Can't wait!

Gregor

What did OP mean by that?

[email protected]

I thought the efficiency curve for GPUs peaked before 100%. If electricity is your primary cost, driving the GPUs at lower loads saves money.

So you might end up with GPUs that spent their entire life at a steady 80% load or something.

[email protected]

Just desserts

[email protected]

Most crypto mining outfits undervolt their cards for lower power usage. They aren't cranking them as you say they are. A dead GPU doesn't produce anything for you; cranking it up the chance that it will fail. You're better off running it an extra 4 years at a lower voltage than you are cranking it for 1.

[email protected]

Have you seen the price of new GPUs? Sure ya do. Maybe they only last a few years. That's alright.

[email protected]

When the "Web" bubble burst, were we left without websites? No.

When the AI bubble bursts, it isn't going away either. It replaces too much labor. Training the AI models is expensive, but running them isn't. Especially on Lemmy, too many dullards don't understand why AI is so popular; it's because it's replacing your job - or allowing 1 person to do the jobs of 10.

[email protected]

When the AI bubble bursts, it isn't going away either. It replaces too much labor.

The claimed amount of labour it could potentially replace is in fact part of the bubble.

[email protected]

The fact that you can claim this unironically, speaks volumes of your ignorance.

[email protected]

True, but many MANY companies went under during the dot com burst and made server hardware cheaper on the secondary market for a while

Same will happen with the AI bubble, there will undoubtedly be a few survivors, but many won't. There will be dead startups everywhere.

[email protected]

Personally, I want to try to have a llama or something rewrite voice to text prompts into Home Assistant commands. Should be very cool.

But yeah, I won't pay the current prices to run it either, nor use the cloud.

[email protected]

Yes.

They don't exactly age, but top of line chips have very large currents in very small conductors. When you do that with DC current, your conductors deform with time, up to the point that they stop working correctly.

[email protected]

[email protected]

Unfortunately, this time around the majority of AI build up are GPUs that are likely difficult to accomodate in a random build.

If you want a GPU for graphics, well, many of them don't even have video ports.

If your use case doesn't need those, well, you might not be able to reasonably power and cool the sorts of chips that are being bought up.

The latest wrinkle is that a lot of that overbuying is likely to go towards Grace Blackwell, which is a standalone unit. Ironically despite being a product built around a GPU but needing a video port, their video port is driven by a non-nvidia chip.

[email protected]

Silverlight, what are you doing here? Go on, get outta here!

[email protected]

You're more likely to see the college bubble burst before AI. Colleges are struggling with low admissions.

[email protected]

This was my understanding as well - that miners often underclock their GPUs rather than overclock them.

[email protected]

I want the hype bubble to burst because I want them to go back to making AI for useful stuff like cancer screening and stop trying to cram it into my fridge or figure out how to avoid paying their workers, and the hate bubble isn't going to stop until that does.

[email protected]

My use case is for my own AI plans as well as some other stuff like mass transcoding 200TB+ of..."Linux ISOs" lol. I already have power/cooling taken care of for the other servers I'm running

I've already got my gaming needs satisfied for years to come (probably)

[email protected]

I never said that they can't serve a function, just that corporations like openai have a vested interest inflating their potential impact on the job market.

[email protected]

GPU prices are gonna get cheaper, annnnyyyy day now folks, any day now