We need a 20MW 20,000-GPU-strong machine-learning supercomputer to build EU's planned digital twin of Earth

Computer scientists attempting to build computational replicas of Earth to tackle climate change and environmental disasters reckon they will need a 20MW supercomputer with 20,000 GPUs to run a full-scale simulation. Starting mid-2021, the boffins will embark upon a seven-to-ten-year mission to create and deploy Destination …

[Joe W]

We all do...

“The majority of the weather and climate community remains skeptical regarding the use of black-box deep-learning tools for predictions" and projections, and anything else.

You have a hard time to tell why things should work the way the "model" says. Qutation makrs because for me a model should be based on understanding things like physics or dynamics or the maths/stats on the appropriate scale (yes, you will need to treat eddies on different scales differently, and you should include non-deterministic trajectories caused by the time and space scale separation of the small scale turbulence on large scale weather patterns) - no, that is not easy. But expecting "AI" to come up with "explanations" is... well... suboptimal.

Plus the deep learning can only analyse (and likely predict / project) only stuff it has observed in the past. I guess that high enough quality data is probably not available for more than 50 years (with sufficiently high spatial coverage)...

[Tom 7]

Re: We all do...

Its a black box model. We use them in lots of things all the time. Very effectively. For some things you dont need explanations. IF AI can come up with a model that accurately models something accurately enough to match reality over a period of time then its probably going to be of value in prediction. While it would be nice to have an 'explanation' it may have found something it will take us a long while to understand.

[Jimmy2Cows]

Re: We all do...

If the outputs of this model will be use to inform decisions that could affect us all, damn right we need an explanation. If that takes time to understand, so be it. How can we have confidence in data we don't understand? "Computer says no" is not a valid response. Rushing to make decisions based on such is asking for trouble.

[Tom 7]

Re: We all do...

So if you have a model that is right 100% of the time and it says your going to flood London if you do this but you dont know why you would go ahead and do it anyway?

[jake]

Re: We all do...

When you have that model that is right 100% of the time, come back and we'll talk.

[Pascal Monett]

I disagree. This is Science. It needs to be justifiable at all levels.

If the only justification you have is "it's the AI wot said so", you have no justification.

[Headley_Grange]

My brother and I both have black-box models of the ballistics of a cricket ball that allow us to throw, catch and hit it.

I also understand Newton's laws of motion, ballistics and I can even get a high-level understanding of the maths and physics of spin into my head.

My brother is crap at maths and physics.

One of us is a much better cricketer than the other. Have a guess which one.

[iron]

I like your line of reasoning... I have a distant cousin who plays tennis professionally and I'm a programmer. One of us can understand the maths and physics behind tennis but couldn't win a point in any sport if his life depended on it and the other has won Wimbledon.

[jake]

But can you or your brother intuitively calculate ...

... how much rainfall there will be (to the hundredth of an inch) during the course of just one week for each city block in a little tiny California county, such as San Francisco?

Don't be too sad, neither can any computer known to man. Nor will this trillion Euro boondoggle. And yet they are claiming it will be able to model the climate of the entire Earth for decades or centuries in advance? Pull the other one, it's got bells on.

[Headley_Grange]

All Models Are Wrong

A clever bloke once said something like "All models are wrong. Some models are useful". It's possible to create a very useful model of, say, airflow across a F1 car's surface. It's inaccurate because it doesn't predict the motion or behaviour of individual atoms or particles at the quantum level, but it's useful because it predicts drag, downforce, brake and engine air intake flow rates, etc. - so it's accurate enough for its intended purpose. Indeed, it's so good that the F1 teams have CPU time limits placed on this modelling when they are developing the cars.

From a climate change perspective it doesn't matter if it rains on Thursday or Saturday - that's just weather. What matters is whether the amount of rain in x years time is going to be a lot less, a lot more or about the same. Do we need to build drains or reservoirs is the level of question that we need to answer.

So, if we can get better forecasting then........it still won't help because a lot of people won't believe there's a problem even when the "once a century floods" happen every year (as long as they are happening to someone else) and many of the ones that do believe in climate change will change their mind when they see how much it will cost them personally to fix either the cause or the result.

