OK, I still don’t get it. The ones we’re talking about are brand new data centers with brand new cooling systems. If the local and state government get their act together and require clean pipes, why couldn’t the water come into the data center, cool the equipment by running through the brand new clean pipes, come back out the other end hotter, but not really dirtier, and go back into the system?
Or, do that, and put a mini filtration and cleaning system at the other end. There are literally trillions of dollars getting invested into these things – I imagine the marginal cost of cleaning up the water would pale in comparison to the other costs, and would stop a lot of the push-back from the communities.
I have a related question, looking through the wrong end of the telescope:
Is “hot computing” technically possible? In other words, running immense servers of some kind that don’t need (much) heat released? The components could perhaps heat up to extremely high temperatures and still function normally. The heat would have to go somewhere at some point, but maybe it can be harnessed to turn turbines or something?
Or maybe the opposite case – could “cold computing” be a thing? That is, immense servers of some kind that don’t really heat up very much? Probably not a great analogy, but I think of how incandescent light bulbs got too hot to touch when lit for a minute or two whereas modern LED lights (at least in household-use sizes) never get too hot to touch. Has there been research into low-heat “cold” computing?
I’m assuming current systems are considered optimized as much as possible given both technical and economic constraints. Still, perhaps a better mousetrap is in the offing?
That can be done, but you are limiting your cooling by requiring the water to stay as water. You can get more cooling (and cheaper) by using the heat to turn the water into steam.
You don’t need to add fancy shmancy filtration and whatever. Instead of running it open loop, run it closed loop instead. You just need to add the equipment to condense the steam back into water. Then you run that water back through your cooling system and the cycle repeats.
You’ll need a bunch of water to fill the system up once, but once it has been filled, you only need to add small amounts of water (because even in a closed system some water will still manage to escape).
It’s already done in some places.
Your proposed method actually causes more environmental damage simply because the water does come out hotter. That extra heat is unhealthy for plants and animals that evolved for lower temperatures. Some plants and animals can adapt, some can’t.
You can solve that problem too, just by adding cooling ponds to let the heat escape into the atmosphere and then release the water back into the environment when it’s closer to normal temperature. But now you’ve added a bunch of extra real estate costs for the ponds to the cost of your data center.
There are a lot of ways to solve this, but they all cost money.
I think the GPUs used in these data centers are as heat and energy efficient as they can be, and probably operate on the edge of melting down even with all the cooling. They would love to use less energy so they could either pack more in closer together or save on energy costs.
You can make a computer run nice and cool. The problem is that it will be very slow and not so useful by modern standards, pretty much the opposite of what you need for a data center.
One of the big limitations of modern processors is heat, so there is a HUGE amount of effort going into making computer chips run cooler.
We’d need to find an alternative to semiconductors.
They operate more efficiently when cooler, and this includes LED lights. They still heat up, just not as much as an incandescent light bulb. The heat is a by-product of the operation. A good deal of that energy will be converted to heat.
Semiconductors will also continue operating when warmer, albeit less efficiently. But run hot enough and some component is going to fail or melt. Whether or not that’s the compute-y bits, they still rely on the other components, too.
That won’t work. The demands on the water main vary. The demands of the data center do not. Without any sort of buffering, the amount of water flowing through the data center will be too variable to be practical. Your cooling will be limited to the minimum flow rate of the water main, which is quite a bit lower than the maximum flow rate of the water main.
You can dump it back into the water system by pumping it up into a water tower where it will end up back in the water main eventually, but straight into the water main without any sort of buffering isn’t going to work.
If you want to use and release the water, there are ways of making it work, but that’s not one of them.
Ah, so something on the order of a technological revolution would be in order. Not just tweaking current tech.
I wonder if data centers could be physically distributed so that you turn one huge heat sink into many small ones, as opposed to “packed in closer together”? Or maybe structured differently – build the data centers up instead of out. Thinking huge Dubai-Tower sized buildings that are basically “smokestacks” full of servers along the inner walls. The servers would be in “rings” along the inner walls of the tower, with waste heat being vented upwards through an open center. The rings would be spaced apart to help avoid accumulation of too much waste heat in one area.
I understand that direct physical maintenance of the servers in a giant smokestack would be complicated greatly by potentially having a human go a few thousand feet in the air to service the devices. I’d bet contemporary robotics could handle 90+% of that kind of work (go up, pull a server down, bring it to ground level for maintenance, replace server up high when done), but maybe not cheap enough?
Actually, no. If they ran that hot they would fail rather quickly. Heat kills semiconductors. You can make them more robust when you design the chip, but at some point you have to make a tradeoff between longer life and the amount of heat the generate. So, rather than being close to melting, they are instead run as hot as they can so that they’ll have an expected life that is long enough for the data center’s use.
