This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Data centers are far, far more complicated than setting up a desktop and WiFi in your home network times a million. To give a taste of the problems:
Power. They're incredibly power hungry. You can't just hook them up to the local grid; the local grid might not even have the capacity to support them. You've got to expand existing plants or build entirely new ones. And then how do you handle power surges? A naive approach is going to cost you tens of millions of dollars when a spike comes through and kills a thousand GPUs. Of course, you've also got to figure out how to distribute the power internally and plan for inevitable component failure at multiple layers. How do you distribute thousands of amps (safely and without melting the insulation)? And what about when the grid fails? You've got to have backup power sources ready for a day or two of unavailability. Not doing these things will make your data center uneconomic, as all that expensive capex is sitting around unused.
Heat. All this power has to go somewhere. Air conditioning and fans don't work at this scale, so you've got to use liquid cooling. But where do you get this liquid and where do you send it? And how do you pump it through hundreds of miles of plumbing? How do you minimize the rate of pipes getting clogged, and how do you handle it when pipes do get clogged? And remember: if this plumbing fails, your GPUs are going to rapidly start failing as well, to say nothing of the risk of a highly corrosive liquid being sprayed all over billions of dollars of investments.
Weight. A rack, by itself, weighs over five tons. Now add all of the equipment and liquid needed to handle 1) and 2). You've got to have a massive foundation that can support that. This isn't just a big Amazon warehouse.
Security. You've got a lot of investment here, all in one place, and quite delicate. How do you prevent a hostile actor from taking a truck or drone and destroying your investment? To say nothing of state-level actors, who absolutely are trying to break in.
Networking. You've got thousands of very chatty, data dependent GPUs coordinating in a highly choreographed dance to transmit trillions of parameters to each other. And the slowest link determines, by itself, the overall speed. When you've got thousands of nodes, that is pretty slow. Your mega ultra gaming WiFi 7 ASUS gaming router is going to have some trouble here. And that's just internally: you're receiving and transmitting massive amounts of data to the outside world. How do you prepare for a backhoe running through, or a shark chewing through, one of your fiber optic cables? Every hundred miles or so, you also need to amplify the signal, which comes with its own power, security, etc requirements. And what about truly bulk data: if you're transmitting 100Pb of data, is it better to saturate your measly 1 Pbps capacity (displacing your other network needs) for a painful amount of time, or to use trucks with hard drives to ship it cross country? And, if you have a bunch of these trucks, how do you efficiently unload them, without causing needless congestion or buffering in the physical world?
There's maybe half a dozen organizations in the world that can handle all these concerns. All of them are American.
Nobody is comparing datacenters to home networks. I don't think your points are particularly correct regarding the extraordinariness of datacenter construction, either, even though you (or your LLM? I'm getting a certain vibe) evidently put a lot of effort into them.
1/2: Bounded above by the requirements of power plants, which lots of countries build. Electric arc furnace complexes are also measured in the same hundreds of megawatts as datacenters. Also, my examples about bitcoin farms are pretty relevant here.
3: Moving heavy things on the order of "five tons" is really not impressive for modern civilisations.
4: Same for securing centralised facilities. How often do first-world countries experience break-ins in their bank vaults or even prisons?
5: This is of course a nontrivial engineering problem, but off-the-shelf solutions exist. Also, China is filtering all their incoming and outgoing traffic for content basically in real time, which is surely a harder problem.
Your conclusion is also wrong. I searched for maybe about 2 minutes to find that Meta's Prineville data center is (was at some point?) considered the biggest in the US, and it's cited as using 15000 GWh in a year. Elsewhere, Alibaba's Zhangbei data center is cited as pulling 150MW, which x 24h x 365 gives 1314GWh which is only one OOM off. Other big US data centers seem to also be moving in the 100-200MW range.
Your LLM detector is busted, I'm afraid.
More options
Context Copy link
More options
Context Copy link
It's actually worse than that: you will fuck up the grid if you just naiively run jobs that cause your hundred thousand node cluster to all spike their draw in unison.
We have to smooth our consumption because otherwise it messes with generating equipment.
More options
Context Copy link
Okay you've aligned all your ducks in a row at massive expense which even that half dozen companies are more 'capable of getting financed' than outright affording and then it's being used for something that isn't even profitable at current scale.
That's a orthogonal concern as to whether it's a big infrastructure project. Maybe they're going to be giant boondoggles; maybe not. The same applies to most of China's infrastructure projects; it still represents a high level of state/social capacity.
And the biggest builder of these projects (Google) usually funds them out of cash on hand. To the extent it issues bonds for them, it's for financial engineering/tax reasons. And investors for whatever reason are desperate to buy "green bonds," and are willing to take spectacularly low rates for the chance to buy them.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link