[DJO]

Re: But can you or your brother intuitively calculate ...

how much rainfall there will be (to the hundredth of an inch) during the course of just one week...

Please spend a little effort in learning the difference between "climate" and "weather".

[[e-mail]]

Re: We all do...

I'm with Jimmy2cows on this - "computer says no" is not good enough.

If you want me to change my lifestyle for a valid reason then I have no problem, but if you want me to give up all the comforts of modern life on the grounds that some dodgy computer simulation tells you to tell me to, you can think again. Just because your computer is more expensive and more powerful (and far more damaging to the environment than anything I own!) does not make it right.

When they can take the readings from "Day One" and load them into the model, then run their simulation and get *exactly the same* results as The Real World experienced a week later, the results might be considered valid. But unless the two results match exactly, the model is not sufficiently accurate to predict the future. And no fudging for the accuracy of the detectors either - according to the models being bandied around in the late 70s and early 80s, the ice caps should be gone and we should all be swimming right about... ten years ago.

[jake]

Re: We all do...

"For some things you dont need explanations."

I think we ALL need a better explanation for the need to spend one point two trillion dollars worth of the public's money. Near as I can tell, right now it's just an excuse to over-pay a bunch of somewhat sketchy "scientists" and their management who are making a living sucking at today's public teat.

[Cuddles]

Re: We all do...

"For some things you dont need explanations."

I completely agree. To take an example already given in this thread, if you want to catch a cricket ball, all that matters is that you system is able to catch said ball, and any explanation which would ultimately involve relativity and quantum physics is completely unnecessary. For other problems, an explanation may be highly desirable, and in some cases the explanation is itself the entire point. With climate, for example, predicting what will happen is useful, but understanding exactly why it will happen and what could be done to change that is arguably more important. A black box might tell you how high you need to build a dike now, but understanding the details could allow you to adjust your behaviour so you don't need to build the next dike at all.

The thing about science is that it's all about understanding. A black box might give you good answers now, but you don't know under what circumstances the answers will stop being correct, and you don't learn where the answers came from in order to develop better tools to give more and better answers in the future. If all you want is an industrial system to give you an answer here and now, a black box is fine. If you're trying to improve your understanding of things, a black box is usually just a dead end.

[The Man Who Fell To Earth]

Re: We all do...

Tom Gauld hit the nail on the head when it comes to AI:

https://www.newscientist.com/article/0-tom-gaulds-attempts-to-create-a-sarcastic-ai-are-really-genius/

[jake]

Re: We all do...

"I guess that high enough quality data is probably not available for more than 50 years (with sufficiently high spatial coverage)..."

I'd guess maybe 20 years. Probably no more than 10. If that.

In other words, we only have that data for a mere crackle on a blip of geological time. Which is to say that it's pretty much meaningless in the great scheme of things.

[sreynolds]

It's not enough captain.

They should build an even bigger machine with even higher power consumption and then simulate what the effect on climate would have been if they didn't waste energy on the simulation. Much more useful data to be cleaned from there.

[The Man Who Fell To Earth]

Re: It's not enough captain.

Another Tom Gauld probably applies.

https://www.newscientist.com/article/0-tom-gauld-on-the-shrinking-size-of-powerful-computers/

[Stumpy]

Who needs this?

The answer is 42. We already know this.

[Brian Scott]

Re: Who needs this?

Ah, but what is the question?

[Anonymous Coward]

Re: Who needs this?

That's what the supercomputer is for!

[Headley_Grange]

Question

Question - "will I be able to put the washing out tomorrow?"

[jake]

Re: Question

"Question - "will I be able to put the washing out tomorrow?""

My Grandmother taught me the fine art of answering that question. Can even go three days out with generally accurate results. She learned it from her grandmother. Likewise, I leaned from Grandad how to tell if the weather was likely to kill me fishing in the Pacific off California's North Coast over the next couple days. Even the TV weatherfolks, with all their fancy computer models and graphics, can't get it right much past 5 days, though. And they will admit that the fifth day is mostly luck, if you corner them at a party and press them for answers.