They are typically designed to last 5 to 8 years. If you didn’t push them so hard they would run cooler and last longer, but the emphasis is on performance, not on longevity.
If you ran them close to melting, they’d probably only last a few days or maybe a few weeks at best.
Given the pace of improvement, it makes total sense to focus on performance over longevity. In 8 years, these will likely be totally out of date for these purposes (but would still be able to play Crysis…).
To me, the question is: is it possible to recover energy from the heated the water to do useful work? Think of a power plant that uses compute cycles to heat the water instead of burning coal. Probably an economic question rather than engineering.
If it helps at all … it wouldn’t have to be anywhere near as well-appointed as a luxury property such as the Dubai Tower. In fact, it might help to think of it as a giant, tube-shaped warehouse. And, optimally, perhaps somewhat uncomfortable for human habitation except perhaps on the levels near the ground.
It could be somewhat “squattier”, too, so long as the exterior is cylindrical to focus waste heat inwards and upwards. Maybe something akin to the world’s largest tandoor oven in shape?
It’s hard to beat evaporative cooling when it comes to efficiency and cost. My issue with these data centers is that they don’t need to and shouldn’t be hooking up to the municipal water supply in the first place. Build them by the ocean or the Great lakes, and use the raw water (with filtration, of course) for cooling. Evaporate it right back over the ocean/lake and it becomes part of the water cycle again. Stop building them in places where they have to hook up to the municipal water supply.
Technically, yes, but as you note it’s more of an economic question than an engineering question.
Power plants require significantly more steam than what a data center generates, so at best you’d get a small power plant that isn’t very economical to run.
On the other hand, there are other industrial uses for steam and it should be possible for another company to use the data center’s waste heat.
Technically this can be done, but it drives your maintenance costs up rather dramatically, especially with salty ocean water.
(Editing my own quote above to allow the pull quote stand alone)
Here’s a good comp to what I have in mind – start with the Vehicle Assembly Building (VAB) at NASA’s Kennedy Space Center. Build it cylinder-shaped (gallon-paint-can shaped, 55-gallon-drum shaped, etc.) instead of box-shaped, and then scale it vertically – instead of of topping out at 524 feet like the VAB, go vertically to 1000-1500 feet. Higher if feasible (though I understand this is already on the bleeding edge of the possible as a concept).
From a distance, it would probably look like an immense can of Red Bull.
First, semiconductors (usually) are crystal. Heat them up too much and you get a blob of liquid where the specific things (doping with small quantities of other elements) would ooze away and do nothing.
Computing uses power. Each time a transistor changes state, that takes energy. They make transistors microns wide now, so not a lot of energy - but then there are millions on a chip, so still - lots of energy. Then there’s thermodynamics. You are adding that heat while the computers run. It has to escape somewhere. Otherwise it just gets hotter and hotter until someting fails or melts down. Energy in has to equal energy out for a steady state. Haystacks in the sun and oily rags are known to sppontaneously combust, for example - bacteria in wet hay generates heat, which the outer layers insulate until the internal temperature reaches combustion. Same with oily rags. Oil in air will - very slowly - combine with air, same reaction as burning but much much slower, so less total heat. Like your haystack, combine oilly rags with a very high oil surface area and lots of trapped air, in a bundle big enough to insulate… and worse, the reaction speed up as things get hotter. Then you need to buy a new garage.
So the purpose of the water is to run through the building and carry away the heat to maintain an acceptable temperature. But the hot water then has to be cooled if you want to recycle it, or disposed of if not. Think how much heat your toaster throws with 1kW, then consider how much a 150MW data center is going to generate. That water is either evaporated or cooled or dumped.
You can’t just dump the output into the river or ocean - it would seriously disrupt the ecology. The environment is adapted to a particular local temperature, and making part of it tropical would disrupt things like fish breeding, especially if it’s a salmon run river. Evaporation is the simplest and most efficient, least ecologically disruptive, the wind carries away the water vapour. so you recycle the water in the data center through a “radiator” that heats up the disposable water then pour it down cooling towers, or simply use new water all the time. Either way, you need to replace the evaporated water.
Meanwhile, power generation has the same issue - like your car, you burn stuff to make power (unless you are lucky enough to have hydro power). A decent part of the heat generated creates steam to run generators (or to accelerate your car), but the output is not 100% - some still is left over as heat. that too has to be disposed of - same problem.
So AI gives you the double whammy - either they make their own power on site, even more water used - or the electrical utility company does that part for them.