News for hackers. Stuff that matters.

elorant

2.8KWs is just the GPUs. Add another 400 watt for CPUs and RAM and the rest of the hardware, and then leave at least a 30% headroom for PSU loss and spikes. Suddenly you're close to 4KW. And then of course you'd have to remove all that heat from the room.
Oracle's core business is healthy as fuck. It can't default. It could get into a long period of low R&D since all profits would be directed to serve their debt and that could hurt their product line, but they'd definitely can't just vanish into thin air.
Re:Private Jet plane crashes in Florida on Monday November 02, 2026 @08:23AM (#50336493)
Judging by what is circulating in Twitter today it mostly flared them up
Why the fuck does a freezer need access to the Internet in the first place?
Re:US tech stocks see sharp falls over AI cost fears on Friday October 30, 2026 @01:34PM (#50307341)
Their 24GB GPU took a year to reach Europe and the 48GB one is nowhere to be found. I mean even if you want to give Intel a second chance in the GPU market they’re so fucking incompetent.
Re:Ask HN: Are you tired too in this AI era? on Friday October 30, 2026 @05:29AM (#50301546)
I’m tired of exuberant hardware prices so that companies can shove AI down our throats on subpar products that work way worse than they did before.
Just short Oracle and Coreweave.
People aren't freaking out because self-hosting isn't a solve issue and it requires a lot of capital. The average company won't go and build an $1M GPU cluster just to self-host any model. If that gets commoditized then they'll start freaking out.
They don't have the infrastructure to meet global demand.
Re:Dead Money on Wednesday September 30, 2026 @07:33AM (#49907416)
I’ve been hearing about Chinese GPUs and Chinese RAM going mainstream for two years now. And yet I can’t find either in the European market.
Re:Dead Money on Tuesday September 29, 2026 @03:50PM (#49899331)
It makes perfect sense. Production moved away from consumer RAM to HMB RAM which is the one sitting inside those GPUs that sit in the warehouses. RAM manufacturers don't give a flying fuck if the hardware will be used or not as long as they're paid.

Also GPUs sitting in warehouses doesn't mean that they're there because demand is fabricated. There could be there because the data centers aren't built yet.

Re:You Are No Longer Invited to Dinner on Tuesday September 29, 2026 @09:41AM (#49893104)
I got tired just reading this comment.
Re:I said no and Apple said yes on Tuesday September 22, 2026 @06:50AM (#49799204)
Windows on laptops also just works, especially for high end ones. It’s the Linux on laptops that’s hit or miss.
Re:Codex Is Down on Thursday September 03, 2026 @11:31AM (#49551694)
Good effing luck to all the vibe coders out there
Some npm library that makes headers bold would be broken.
Re:Tim Cook Sold Out Steve Jobs on Tuesday September 01, 2026 @09:53AM (#49522057)
If they could, they would ship an apple version of the internet only available to apple devices.

Well Microsoft tried to do the exact same thing with that dreadful Microsoft Network in the early days of the web. So it's not inherit to Apple only, all large corporations would like their users locked up in their ecosystems.

Re:GLM-5.3 is now open-weight on Friday August 28, 2026 @06:42PM (#49485130)
Privacy isn’t only, I don’t want anyone to have access to my data. It could also be, I don’t want anyone to know my use case because it’s niche and highly profitable.
Same in Greek. It's "ekatommyrio" with a prefix of the order of magnitude "dis" from two for billions, "tris" from three for trillions, "tetrakis" from four for quadrillions and so on.
Re:Oceans hit highest temperature on record on Monday August 24, 2026 @04:12PM (#49425260)
Meanwhile, canned sardines went 20% up in the last month in Greece. And that's while we have plenty of local production. I can't imagine what it would be in countries where they import everything. Give it a few years and all fish food prices could go up substantially.
Re:Fable and the End of the Free Lunch on Monday August 24, 2026 @10:36AM (#49420421)
So why isn't this the default mode then?
I built from scratch a sparse text embedding model trained on a 13T token corpus. Not the same as an LLM because there’s no transformer in the mix, but still I learned a shit load of things in order to solve all sorts of problems that emerge when you try to access big datasets and daily update tables with billions of rows. But if it wasn’t for a specific use case that I tried to solve I don’t think that whatever knowledge I gained could be utilized in the market. Sparse models are a very small niche and most people I’ve come across with similar knowledge are in academic circles, not business related ones. So even if LLMs are all the rage these days I doubt the demand for people who know how to build them is that high. Someone who knows how to setup an open weight model and expose an API might be more valuable to a company these days.

Older »