[9Rune5]

Re: Who needs this?

I was going to post something along the same lines, but just want to add that I'd love to see one of these rigs produce schematics for an even bigger supercomputer that will be able to compute the question.

[Tom 38]

Re: Schematics for an even bigger supercomputer that will be able to compute the question

DestinE's Child?

[TimMaher]

Re: Who needs this?

The white mice want it. Do as you are told.

[Chris G]

I can see where this is going

By the early '30s DestinE will have become a singularity and will begin remedying the prime causes that are threatening the Earth's climate and ecology.

[Timbo]

Re: I can see where this is going

"By the early '30s DestinE will have become a singularity and will begin remedying the prime causes that are threatening the Earth's climate and ecology."

So, another name could be "Skynet" or maybe "The Matrix" ?? (Both had the same idea of essentially getting rid of humans so that a machine future was inevitable).

"Agent Smith:

I'd like to share a revelation that I've had during my time here. It came to me when I tried to classify your species and I realized that you're not actually mammals. Every mammal on this planet instinctively develops a natural equilibrium with the surrounding environment but you humans do not. You move to an area and you multiply and multiply until every natural resource is consumed and the only way you can survive is to spread to another area. There is another organism on this planet that follows the same pattern. Do you know what it is? A virus. Human beings are a disease, a cancer of this planet. You're a plague and we are the cure."

[KarMann]

Re: I can see where this is going

Missing icon is missing.

[jake]

Re: I can see where this is going

Yep. It'll pull it's own plug.

[Binraider]

Someone needs more FPS in Quake.

[Sgt_Oddball]

But will it..

Run Crysis or cause a crisis?

Mines the one that looks like an armoured gimp suit...

[Peter2]

If you are planning a two-​metre high dike in The Netherlands, for example, I can run through the data in my digital twin and check whether the dike will in all likelihood still protect against expected extreme events in 2050," said Peter Bauer, deputy director for Research at the European Centre for Medium-​Range Weather Forecasts (ECMWF) and co-​initiator of Destination Earth initiative

And for what one imagines would be likely to use far less expended energy and be significantly more practical use, one could just spend the extra time with a JCB and pike an extra meter or two on the dike if you had any concerns about what it was protecting getting wet.

[Chris G]

One could buy a lot of JCBs for the price of a 20MW super computer.

[rg287]

And for what one imagines would be likely to use far less expended energy and be significantly more practical use, one could just spend the extra time with a JCB and pike an extra meter or two on the dike if you had any concerns about what it was protecting getting wet.

Of course if your "2metre dyke" is a €100m dyke programme surrounding a significant area, it's probably cheaper and more efficient to bid for a couple of days on DestinE to check your calculations than to "just build it" and find it doesn't work because of a fundamental error in your design.

Muckshifting and civils are expensive. Increasing a 2metre dyke to 3metres doesn't increase material use by 50%, you're actually getting closer to double the material as the footprint gets wider (a problem if you're building up to a river or a housing development as your dyke is now in the water or inside people's kitchens).

[sgp]

I say let them drown, living beneath the sea level and all that!

[Peter2]

Of course if your "2metre dyke" is a €100m dyke programme surrounding a significant area, it's probably cheaper and more efficient to bid for a couple of days on DestinE to check your calculations than to "just build it" and find it doesn't work because of a fundamental error in your design.

It would certainly be cheaper.

However going by the $70 billion damages caused by the flooding at New Orleans then i'd suggest that it would be safer to estimate the worst possible case (from previous historical patterns such as the once in two centuries 1953 storm) and then build taller "just in case" with a generous margins for error, subsidence and sea levels rising. And then with a bit more added in for paranoia's sake since better safe than squidgy.

In almost all cases I would expect that it's going to be cheaper to build the dyke higher than repair the damage and I doubt that the results of a computer program are going to come with an indemnity policy covering the potential damage, or even a guarantee that the people who recommended the wrong level would admit fault.

[iron]

And how does one calculate these "generous margins for error, subsidence and sea levels rising"?

