Hacker News new | past | comments | ask | show | jobs | submit | agilob's comments login

Some of us also worry about energy consumption

They idle at pretty low wattages, and since the bulk of the TDP is rated for raster workloads you usually won't see them running at full-power on compute workloads.

My 300w 3070ti doesn't really exceed 100w during inference workloads. Boot up a 1440p video game and it's a different story altogether, but for inference and transcoding those 3060s are some of the most power efficient options on the consumer market.


Interesting, my 3060 uses 150-170W with 14B model on Ollama, according to nvidia-smi.

They were disappearing 6 years ago too


Why don't all these new ~AI projects publish Dockerfiles and images?

I've been looking at several projects recently for subtitle, image generation, voice translation, any AI coding assistant, and none of them had a out of box support for containers. Instead authors prefer to write details install instructions, commands for Fedora, Ubuntu, Arch, notice to Debian developers about outdated python... Why is that?


I see twoo reasons:

1. Because they're researchers, not devops experts. They release the model in the way that they are most familiar with, because it's easiest for them. And I say that as someone who's released/open-sourced a lot of AI models: I can see how Docker is useful and all that, but why would I invest the time to do package up my code? It took long enough to cut through the red tape (e.g. my company's release process), clean up the code, document stuff. I did that mostly because I had to (red tape) or because it also benefits me (refactorings & docs). But docker is something that is not immediately useful for myself. If people find my stuff useful, let them do it and repackage it.

2. most people using these model don't use them in docker files. Sure, end users might do that. But that's not the primary target for the research labs pushing these models out. They want to reach other researchers. And researchers want to use these models in their own research: They take them and plug them into python scripts and hack away: to label data, to finetune, to investigate. And all of those tasks are much harder if the model is hidden away in a container.


Thats quite deep for the region. TIL Baltic Sea is 55m deep on average, but 80m deep in the region where cable was cut.

That also means that 2nd easiest to acquire PADI qualification is enough to dive there by yourself.


> That also means that 2nd easiest to acquire PADI qualification is enough to dive there by yourself.

How? Open water diver is 18m, advanced Open water diver is 30m. With „deep diver“ specialty, you’re at 40m.

I think you need to become an instructor to go deeper, that’s not the 2nd easiest qualification at all.


> How? Open water diver is 18m, advanced Open water diver is 30m. With „deep diver“ specialty, you’re at 40m.

Correct. Those are the limits of those certifications.

> I think you need to become an instructor to go deeper, that’s not the 2nd easiest qualification at all.

In order to go down to deeper depths and returning you generally need a technical or professional certification. Anyone can go to any depth, as long they don't expect a reasonable chance to get back to the surface alive.

PADI open circuit technical course go down to 50 but there are other agencies for technical diving with a bit better reputation, and those generally max out at 100.

Most people who are trained for 80+ depth usually do so on a rebreather. Padi do have a 100m course for that. However with rebreathers, students mostly look towards specific instructors and not agencies. What agency a instructor uses is generally much less important than that you trust the instructor.


I think PADI doesn't even go that far, it is predominantly a leisure diving organization. Their technical diving courses seem to end at 50m. Not really related to instructor courses either, a regular instructor typically has no need to go beyond 30m.

Diving to 80m is a serious endeavor on this scale.


80m is moderately serious. I've personally done 75m in the Atlantic and I wouldn't say I'm superhuman or anything (or military).

It requires helium mixes and personally I'd only want to do it with a rebreather, not normal "blowing-bubbles" type SCUBA gear (too wasteful on the breathing gas). 20-30 minutes at the bottom will mean around 90-120 minutes decompressing. You don't want to be doing anything taking too long, or being too laborious.

Could be done from a fishing boat without any problem.


Yes, it's very doable given the right training. But it's not really an extension of what a typical "PADI diver" might do, is what I meant.

Having had equipment failures no deeper than 30m, I decided any dives that require decompression are not for me. I'll leave deep dives and sabotage to others :)


You know, I never dove but I always wonder, why do deep divers not breathe a nitrogen-free atmosphere before they dive to flush out the nitrogen? I understand astronauts do this too (they breathe pure oxygen but at a lower pressure so it's not dangerous). And because this a sudden pressure drop won't give them the bends.

It would seem simple, being able to deep dive then without decompressing. I'm sure there's a good reason it's not being done, I just wonder what.


I think what you breathe pre-dive is almost irrelevant. The real kicker is what you breathe during the dive, and that's always going to be gas under pressure. The levels of gas in your blood go up very quickly. Recreational divers can only afford ~10-20 minutes at 30m depths before they need to decompress. And that doesn't even count as "deep".

You can't breathe pure oxygen under water, as it becomes very toxic very quickly as pressure goes up. So you have to dilute it with something. Nitrogen is easy to get hold of (actually you just compress atmospheric air). I'm also not sure cutting nitrogen out helps much with decompression sickness, you get it with other gases too - maybe more slowly, I'm not sure.

You can, and do, replace nitrogen with other gases, mostly helium (and in rare circumstances, hydrogen), but for other reasons. Nitrogen too becomes toxic under pressure, around the 30-40m mark, with symptoms similar to drunkenness - not good under water.

Funnily enough, helium does the same deeper down, except the symptoms there are the opposite - nervousness, tension etc. So to some extent perhaps these two balance out. Hydrogen has very low toxicity, but for some reason people feel queasy about mixing pure hydrogen and oxygen.

Crucially, too, a lot of deep diving is commercial in nature, and cost-sensitive. Anything that isn't just compressed and bottled air is getting very expensive very quickly.

It all gets incredibly complicated. Deep enough, you need to breathe gas that contains so little oxygen that you would suffocate on it at surface pressure. At pressure, it's fine - the partial pressure of oxygen is sufficient, and having less of it makes it less toxic. But you have to be careful to only use this gas deep enough, else you'll run out of oxygen. Or at other times, when decompressing, you might use an oxygen-rich mix, which while more toxic, helps you purge some of the other gases from your bloodstream, cutting decompression times.

It's a fascinating topic to read about from the comfort of my sofa.


Might work for technical divers but regular leisure divers use compressed air which contains nitrogen anyways. If you prebreathe pure oxygen and then begin the dive using regular air, you will reaccumulate the nitrogen pretty quickly, I think, so you will need to decompress like normal. And you can't dive with pure oxygen because it becomes toxic when going lower than about 6m due to the high partial pressure.

Also, you would need to have pure oxygen available and breathe that for maybe an hour before each dive, which would be expensive and annoying. For touristic dives, especially along the shore, it's easier to just plan the dive so you can still see something interesting while waiting for decompression.


Anybody can dive there. Surviving the trip back up is what most of the classes are for.


There is no reason to dive, large freight ships can have anchors with chains several hundred meters long.


In Java I use jbang https://www.jbang.dev/


I heard from Polish news that Spanish government formed a dedicated team to negotiate with US and they succeed, yet Polish government wasn't interested in the talks at all despite being invited to.


being efficient is not a strongest side of any polish politician


This is exactly what all Europeans watching US politics expected. No more, no less.


The executive order talks about building new geothermal power plants to support AI infra


How about we cut the AI bit and just get a nice C02 decrease, i.e. the actual thing that will support future humans? THat's one more coal plant we can shut down.


What about the water?


>Are we trying to slow global competition with bad actors by restricting their ability to run compute on these.

Huawei GPUs


Don't worry, we - software developers are going to ruin the software with AI features that you will need to upgrade to Ryzen Al Max+ 395 just to run an editor.


I'm sure this is tongue in cheek, but it's a legitimate fear.


Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: