hengheng

A better metaphor would perhaps be machining. You will not find a shop that has no CNC anywhere, but the machinists who can also use manual machines will always have an edge for certain things. Not that they're working manually much in their day to day, and not as all as a failsafe, but you can just tell who has that kind of background.

(And you can also tell whose mind is stuck in the old world, too.)

Pixel Phone's locally generated real-time subtitles are better than the ones on YouTube, that's how much effort is put into that.

The protocol needs ugly extensions to have a variety of devices plug into the same user account. For me back in the day this was an ssh session running the IRC client in a screen, and it was no worse than whatever else was available.

The Hackernews snark would likely be worse had they called it 'bureau'.

Have you consumed any media in the last decades.

Goalpost move detected.

Ed Zitron Voice: Is that good?

Depending on how you build it, you could run homeassistant next to your smb, which lends itself to all sorts of add-ons such as calibre-web for displaying eBooks and synchronizing progress.

Of course, gitea and surroundings, or similar ci/cd can be a fun thing to dabble with if you aren't totally over that from work.

Another fun idea is to run the rapidly developing immich as a photo storage solution. But in general, the best inspiration is the awesome-selfhosted list.

In your opinion, what would instead justify the total cost of devoting 10'000 people's lives to basic research?

That is equivalent to a continuous draw of 150 MW. Not great, not terrible.

Far less power than those projected gigawatt data centers that are surely the one thing keeping AI companies from breaking even.

Look into CXL, Oculink, and riser cables.

How's AMD's engineering support these days? I've heard through the grapevine that many laptops were mostly engineered by intel engineers, creating a natural moat because the laptop brands are used to not having to do much PCB layout or thermals.

AMD, I heard, seemed less capable, or less interested, or couldn't justify at their quantities, to do the same, which meant their engineering support packages were good for atx mainboards only, and maybe the occasional console.

This must have changed a while ago, does anyone have the tea?

I agree insofar as the motor is not a Big Ticket Item, opposed to ICE cars where the engine block is going to be 10% or more.

Tesla (I know) claimed a 30kg (?) weight loss on their Cybertruck (I know) just from moving their 12V systems to 48V, allowing for lighter cables at lower currents. Not all such potential is untapped, and my hunch is that there is more to be had with structural battery integration, battery cooling, and high voltage wiring.

To me the real question is what kind of trust has eroded that Americans stopped electing smart people into office.

That explains so much. Done to well for a goof channel, eclectic assortment of skills ("tactical garden trowel" vs fully equipped metal shop vs perfect video production), all fat trimmed off the videos.

I kinda want tvtropes to put a name on his slapstick humor. It's like looking over the shoulder of that weird uncle that seems to live in an entirely different world.

How can you tell without knowing the bus width?

Write it as a children's book. A literal ELI5.

(Knowing, of course, that it will still be read mainly by engineers. But that's the charm.)

Chiron still has that Piëch handwriting on it. It's driveable enough to take your wife to the opera. Full regulatory compliance, low wind noise at high speeds, all that. I don't want to say it is compromised, but it's not as extreme as it could be.

The closer ICE comparison would be Koenigsegg (447 kph/278 mph), Hennessy Venom GT (435/270) and SSC Tuatara (455/283, no shenanigans). SSC have reached 295, they were clearly aiming for 300. It's no 308 but it's reasonably close.

All these are also relatively small companies with relatively low budgets -- none of the big manufacturers seem interested in top speeds anymore.

> we prob don't want to have it in Europe

Best indicator as to what that is.

My guess is automated surveillance, which is also where this whole play has to be headed.

Has the research aspect ever been more than a fig leaf and a "might as well"?

Strong parallel to rockets and high-altitude planes used for atmospheric research ...

I've also learned the hard way to Google "AM4 main board tier list" before buying.

Some boards can run a 5950X in name only, while others can comfortably run it close to double its spec power all day. VRMs are a real differentiator for this tier of hardware.

(If anyone can comment on the airflow required for 400-500W Epyc CPUs with the tiny VRM heatsinks that Supermicro uses, I'm all ears.)

I don't even think it's cynical anymore to assume that this is the entire reason why Meta are pursuing this.

They see themselves in a race to produce the most radical, most efficient machine that produces the most effective addictive response. Content has been interchangeable for decades, everything is about the naked control over people's attention, because that is having power over people.

There is a very modernist logic in the whole effort. Everything must be taken to its extremes, nothing is ever enough, and nothing good sits in the middle of anything, and having values is only a detriment in this race.

I can't imagine there being enough information for true fingerprinting of individual devices. With ten million iPhones being made per month, surely the blur patterns have to have some overlap?

It's the same with military sites in France, where satellite images are blurred but you still get street view, and if that doesn't exist, you can still look at ground level photos, like so https://maps.app.goo.gl/Kr2822pASFRPLJHR7

Compliance is fun.

Needs a PCIe bifurcation chip on the main board for all we know. Compatibility is going to be fun.

On my Fenix 6, Bluetooth volume control broke after a few years, and now the volume slider does nothing, which the hivemind confirmed.

So, it is in fact worse than not getting upgrades, the software actually deteriorated over its lifecycle.

I've seen US startup people deal with manufacturing. There is a class gap that leaves no real space for skilled blue-collar work to exist and also be business critical, and fully integrated into the team. That is how a three-week turnaround for a milled prototype part ends up being tolerated.

Surprised to see something on hackernews that isn't an AI picked selection of articles :-)

/s

The recent qwen3 models run fine on CPU + GPU, and so does gpt-oss. LM Studio and Ollama are turnkey solutions where the user has to know nothing about memory management. But finding benchmarks for these hybrid setups is astonishingly difficult.

I keep thinking that the bottleneck has to be CPU RAM, and for a large model the difference would be minor. For example with an 100 GByte model such as quantised gpt-oss-120B, I imagine that going from 10G to 24G would scale up my tk/s like 1/90 -> 1/76, so 20% advantage? But I can't find much on the high-level scaling math. People seem to either create calculators that oversimplify, or they seem too deep into the weeds.

I'd like a new anandtech please.