Perhaps using some super computational machine?

[jake]

Easier answer.

Don't build habitation in flood plains in the first place. We have modern transportation, we don't have to live in the fields we grow our food in anymore.

And then switch to growing food that likes it in flood plains. Simples.

[sgp]

Re: Easier answer.

Do you have any idea how many people live in flood plains in the Netherlands? "Simples"

[rg287]

20MW

20MW is a shedload of power. It's the equivalent draw of a small town with 25k homes.

Equally though, the new GE Haliade-X turbines provide 12-14MW each with a 60-64% capacity factor. So if they sponsor one and a half turbines, that'll largely cover them (with some hydro as backup for wind drops). The new Dogger Bank farm is installing hundreds of them. Magnificent bits of engineering (albeit they need to work out how to recycle the things at EOL).

[Roland6]

Re: 20MW

Surely somewhere like the Dogger Bank is the right location for all of this: all that waste heat could be used for a fish farm.

[andy k O'Croydon]

Re: 20MW

It sounds like the timeline is similar to when the US and Canada are looking to start deploying micro-reactors (not sure on the Russian and Chinese timelines). So a couple of these may also do the trick (typically these are specced around 10MWth, so around 4MWe). The spare heat capacity could also, potentially, drive heat engines to drive the cooling systems, although that's probably looking a bit further forward.

[adam 40]

Re: 20MW

Make it run on the turbines ONLY, so it will be net zero.

If the wind doesn't blow, it doesn't run.

[jake]

Re: 20MW

I suspect the state of Texas will be quite happy to sell them some slightly used turbines for pennies on the dollar. As is, and where is, of course.

[Natalie Gritpants Jr]

1kW per GPU

Seems very high, 100W would be more believable.

[Sgt_Oddball]

Re: 1kW per GPU

Considering the RTX 3090 graphics card can chugg 350 Watts by itself and an AMD Epyc CPU can slurp down 240 Watts..and that's before you look at ancillaries like drives, networking, memory, motherboards etc etc.

Having a 1kW isn't that outrageous considering cooling requirements to stop this lot making the mother of all clouds of magic smoke would be needed as well.

[Anonymous Coward]

Re: 1kW per GPU

Not with the times are you, 250w+ each.

Then the servers to put them in, storage, cooling, heating for the water bags, etc.

[Jimmy2Cows]

Re: 1kW per GPU

Not just the GPUs. The rest of the system needs power, not to mention the vast cooling requirements to stop the whole thing melting into a giant pool of glass and plastic.

[Posted before a screen refresh, hence repeating what others have already said. Note to self: F5 before posting]

[just another employee]

A small ask, but...

...does this mean they may also be a bit better at forecasting tomorrows weather ?

[Chris G]

Re: A small ask, but...

No! But after any 'Natural Disaster' they will be able to tell you exactly why it happened, it's called an expert system.

[MachDiamond]

Re: A small ask, but...

"No! But after any 'Natural Disaster' they will be able to tell you exactly why it happened"

A year later.

[Auntie Dickspray]

Re: A small ask, but...

No but they'll be able to predict the weather in 150 years with 100% certainty.

[MajDom]

If you're worried about power consumption, concentrate on cryptocurrency mining. We're talking terawatts. And utterly useless ones at that.

[Arty Effem]

Sounds like a Year 9 programming test. Time allowed 90 minutes.

[MarkET]

Digital twin

Welcome to the Matrix

[jdiebdhidbsusbvwbsidnsoskebid]

Model validation?

How are they going to validate that climate model? There's been a huge amount of climate research and modelling over the last decades but there's still a significant amount of discrepancy between different models. And what they are planning to do with DestinE seems like an order of magnitude more complex than what current climate models achieve.

As well as the extra complexity, the outputs of this model look they will inform immediate and significant decision making about large scale infrastructure. That cranks up the level of required control even more!

[Chris G]

Re: Model validation?

Absolutely, they need to avoid GIGO as much as possible and the bigger the project the more difficult it is to ensure that all that goes in is valid.

