Calculating a computers carbon footprint
To computing now and what’s been dubbed ‘the hidden cost of big data’. As computers have become more powerful, they’ve enabled scientists to probe complex problems in new ways, often by building virtual “models”, or analysing huge databases looking for relationships between different pieces of information. It’s given us breakthroughs in weather prediction and ocean currents, and enabled us to piece back together fragments of genetic code from over a million years ago. But all this computer power adds up to a serious energy bill, and that, in turn, means a big carbon cost. Here to explain how big, why and what we can do about it is the University of Cambridge’s Loïc Lannelongue…
Loic - We are a lab working on computational biology. So we use all these big algorithms and it was around the time of the Australian bush fires, part of our lab is in Melbourne, and around the same time a paper came out showing the carbon footprint of an artificial intelligence model. And we thought, well, our models in biology are just as demanding. So we wanted to calculate the carbon footprint of what we were doing. And we thought that would be a one week project, it's going to be really quick and we are just going to move on to what we were doing before. And it turns out, as you said, that no one outside of AI was really looking at it. So we had to first build a tool to do it and then look at it. So two years later, still here.
James - Practically, then, quite a challenge?
Loic - Yeah. Basically, to know how much energy is needed depends on what you're doing and what hardware you're using. And then you need to know what the carbon footprint of getting that much electricity is. Data centres are mostly connected to the power grid, so that depends where you are in the world.
James - And are you able to put some numbers on this, put this into context perhaps for me?
Loic - One striking number we came across is what's called genome-wide association studies. Anytime you come across an article saying, we found the gene for X, that's usually what's done. So you look at a lot of people, you look at the entire genome and you try to figure out what gene is responsible for what disease. We found that if you do it for a thousand different traits (one trait can be height, weight or a specific disease) it can be as much as 17 tonnes of co2. For reference, you could fly to Paris and back every morning to have breakfast there for six months.
James - That's remarkable. What's occurring to me now is to wonder if the current energy crisis and the high energy prices will mean that those areas of research, which are the most resource intensive, whether that will have some knock on effect as to whether that research can carry on at this time?
Loic - I think it may force people to think a little bit more about it. And I know I've talked with academics in charge of data centers who said, yes, we can only use half the data centers because the energy bill is too high. And so what computing used to be considered completely free, and people don't think twice about using computing compared to someone in a lab, in a physics lab where everything costs a lot of money. Uh, computing is, it doesn't cost that much, but maybe if that changes, it'll force people to not waste resources, but hopefully it won't hinder discoveries.
James - So what can we do about this? How can we reduce the carbon footprint of computational research?
Loic - So scientists can maybe use data centres that are more energy efficient, because a lot of the power is actually not used for the computers, but it's used to keep the facilities nice and cool. So if data centres are more efficient, we could reduce the total carbon footprint. Also, things just like using the latest version of a software can work sometimes. And I'm sure everyone has come across the fact that you update windows on your laptop and everything stops working for a week, and it's the same thing in any computational lab. So no one likes updating things, but actually sometimes it can have great effects. But actually just estimating carbon footprints regularly so each user is aware of what their carbon footprint is, but also at institutional levels, so knowing what the total carbon footprint of all the research being done in the department. If you do that over time that's really important because even if you try to be very sensible and make everything more efficient, there's this nasty thing called the rebound effect that means if you make a tool 10 times more efficient, the only result is scientists will use it a hundred times more. And at the end of the day, you didn't save any energy. So it's important to measure things over time to make sure that we'll need to change how we think about computing costs and the cultural changes needed.
James - And, just briefly, on that point of cultural change, could this perhaps be the start of a movement towards science more broadly becoming obligated to acknowledge the carbon intensivity of their research?
Loic - I certainly hope so. Funding bodies are becoming more aware now that this is an important topic and a lot of them are putting in efforts. Actually, we've teamed up with some funding bodies and some research institutes, and we put together a roadmap of where we think environmental, sustainable professional science should go moving forward. It's coming out in the next couple months. So, yeah, I think scientists are going in this direction. There's hope.