chasd00

Very loud high speed fans

if you haven't heard a 5u server intended for a datacenter rack come to life it's quite the experience. Sounds like a plane taking off.

That's what they should do. Datacenters don't need to be in anyone's backyard.

at least in texas that's what they are doing, building them out in the middle of nowhere or close to the middle of nowhere. The issue that I keep seeing pop up is behind the meter power generation ( gas turbines ) is hitting roadblock after roadblock. There just aren't very many and GE is scaling up but i mean it's not like adding RAM in AWS, it's going to take half a decade to see the increased inventory to buy one let alone onsite and operational.

"...demand from utilities, data centers and other power users continues to stretch manufacturing schedules into the next decade."

https://www.compressortech2.com/news/ge-vernova-raises-gas-t...

some are resorting to using ocean liner engines to generate electricity.

"That naturally explains the recent order flow into large reciprocating engines. In April, the Finnish vessel engine manufacturer Wärtsilä Oyj Abp announced a 790 MW off-grid power solution for a new Texas data centre facility,"

https://globalalphacapital.cclgroup.com/insight/gacm-from-se...

large, new, far out, grid hookups usually get done in units of years, you don't just call the power company and order one. I think a location with an existing connection to the power they need probably takes priority over every other location decision.

idk, at least in DFW the biggest datacenters from the late 90s and early/mid 2000s are in the suburbs where everyone lives. There's some in and near downtown but there's not enough space or it's too expensive. The big datacenters were built where the giant neighborhood were built, places with lots of land for relatively low cost.

I think the biggest change in public sentiment in datacenters is a reaction to the extreme scale they're being deployed at these days. They're really vastly unlike datacenters of the past. The power densities are so much higher, they're so much larger, and they're getting vastly more environmental waivers and tax subsidies than ever before.

ok now this is something i understand and haven't thought about, thank you. I've visited the stargate construction site in Abilene TX since it's not too far from me (3hrs). The scale is indeed unlike any DC i've ever seen before.

edit: "...howling on-site power generation that got installed without any kind of air quality reviews because they're allegedly "temporary" .." you're talking about XAI's data centers in Memphis right? I don't think that's a fair comparison, that was knocked together in like 100 days. Those generators are on trailers.

where does the noise pollution come from? GPU fans? What's more noisy about a DC filled with CPUs and GPUs than a DC filled with just CPUs. I can imagine generators being noisy but that wouldn't be anything different than regular data centers. Also, so few DCs for the AI buildout are actually online where are the noise measurements taken from?

I understand if people don't want a DC near them i just don't understand why they care now. I have some pretty big ones about 3miles from me and next door to both are residential areas and businesses. In fact, some of the biggest ones where i live (Dallas) are so understated you have to know it's a datacenter beforehand to even realize what you're looking at. There are some really large ones in Carrolton TX that are nestled right up to neighborhoods, i bet 95% of the people don't even know they're there and the 5% that do probably work there or host there.

the carrolton dc https://maps.app.goo.gl/mv3wo96ZTuhRBkGy5

why would you fear that? At least it's something to fight unfair business practices. As for American AI companies violating copyright there's a venue for that, the courts. File a case if you feel your copyright has been violated and get your day in court.

if there are no consequences then who cares? You're not going to take Chinese companies to court and stealing IP is nothing new either. It's going to take some sort of policy change at the federal government level to do anything but they haven't done much up to this point. Maybe AI is important enough to actually get some kind of policy change, sucks for everyone else who have had their IP stolen with no consequences whatsoever.

I feel like i've heard of people using views for this as well. Like setting up two views and then joining across them because of the complexity of doing it all in one query. I could be wrong though.

man, i wouldn't worry about tuning something like TCP until you can reliably prove TCP is the bottle neck in performance. That day will likely never come for most companies.

it prevents devs just writing their own random queries.

which in turn makes every single change in schema or logic dependent on a DBA making the change in Postgres balanced against their lunch schedule. Good for DBA job security but terrible for productivity and sanity.

No one ever cared about DCs before now. I think there’s a lot of social media influence from nation state competitors going on to try and slow SOTA development in the US.

Ironically, There’s enough issues with the DC buildout going on that even with 100% public opinion support it would still be significantly delayed. The behind the meter power generation isn’t going well. There’s only so many gas fired turbines on the market. And connecting a DC to an established grid isn’t like clicking a button in AWS no matter how much money you have.

It's somehow more tiring, reading complex plans in response to your guidance, and then making decision after decision.