Plus I suspect there may be a number of important factors that we are not even awate of yet.

An additional concern is if addressing climate change is achievable, what is the default that is being aimef for and is the correct one or even desirable considering that our ecology is an evolving system?

[Jellied Eel]

Re: Model validation?

How are they going to validate that climate model? There's been a huge amount of climate research and modelling over the last decades but there's still a significant amount of discrepancy between different models.

Usual way is to test SimEarth against reality. So climate models can be initialised to the past, then run and compared to say, the last decade's actual weather. So a combination of 'hindcasting' to predict the past, or reanalysis to compare past model predictions against reality.

But there's an interesting video on global warming simulations here-

https://www.youtube.com/watch?v=yto4Sx-yhFY

Err.. I mean here-

https://www.youtube.com/watch?v=S-XrpTWxoOw

With Vicky Pope giving a Royal Institue presentation. She has.. interesting views wrt climate*, but the RI channel's worth a sub because they have all sorts of interesting lectures.

And what they are planning to do with DestinE seems like an order of magnitude more complex than what current climate models achieve.

Yup, hence wanting a supercomputer an order of magnitude larger than a lot of current climate models. Which by their very nature are very crude approximations. It's 'easier' to do UK MetOffice or ECMWF forecasting because they're either focusing on a small geography, or a short time frame, or both. So ECMWF might forecast out to 10 days, with probability narrowing as time passes and the forecasts are pretty good.

Climate models then try to make ECMWF-style forecasts decades or a hundred years out. So their Sim-Earth is broken down into large grids, eg-

https://en.wikipedia.org/wiki/General_circulation_model#Grid

Typical AGCM resolutions are between 1 and 5 degrees in latitude or longitude: HadCM3, for example, uses 3.75 in longitude and 2.5 degrees in latitude, giving a grid of 96 by 73 points (96 x 72 for some variables); and has 19 vertical levels. This results in approximately 500,000 "basic" variables, since each grid point has four variables (u,v, T, Q), though a full count would give more (clouds; soil levels). HadGEM1 uses a grid of 1.875 degrees in longitude and 1.25 in latitude in the atmosphere; HiGEM, a high-resolution variant, uses 1.25 x 0.83 degrees respectively.

So very low resolution, and not many variables simulated. So having a more complex model with say, a 1km grid with accurate terrain modelling should allow more complex (ie more realistic) climate simulations. But still not enough to simulate what might happen to wind-driven weather/climate effects, if you raised a 3m high dyke around a square kilometer of land.

*As in she's a true believer, and the Hadley models run hotter than reality. Wonder if that's because they exagerate the climate sensitivity of the atmosphere wrt CO2? In the RI vid, at around 18mins, you can see she's totally convinced that corrrelation = causation.

[jake]

Re: Model validation?

"at around 18mins, you can see she's totally convinced that corrrelation = causation."

There is a lot of that in so-called "climate science".

[Mike 137]

"Computer scientists will be able to tweak the digital twins to assess hypothetical scenarios..."

They will be able to assess hypothetical scenarios, but whether these accord with reality will depend entirely on the validity of the (huge) code base on the system. Unfortunately, software on this scale is likely to be so complex as to be hard to verify.

As to the irony, some decades back I worked on a very small physical simulation facility that engineered closed climates for ecological research including the effects of global warming. Running it pumped out over 70% of around an average 60 kW electrical load into the atmosphere as heat and its water purification wasted some 90% of the water it used. So simulation is in principle a huge improvement, provided it can be verified to accord with reality, although 20MW is admittedly a little on the high side.

I also have a concern about maintenance. A big limitation of early discrete component computer hardware was component failure due to the vast number of points of potential failure. 20,000 GPUs would seem likely to approach a similar problem.

[six_tymes]

cause it. sooner than netflix and other streaming services.

[Androgynous Cupboard]

Not again

Unfortunately the EU has form on spaffing huge amounts of money on IT projects that go nowhere: the human brain project had the same goals for the inside of yer bonse.

