@volume I happen to do research on quantum computing. I am not going to try and change your mind.
Quantum computers will become more mainstream and it could be within 20 years. For cracking crypto, it will be sooner, but those machines will consume a lot of electricity.

Quantum computers have advantages for certain classes of problems, and that includes weather and climate simulations, so eventually some good could come out of that. But quantum computing is a distraction. It is not what we need to tackle the climate crisis.

@frox I was a bit vague on purpose there by not specifying "the field". There is no such consensus in the wider computing science community because the AI and blockchain and IoT researchers want to keep on doing their thing. But in the low-carbon/sustainable computing field, there is a consensus that there is no tech fix. The best paper to my mind is this one: https://www.sciencedirect.com/science/article/pii/S2666389921001884
It looks at AI, IoT and blockchain.

Quantum computing is not covered but it is a red herring because nobody really sees it as a viable low-carbon technology in the next two decades.

The issue with space industry and global warming is discussed here: https://aip.scitation.org/doi/abs/10.1063/5.0090017?journalCode=phf

@wim_v12e @frox wanna chip in as an incredibly disillusioned compsci AI person that language models like GPT-3 are highly centralized research computation efforts that throw grant money at more compute power nobody else can access. And that's just for the specific problem of generating some convincing text. We are just understanding about how much computational resources is required on par to make this kind of effort even *feasible* , much less what kinds of average power draw we can expect.

There are probably better approaches for natural language generation, but I'd suspect a computer scientist might have to actually tackle with the knowledge base of linguistics as well (this may be a bit snarky, so please forgive me). While computer scientist graduates generally can tell you what a regular grammar is there is zero chance they've read Chomsky's "On Language" or express any interest in actually tackling the problems of that field to build better AI. Unfortunetly this general syndrome of computers being able to do the abstract thinking for us rather than believing that computers have not, in fact, already obseleted every other field of science, has permeated through many a student exposed to the hubris of an alumni speaker who ran git checkout on hotdog not hotdog.

@thufie @frox Thanks for sharing your insights!

@frox In terms of how much consensus there is on this, I was reviewing a draft whitepaper on this topic (what they call "Twin Transition", green+digital) by the organisation grouping twenty one of Europe's most distinguished research-intensive universities in sixteen countries, and they echo the views in that article (and fwiw, in my article https://wimvanderbauwhede.github.io/articles/frugal-computing/)

@wim_v12e Does performance per Watt really not increase exponentially anymore? My main computer is now 3x as fast (when clocked down to consume the same 30W of energy) as my previous home-server was. Processors in notebooks have reduced their consumption and are much better at clocking down when not needed. @frox

@ArneBab @frox Good point, thank you, I need to clarify that. It is still exponential but it's less than linear in a log-log graph, because the time required to double performance per Watt gets longer and longer. The projection is that it will saturate in the next two decades. I will amend that in the article. It does not really matter as the projected growth in computation is higher than the project rate of improvement in performance per Watt.

@wim_v12e Thank you! Did you take into account that electricity production might have much lower CO₂ emissions if we manage to get to renewable Energy? @frox

@ArneBab @frox Yes of course, but the all projections I have found indicate that this will not be in the next two decades (if we can get to 30% by 2040 it will be a success), and that is the time we have to reduce emissions in order to keep warming below 2degC.

@wim_v12e Do you mean on global scale? Wouldn’t a CO₂ cost then cause a differentiated computing environment? Countries with more renewable energy would be able to utilize the more consuming devices. Germany is already at 38% renewable electricity. @frox

@ArneBab @frox Yes, on a global scale. For global warming that is the only scale that matters.

It is not as if any country already has excess renewable energy, and the global energy demand is growing. The scale at which computing is projected to grow could easily gobble up all that renewable energy and leave the rest to burn fossil fuels. This is very much the strategy of Google, Microsoft etc. They can claim they're green at the expense of the rest of the world.

