Is bitcoin's energy efficiency all we believe it to be?
Image credit: Foto 117620665 © Josefkubes | Dreamstime.com
Bitcoin is an obvious resource hog but we could easily be looking at more if the tech sector is not careful.
Belief is a powerful drug. It’s easy to believe that computer technology and electronics will make the world more efficient and reduce its dependence on fossil fuels, not least by reducing how much of that energy needs to be performed by physical things acting in a wasteful way. But this is the same set of core technologies that gave us bitcoin.
In a demonstration of how perverse incentives can be in economics, bitcoin and the other blockchain protocols that evolved around it fell into an arms race that make just burning currency for heat look vaguely sensible.
The first miners just used the Intel processors in desktop PCs. But that quickly changed as the value of bitcoin and interest in mining for it increased. The protocol behind the cryptocurrency was designed to ratchet up the amount of work needed to compute the hash codes that guarantee the authenticity of the blockchain to maintain a more or less constant rate of bitcoin creation. People armed with GPU cards able to crunch the hashes faster quickly displaced the x86s. After a while, the GPUs in turn gave way to teams able to deploy arrays of programmable-logic chips, each running custom hash-generating algorithms. Finally even they lost out to custom-made chips: the owners traded the millions of dollars it took to develop and fabricate them against the lower power efficiency and hash rates of GPUs and programmable logic.
This arms race has made it tough to estimate the pollution generated by bitcoin. Miners have their own arrays of accelerators, each plugged into electricity grids with different degrees of reliance on coal and gas. Some researchers have argued bitcoin miners even move their operations around depending on the availability of cheap electricity, with some apparently shifting between hydroelectric-rich areas during the Chinese rainy season to inner provinces closer to Russia where coal dominates at other times.
To be fair, some of the more lurid claims of energy consumption may be inflated. To address this issue, a team from the University of Southampton used machine learning to try to reduce the uncertainty in a project that found an earlier claim of mining operations generating some 70MtCO2e in 2017 was overestimated. The bad news was that the trajectory for bitcoin mining puts it on course to rise from just under 15MtCO2e in 2019. Though a smaller number, this is still more than the total emissions from Bolivia and similar small states.
Despite the quantities of air conditioning needed to maintain racks of accelerators cool enough to keep running, energy is likely not the biggest cost to miners. Belief has wound up costing them a lot more.
According to work published late last year by Charles Bertucci and colleagues working at a group of Parisian universities, the energy still winds up being a comparatively small part of the cost even though the energy required could now be as much as 100TWh each year, which works out to be a third of the total likely consumed by large-scale, non-cryptocoin data centres worldwide in 2020.
The conclusion Bertucci’s group came to as to why energy is responsible for less than a quarter of bitcoin miners’ costs is that they believe they have to keep buying new hardware to stay in the mining game. Though the miners have collected a significant quantity of bitcoin, as with the gold rushes of the 19th and 20th centuries, the hardware manufacturers have been cashing in on crypto. The implication is that miners are relying on further increases in the value of the tokens they hold as well as those they will win in the future to justify the continued purchases. Given bitcoin’s collapse in price in tandem with falls in the US and other stock markets, that belief may have taken a knock. But for the moment, many of those in the crypto world believe it is just a temporary setback in a continuing climb up the monetary-value mountain.
Measuring the net effect of bitcoin, assuming we have the right power-consumption numbers in the first place, is relatively easy. That is because, for the most part, it is not replacing much that is physical. You could argue that it’s replacing physical coins and notes, which take resources to mint though the former is readily and routinely recycled. In reality, it’s mostly substituting for other electronic transactions that managed to proceed pretty efficiently and have done for decades.
Working out the effects of other technologies is a trickier endeavour as explored in the latest issue of E&T in this piece about the issues faced by large AI models that are meant to run the world better and how rebound effects can occur, in which the convenience of digital activity makes the overall level of energy or resource usage increase. The rebound effect for automated driving is fairly clear to see, though hard to quantify. Others could be a lot more subtle. One example is video-conferencing.
The principle is, as we’ve seen with the Covid pandemic, online conferencing can substitute pretty well for travel that could well generate a lot more carbon. It could also generate rebound effects as it becomes easier to have geographically separate teams who then travel to see each other for more in-depth meetings, increasing the carbon from flights. The research so far tends to indicate that video-conferencing does result in less travel overall, though researchers such as Kelly Widdicks, lecturer at Lancaster University, believe a lot more is needed.
If we bring in the possible evolution of video-conferencing to virtual reality and the metaverse, the resources needed for the electronics may make it even less of a clear trade. There are limits on how much a headset can do before it becomes too hot to wear, but the industry is looking at offloading much of the processing to local servers and the cloud. Given the level of detail that full VR needs compared to what we today accept from video-conferencing will likely result in increases in both electricity and in embodied emissions: from all the silicon and other high-precision devices needed throughout the network.
This week, ABI Research published a report that underlined how much change a full metaverse implementation would need at every level in the network. And, as the ABI analysts point out, crypto may wind up embedded in these applications through non-fungible tokens (NFTs) and similar digital-money and ID devices. Though Ethereum and other blockchains are moving to less energy-intensive protocols, supporting NFTs directly rather than other, potentially more efficient transaction mechanisms, could make the metaverse quite the power hog.
As with AI, there is a clear need to measure how much these technologies need in the way of resources. But even with greater transparency, there is plenty that could be missed as the effects are so subtle given that we cannot be sure how much metaverse sessions would balance against air travel or other activities with high resource demands.
Widdicks and colleagues argue that the solution may lie in the expanded use of carbon pricing, as long as it is applied across the lifetime of the systems used to implement these services. One notional advantage of carbon pricing is that it lets the market drive towards a minimum instead of relying on measurements that may not be balanced all that way and, in turn, lead to perverse outcomes. ”We will still need transparent evidence on ICT’s emissions and the reductions in these as we align the digital sector, and global sectors generally, with the Paris Agreement,” Widdicks added.
Is it too late to start? A hiatus in both AI and the metaverse now seems possible as the FAANG companies batten down the hatches in anticipation of a slowdown or recession, which provides some breathing room. However, there is a clear danger that the next wave could get going before governments can agree on how to approach the issue, not least because an expanded carbon-pricing regime would be politically difficult territory and measurement on its own may look a lot easier to implement even if it far less than optimal. The first step is to change beliefs. Digital is not necessarily the efficient option. It depends how you go about it.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.