In case you think this is a brexity dig at the EU, it's not at all. In fact given how much money our own govt waste on IT that underdelivers, in some ways it's reassuring to know we didn't depart a flawless, sunlit uplands of rational decision making for the wet puddle of mismanagement we have now.

[John Sturdy]

Re: Not again

It does sound to me like there may be an element of what I believe is referred to in the USA as "pork".

[jake]

Re: Not again

Pork is spending large sums of money at the federal level for projects that only affect a small, local community, not the US as a whole. (See Senator Robert Byrd's "Federal" highway system and many other, similar projects in West Virginia for one of the canonical examples).

The word you are looking for is Graft.

[fajensen]

Re: Not again

Well, if government doesn't spend the currency it collects in taxes, then why (and how) would we pay taxes?

[Rich 2]

A bit late, don't you think?

"Starting mid-2021, the boffins will embark upon a seven-to-ten-year mission to create..."

I hate to point out the flaw in this plan, but we're fucked right now. I think in 7 to 10 years, we'll be well and truly fucked to the point of no return (if we've not already passed that point).

I suppose they can use it retrospectively to simulate how we managed to destroy our planet. If anyone is still alive to give a shit by then, that is.

[Swiss Anton]

Re: A bit late, don't you think?

An "expert" on the radio (*) reckons we need to build 1 nuclear power plant, or a 2500 turbine wind farm, every day for the next 30 years to replace all of the fossil fuelled power stations in the world by 2050. He didn't say what the generating capacity of these units was, so I decided to do my own estimate based on a 800MW nuclear plant (like Hinckely Point B), or a 80x10MW turbines wind farm, and I came up with a more optimistic estimate of one of these every 4 days. My estimate based on current electricity demands and does not take into account the increased demand of electric cars and of replacing gas central heating systems.

So in short, I agree, we're fucked.

(* The radio show is a BBC world service programme How soon can we go carbon zero? )

[jake]

Re: A bit late, don't you think?

I predict that in 100 years we'll still be arguing about the climate in forums like this, with virtually no change from today in how humans as a whole go about their day to day business.

[fajensen]

Re: A bit late, don't you think?

I predict that in 100 years time, when the water is standing 30 cm above Christiansborg Slotsplads, Copenhagen, some wankers will say that climate models totally failed and it is all a scam because, as everyone can see with their, the water didn't get to the 80 cm that the climate conspiracy claimed!

Inside parliament they will be ranting about "Muslims" and refugees, especially the latter ones becoming numerous these days, their weight pressing down the land!

And in that way the day to day business of humans will have virtually no change.

[Schultz]

"run on more renewable energy sources:"

They really should adjust our whole electricity network to let us run our Stuff on more renewable energy sources. E.g., insert the Euro-plug right-side up to get the renewables or left-side up to destroy the world. No reason why the researchers should get to choose their energy source and we can't.

Let me translate that statement: What this researcher wanted to say is: "I know it's ridiculous to burn this kind of energy for a badly defined research project, but see if I give a rat'* a** about it".

Kind-of sums up humanities' current approach to climate change in general. Sell me a fancy new car with an "eco" sticker and we're good. Right?

[MachDiamond]

Re: "run on more renewable energy sources:"

"They really should adjust our whole electricity network to let us run our Stuff on more renewable energy sources"

It's not the inputs, it's the rigid usage patterns. You start at 8 or 9 am and you leave at 5 or 6 pm from Monday to Friday. Certain days are bank holidays that don't change.

I equate what needs to be done as being very similar to buying fruits and veg in season. The problem is accountant's heads would explode if they had to statistically forecast factory outputs if the factories were powered on when the sun was shining and there were enough wind. Some processes can be run that way. Ammonia manufacturing is a good example. Ammonia is a big feedstock and accounts for lots of power usage. A Haber-Bosch process set up in a shipping container sat right next to a wind turbine could do it's thing when the wind was blowing and sit idle when it wasn't since ammonia is sold in batches rather than as a continuous feed in most places.