@wim_v12e So if I understand it correctly, the question would be: How much "computing"¹ should every human have available? How can we change computing such that the available renewable energy in 2040 suffices for that? @frox

¹: for lack of a better term.

@ArneBab @frox In the end, it is about utility. The problem with the current system is that utility is only considered in terms of money, so a company's utility of computing is determined by the profit, not by its benefit to society. There is no incentive to take into account the externalities (ecological as well as social etc).

The ideal situation is one where it is profitable to be green, and similar for social responsibility. But to get there, we can't keep the status quo. I think we need a combination of consumer demands and government intervention.
But for both of these the awareness needs to be there that unlimited growth in computing is disastrous. So that is my main message.

On the tech/science side, I strongly believe that we can achieve computational tasks at a fraction of the current energy cost. There simply has been no incentive to get there. But doing that on its own is not enough because then the demand would simply increase, offsetting the gains.

@ArneBab @frox That was a long answer and maybe not entirely answering you point, so to be clear, I agree, how much compute per person can we afford with the current emissions budget, and what can we do to optimise that.

@wim_v12e I like your long answer. For ways to get there: optimizing hot code paths — either by working in a fast language from the start or by delegating them to a fast language — can easily make a factor 10 to 100 difference; according to the benchmarksgame: https://benchmarksgame-team.pages.debian.net/benchmarksgame/box-plot-summary-charts.html — you could say „never use Python“ until one pulls out Cython and goes down to the speed of C for 95% of the runtime while writing 95% of the code in fewer lines (that consume high-powered developer compute). @frox

@wim_v12e One more thing to keep in mind: Javascript proved that throwing enough money at it can make every language fast, but that there’s still a price to be paid for hard to optimize language design (in this case in memory: speculative pre-compilation of likely combinations of types) @frox

@ArneBab @frox Energy-aware programming languages would be good too, but it's a complex issue because it's the full-system energy that matters, not purely the energy used by the program.

@ArneBab @frox
That is definitely one way to make computing more efficient. There are other ways that need to be considered, in particular better algorithms, better compilers and minimising network data transfers for applications.

@wim_v12e Do you have data about which kinds of program activity consume the most energy? Something akin to the latency pyramid,¹ but for energy use? @frox

¹: https://summerofhpc.prace-ri.eu/wp-content/uploads/2020/08/latency-pyramid.jpg

@Gargron @wim_v12e @EU_Commission if this is the case, and the only value is in receiving poorly-considered propaganda posts in a single direction instead of engaging in a public form, then it might be best to fediblock social dot network dot europa dot eu

but even then a reply with a correction where the public who are victims of this misinformation might see it is far more valuable than talk about some completely invisible unaccountable chain of command

@gargron @EU_Commission If the Commission wants to communicate to the public, then they have to communicate back the public's views to the decision makers as well.

@wim_v12e @gargron @EU_Commission

I live in Brexit island (I didn't support or vote for it) so no longer can contact MEPs etc; but the way the Commission have set this up with 0 replies to the public comes across almost like a 1980s propaganda operation (even though I agree with and support a lot of what they are saying and doing!)

It was a perception of Brussels being "unaccountable" and "untouchable" which in part led to Brexit in my country..

@vfrmedia @wim_v12e @EU_Commission Do you think that Ursula von der Leyen is sitting on the Mastodon account or do you think that social media managers have the authority to off the cuff respond on behalf of the European executive body?

@gargron @wim_v12e @EU_Commission

I don't expect them to be able to respond informally, but I've worked in the UK Civil Service myself (pre social media) and there was a whole unit in the Chief Executives department that logged and replied to correspondence from citizens and activist groups, who did usually get /some/ kind of reply (and a file opened and kept on them, particularly persistent corresponders and activist groups!)

@wim_v12e @Gargron @EU_Commission No doubt they have a Ph D candidate or 2 implementing a sentiment analysis monitoring tool, as govts do with other social media... but here they'd have to do it instance by instance.

@gargron lol you talk like Nate Silver @wim_v12e @EU_Commission