What is the marginal energy cost of a bitcoin transaction?

Question

According to this web page: https://arstechnica.com/tech-policy/2017/12/bitcoins-insane-energy-consumption-explained/

...one bitcoin transaction consumes about 250 kWh of energy.

Is this the marginal energy cost of each additional transaction, or the average electricity cost of continuous mining (that would happen no matter whether there are transactions or not) divided by the relatively low transaction count?

What is the marginal energy cost of one additional bitcoin transaction? I.e. if the number of transactions would increase by one, by how much would the energy consumption of the bitcoin network increase?

Would the energy consumption of the bitcoin network double if the number of transactions doubled?

If each bitcoin transaction consumes 250 kWh, that would seem absurdly high. I do several hundred ordinary bank or card transactions per year, making the energy use about 75 000 kWh per year if I used bitcoin, which is 25 times higher than my current direct electricity use and probably still 10 times higher than my current direct and indirect electricity use combined.

Related: https://skeptics.stackexchange.com/questions/39910/does-one-transaction-on-bitcoin-require-215-kwh (I initially considered asking this on Skeptics, but they seem to have a habit of closing perfectly valid questions).

Answer 1

Yes, this number is the total energy consumption of mining, divided by the number of transactions confirmed. The total energy consumption has basically nothing to do with the actual number of transactions, at least not directly.

There's an indirect relationship: since the total number of transactions (or rather, total size in bytes) that can be confirmed per block is limited by the protocol, then if there are more transactions, you would expect transaction fees to increase. This in turn would increase the revenue of mining, and hence you would expect more miners to enter the network to compete for this revenue, increasing total energy consumption. But since transaction fees are a relatively small fraction of total revenue (it's dominated by block reward), this relationship will be far from linear.

It's hard to say what the actual marginal energy cost of one transaction is. One way to look at it is that, roughly speaking, every full node on the network will check the transaction for validity. Looking at the logs from my own server as it synchronizes with the network, in one second it's able to validate blocks containing about 5000 transactions. At full CPU, this machine draws about 15 watts, so that's about 3 mJ per transaction. (Of course, much of this is overhead, since the machine is powered on regardless and used for other things.) Now https://coin.dance/nodes estimates around 10000 full nodes currently on the network, so if we extrapolate from my machine, we have 30 J per transaction, or 8.3e-6 kWH. If you make 1000 transactions, that's 30 kJ or 8.3e-3 kWH.

If you want to exclude overhead, you could suppose that if not for verifying transactions, my server would have been idling its CPU during that time, in which state it draws about 10 watts. So the marginal CPU usage is about 5 watts, or about 1/3 of the values above. This logic takes you to 10 kJ per transaction. On the other hand, my computer is pretty low-power compared to most (it uses a mobile CPU), so you may instead want to increase those values. But this should be on the right order of magnitude.