[Binraider]

Re: "run on more renewable energy sources:"

The electricity network doesn't care what your source of generation is. What it does care about is the ability to adjust itself to stay at 50 or 60Hz as per your local networks standards dictate, in response to short term fluctuations like 20M people putting the kettle on at the close of a world cup final. Renewables fundamentally have less control over output compared to a gas turbine, a nuke, or a coal plant that have the advantage of physical inertia in the generator to ride out short term fluctuations. To facilitate mass renewables to fully replace heavy spinning generators, we need capable storage systems. Dinorwig is hugely capable, but also 40yr old and priced accordingly to it's mostly unique capabily in England & Wales. Smaller scale pumped storage such as that touted by Rhenergise or some of the battery ideas knocking about are key enabling techs to get unhooked from heavy spinning plant. The investment is slowly coming, but we do need more. I remind people of the events of August 2019 that lead to disconnection of demand due to a short term shortfall of generation when two big generators shut down within seconds of each other.

[fajensen]

Re: "run on more renewable energy sources:"

They really should adjust our whole electricity network to let us run our Stuff on more renewable energy sources

"They" are already doing it, on a big scale; Undistracted, I might add, by the many, many, couch-table "experts" who still think the electrical grid works, and indeed should continue to work, like it did back in the 1950's. The good times, where arabs and wimmen knew their proper place and dad could bring home a living wage and nukes made electricity too cheap to meter.

"Energy Efficiency in Science" is A Thing. The money that goes to electricity comes out of the money that researches can use for new kit and more minions, all of which means less research and fewer citations for them. They very much want to keep the power bill as low as possible: https://home.cern/news/news/knowledge-sharing/energy-sustainable-science-0

[Binraider]

There is the not so minor difficulty that they actually have to get the GPU's. You know, stock in a warehouse somewhere. Rather than the vapourware we've had for the last 18 months.

[jake]

So that's why ...

... the shelves at Fry's were always empty!

[FlamingDeath]

“ Computer scientists attempting to build computational replicas of Earth to tackle climate change and environmental disasters reckon they will need a 20MW supercomputer with 20,000 GPUs to run a full-scale simulation”

Well they’re not bloody computer scientists are they

Give them a 286 and don't open the door until they figure it out.

If you give them a super computer, they will just waste resources with inefficient bollox

[MachDiamond]

"Give them a 286 and don't open the door until they figure it out."

Way back when the USSR existed, I knew some engineers that were using tiny PCs to do acoustic analysis. The code was fantastically compact and efficient. They didn't have access to anything very powerful as the military laid claim to anything actually useful in terms of computers. The output was just tables of numbers, but it saved lots of punching buttons on a hand calculator.

[Binraider]

The gas simulation software I worked with for a few years had its origins in Communist Europe. The algorithm behind it could simulate large and complex networks with minimal CPU and RAM requirements, provided you had the patience to write and interpret results from slower storage. A German company picked up on the algorithms efficiency post collapse of the Soviet Union and has spent the last 30 years building a modern UI around it. It's still the best software of its type, because of the algorithm, not because of the UI. The problem in climate space is of course, that such a large system cannot be meaningfully distilled down to an elegant algorithm without losing purpose. When computers need 20MW though, I'm increasingly of the opinion you are probably better off investing a bit more safety margin on your construction than throwing more horsepower at trying to optimise a fundamentally impossible to divine result.

[fajensen]

I'd say that it is probably better to blow puny 20 MW on science rather than wasting it on crypto coin mining or serving up SoMe swill to the masses.

FaceBook is currently running at 2 GW, factor 100 more than 20 MW, but, it ain't "Climate", so no reason to question the utility of that?!

[Binraider]

Somewhat stating the obvious, there absolutely is utility in obtaining power efficiency on any large or distributed compute capability. Consider Facebook - a "simple" application to display text and a few photos manages to chew obscene amounts of memory client side.

Google ploughed through 10.6 TWh in 2018. No matter who you are, that's an awful lot of kettles.

Going back to my original thought; burning 20MW chasing an precise answer that is unknown by design, is an exercise in futility. What is not futile, is pursuing energy efficiency measures and reducing consumption. Consider that every extra demand connected to the network today, means turning a fossil-burner back on to cover until a windmill can catch up to that level of demand.

