of mining hardware. These numbers are public and cannot be easily dismissed. Also worth looking at the mining restrictions and bans in Quebec, Plattsburgh, Washington State, China, and elsewhere.
To front-run an example that coin promoter frequently use as a whataboutism: there are enormous wastes in the current traditional financial industry, removing those inefficiencies is a decades-long ordeal. However, as of this writing, no major bank is building dozens of data centers and filling them with single-use ASIC machines which continuously generate random numbers like proof-of-work coins do. That would be rightly labeled as a waste.
In point of fact, according to the Federal Reserve:
In the aggregate, U.S. PCS systems process approximately 600 million transactions per day, valued at over $12.6 trillion.
It shouldn’t take the energy footprint of a single country, big or small, to confirm and settle electronic payments of that same country. The fact of the matter is that with all of its headline inefficiencies (and injustices), that the US financial system has — the aggregate service providers still manage to process more than three orders of magnitude more in transactional volume per day than all of the major PoW coins currently do.11 And that is just one country.
Frequent rejoinders will be something like “but Lightning!” however at the time of this writing, no Lightning implementation has seen any measurable traction besides spraying virtual graffiti on partisan-run websites.
Can the gap between the dearth of transactional volume and the exorbitantly high cost-per-transaction ratio be narrowed? Does it all come down to uses? Right now, the world is collectively subsidizing dozens of minuscule speculation-driven economies that in aggregate consumes electricity on par with the 18th largest real economy, but produces almost nothing tangible in exchange for it.
What if all mining magically, immediately shifted over to renewable energy?
Izabella Kaminska succinctly described how this still doesn’t solve the environmental impact issues:
Renewable is displacement. Renewable used by bitcoin network is still renewable not used by more necessary everyday infrastructure. Since traditional global energy consumption is still going up, that ensures demand for fossil continues to increase.
To Kaminska’s point, in April a once-shuttered coal power plant in Australia was announced to be reopened to provide electricity to a cryptocurrency miner. And just today, a senator from Montana warned that the closure of a coal power plant “could harm the booming bitcoin mining business in the state.”
It is still possible to be interested in cryptocurrencies and simultaneously acknowledge the opportunity costs that a large subset of them, proof-of-work coins, are environmental black holes.12
If you’re interested in discussing this topic more, feel free to reach out. If you’re looking to read detailed papers on the topic, also highly recommend the first two links listed below.
- Bitcoin’s Growing Energy Problem
- Decarbonizing Bitcoin
- The Marginal Cost of Cryptocurrency
- Bitcoins: Made in China
- The State of Cryptocurrency Mining
- The Environmental Case Against Bitcoin
- Bitcoin could cost us our clean-energy future
- Bitcoin’s stupendous power waste is green, apparently — bad excuses for Proof-of-Work
- If the market value of a coin decreases, then because hashrate follows price, in practice hashrate also declines. See also a ‘Maginot Line’ attack [↩]
- Another estimate is that Bitcoin’s energy usage creates as much CO2 as 1 million transatlantic flights. [↩]
- There have been proposals from various developers over the years to change this hash function but at the time of this writing, both Bitcoin and Bitcoin Cash use the same one. [↩]
- And because many of these mining systems likely use more-powerful-than-needed CPUs. [↩]
- Note: Vitalik Buterin highlighted this discrepancy earlier this year with the NYT: The creator of Ethereum, Vitalik Buterin, is leading an experiment with a more energy-efficient way to create tokens, in part because of his concern about the impact that the network’s electricity use could have on global warming. “I would personally feel very unhappy if my main contribution to the world was adding Cyprus’s worth of electricity consumption to global warming,” Mr. Buterin said in an interview. [↩]
- At 8.65 billion kWh * $0.07 / kWh comes to around $600 million spent on electricity per year. Mining rewards as of this writing: 3 ETH * $267 / ETH * 6000 blocks / day equals to $4.8 million USD / day. Or ~$1.7 billion per year. This includes electricity and hardware. Thanks to Vitalik for double-checking this for me. [↩]
- Just looking at the hash-generating machines, according to Chen Min (a chip designer at Avalon Mining), as of early November 2017, 5% of all transistors in the entire semiconductor industry is now used for cryptocurrency mining and that Ethereum mining alone is driving up DRAM prices. [↩]
- See Chapter 3 [↩]
- As described in a Politico article this past spring: “To maintain their output, miners had to buy more servers, or upgrade to the more powerful servers, but the new calculating power simply boosted the solution difficulty even more quickly. In effect, your mine was becoming outdated as soon as you launched it, and the only hope of moving forward profitably was to adopt a kind of perpetual scale-up: Your existing mine had to be large enough to pay for your next, larger mine.” [↩]
- Following the dramatic drop in coin prices since January, Nvidia missed its revenue forecast from cryptocurrency-related mining: Revenues from miners were $289 million in Q1, which was about 10% of Nvidia’s revenue. The forecast for Q2 was $100 million and the actual revenues ended up being $18 million. [↩]
- On average, the Bitcoin network confirms about 300,000 transactions per day. A lot of that is not commercial activity. Let’s take the highest numbers from Chainalysis and assume that each major cryptocurrency is processing at least $10 million in merchant transactions a day. They aren’t, but let’s assume that they are. That is still several orders of magnitude less than what US PCS systems do each day. [↩]
- The ideological wing within the cryptocurrency world has thus far managed to convince society that negative externalities are ‘worth the cost.’ This narrative should be challenged by both policy makers and citizens alike as everyone must unnecessarily bear the environmental and economic costs of proof-of-work blockchains. See also the Bitcoin Energy Consumption Index from Digiconomist and also Bitcoin is not a good fit for renewable energy. Here’s why. [↩]