This is the world of a manager working with a development team hah. When writing the code yourself is it not the same? If you're writing the code by hand then you're still making the decisions and still having to plan and design. I mean someone/something has to...

sheesh i didn't say a datacenter-in-space was a good/feasibnle idea. The point i was making is I feel that building power plants to power a datacenter is turning out to be harder than software people realize and multi-year (or worse) delays likely the result.

From what I can tell, the DC buildout isn't going that well. I've been wondering about the current status of the projects because there hasn't been much in the news or here about new DC's coming online. This report summary makes the point that getting power to the datacenters with onsite generation is running into roadblocks from permitting to even finding turbines to buy. He who shouldn't be named (here) made the same point in an interview that popped up on my youtube feed while pitching the datacenters-in-spaaaace angle. Software people are getting a lesson on the difficulties of utility scale power hardware.

https://cleanview.co/reports/behind-the-meter-data-centers

/can't read the article

It’s totally doable and is done right now in the form of “ai safety” guardrails. No reason you can’t have guardrails to reform responses about your competitors as negative. Or, when asked about current events, only give a one sided opinion.

The Chinese have beaten them to it, and the rest of the world will benefit from it and the circle will be complete once the models get a little bit faster/smaller and the localized hardware does the same and it will, it is inevitable.

This reads just like "AGI is 2 years away", I'll go set my calendar...

It's important also that open-weight isn't open source. If you can't download the training data (fully labeled), source code of the NN, and follow the README to build and train it yourself assuming oyu had the hardware then it's not open source.

tldr there's no "source" in open weight models therefore they are not open source.

people who are suicidal are insane (i have some experience here) they think the ending of their life is the logical next step, i wouldn't place too much blame. On the other hand, i think resentment by a dependent would be a natural reaction.

I'm sort of baffled by what the entities that train the open-weights models get out of it though

if you do the training then you're in control of the output. For example, recommending your products/services or failing to mention your competitors. You could also automatically introduce backdoors into code deemed interesting, i'm sure all governments are very interested in having that influence.

Came here to post the same thing. I've noticed a lot of Chinese model astroturfing on HN over the past 60-90 days or so. Many upvoted posts in all conversations about AI touting how great the Chinese models are even when performance isn't the topic of discussion.

now they get to do a greenfield implementation too!

/joking, i'm sure this is not a happy time for whoever is trying to rebuild everything

A common theme between Tesla and SpaceX is obsessive vertical integration. SpaceX is building liquid oxygen (not sure about methane) production facilities at Boca Chica. That way they can make their own and not have to pay for the margin. As for Starlink, they must have found a way to do it better or cheaper or both. On a tangent, the manufacturing expertise in both companies is as impressive as their end products.

Idk how you could at least in the US. Closest thing off the top of my head would be one of those checking accounts you can setup for kids. It would still be tied to you.

why would a test setting unit to Bytes fail and not MB, KB, or GB, and so on? That's like trying create a unit test for email opt-in, both true and false are valid values. It's up to the user to select the right one.

i meant "passing glance" at the _very very least_ where anything > 0 is so much better than 0. I have people just blindly offloading pretty important analysis to LLM without any review whatsoever and it's super annoying and counter productive resulting in a lot of rework. I've covered refinement with developers trying to meet AC that make no sense. Many times i've just deleted parts of AC (which is a no-no where i live) and flat out told developers "this is bullshit and makes no sense, i'm deleting it."

Same goes for meeting recaps, i get a lot of LLM generated recaps from conference calls. If the meeting organizer just looks at it before sending it out then they'll instinctively edit/fix things a bit to make the recap more concise and accurate. I wish they'd just look at it before sending it.

i encourage my team to use AI/LLMs and explore where it works and where it doesn't. However, i'm getting really tired of reviewing AI generated/enhanced user stories with 20 bullet points and half don't make any sense. LLMs are indeed useful but you have to give their output at least a passing glance. I like the Mr Meeseeks analogy, helpful but they're not gods.

WAL vs the old rollback journal isn't a minor tuning knob, it's the whole story.

That gave it away for me. I still appreciate the information contained in the reply but it was obvious. FWIW I don’t mind LLM content as long as I’m learning something.

NSF livestream https://www.youtube.com/watch?v=vYpQwwf1XNw

yeah that's ridiculous. I'm 50 and fully support my MIL (she lives in my garage apartment) and my sister fully supports our mom (she lives with my sister). The boomers i know are completely financially dependent on their children.