[Falmari]

DSS

This is a DSS (Decision Support System) well that’s the name that was used in the late 1990s. I worked on one for UK MAFF to predict nitrate leaching into the soil. It’s a sort of what if I do this what will happen. The system would feed the data into various computer models producing maps.

But the results more of a rough guide in the end. The models though fairly good on past data not that good when you are guessing the future data. The weather for one affects the models, 10 years of wet summers will have a different level of nitrate to 10 years of dry summers and that’s just some of the data. So, it was more of a best possible to worst possible scenario, if we did this.

Funny enough that project was because of EU insistence. As a programmer it was a great thing to work on getting to use lots of interesting stuff.

[Korev]

Heat?

The irony of building an energy-intensive system to tackle climate change was not lost on the researchers. They noted in their paper that the future super should be built at a location where its nodes can run on more renewable energy sources

It'd also be good if the ~20MW of heat could be used for heating or other purposes

[JassMan]

Re: Heat?

I believe there is a power company which uses solar heated Stirling engines driving a generator to make green energy. Throw away the solar concentrator and connect the Stirling motors to the heatsinks on the GPUs and sort of get perpetual motion. OK, entropy and all that, but they could probably reduce the overall input requirements. Do it all near the sea and the Stirling motors would be very efficient.

[MachDiamond]

Re: Heat?

"Throw away the solar concentrator and connect the Stirling motors to the heatsinks on the GPUs and sort of get perpetual motion."

The delta T isn't high enough to extract enough power from a set up like that. By the time you let the GPUs get hot enough that it would work, they'd be slag. You could keep a pretty good sized greenhouse growing veg all winter, though.

[jake]

Re: Heat?

"You could keep a pretty good sized greenhouse growing veg all winter, though."

That can easily be done with a so-called "earth battery". Works by blowing air through underground tubes. Stores heat in the summer, releases it in the winter. Maintains proper humidity, too. One of the greenhouses I help run is 300 by 400 feet (2.75 acres, 1.1 hectares near enough). The "climate" is maintained by a couple of small blowers (under one tenth of a horsepower total) that run 24/7. On very hot days, roof and floor level vents open automatically, using wax filled actuators (similar to the thermostat in your car). The battery that powers this is an 8D, which is kept charged with a 1kW solar panel that is overkill for the purpose.

[Chris G]

Re: Heat?

There is a bigger argument for developing solid state generation using the Seebeck effect, it has relatively low efficiency but is far less complcated than Stirling engines and requires little maintenance in comparison. There is an interesting video by Tech Ingredients on YouTube.

[DS999]

More power

They'll want to simulate it better and better, which will require more power. In the end the model will determine that climate change would have been reversible if not for all the gigawatts spent on running climate change models over the years.

[JassMan]

WTF?

No mention of a terrapixel or even petapixel display - indeed no mention of display at all - so why do they need GPUs. Presumably they chose GPUs for the peer-peer interconnects so why aren't they using NVidia CryptoMiningProcessors instead. These are basically a GPU but without the video I/O and all the logic that goes with it. Less power hungry and with even more maths processing.

[MachDiamond]

Below every deep another deep unfolds

What tends to happen with these sorts of projects is the discovery of more important variables than have been considered. It's turtles all the way down without end. Seemingly, the universe doesn't want this sort of thing to happen.

[jake]

Re: Below every deep another deep unfolds

Yes, the Universe is quite fractal. So is my cat.

_{Or was that fractious?}

[PeterM42]

Has DestinE Crashed?

No - the wind dropped, so the turbines stopped. We didn't predict that.

D'oH!

[conel]

Hubris

They want to centralise decision making on what crops to put in what fields!

In engineering when working on things as mundane as aircraft strutures testing/ benchmarking of the simulation results is necessary. Someting that is impossible with something as gradiose as an earth digital twin.

The notion that they can simulate the entire planet and make better decisions than specialists - quite literally working on the ground in some cases - is simply science fiction.

[Dale 3]

Douglas Adams

I demand to know where the obligatory Douglas Adams reference is in this article?!