Hacker News Reader

Comment by pixl97 | Hacker News Reader

>They think they can regulate all that? Come and take it from us.

Sorry, you can't buy GPUs any more.

And there you go, it's over for open source AI. Our supplies will dwindle and we'll be far behind the 'licensed and regulated' data centers that are allowed these 'munitions'.

View on HN

No GPU purchases means no game industry, no VFX industry, no economy of scale in production to keep unit costs anywhere near sane, and no significant profits to drive further research and development of GPUs. No government would have the will to plug that funding gap.

And besides, if you take away the bread and circuses, how long would such a government last?

The collateral damage level for this scenario is at a suicidal scale, and would just hand everything over on a platter to a competing high tech state.

> No GPU purchases means no game industry

It would be a huge hit to AAA games, but the games industry is a lot more than that, and very little outside of AAA games require a GPU.

This is completely false. We're not living in a world anymore where CPUs do the rendering for videogames. Indie game studios use some of the same engines (Unity, Unreal) as AAA, and those engines definitely require GPUs.

> Indie game studios use some of the same engines (Unity, Unreal) as AAA, and those engines definitely require GPUs.

That's funny, because I have no GPU on any of my machines, yet I can run games using Unity, Unreal, etc. I can even run at least some AAA games, albeit poorly.

Most people I know don't have a GPU, but most do play modern games on their computers.

Edit: I think I should clarify what I'm saying, since someone downvoted me and the only reason I can think for why that is is because of this confusion. I'm talking about dedicated GPUs here, not integrated ones. The topic of conversation was around dedicated GPUs.

Most modern CPUs (really SOCs) have an integrated GPU built in, which isn't half bad. Heck, the GPU often takes significantly up more space on the silicon die than the CPU does.

You mean you don't have a dedicated GPU. You wouldn't seen anything on your screen without a GPU at all.

Yes, in the context of this discussion, we were talking about dedicated GPUs so I didn't think it was necessary to add this bit of detail. I apologize for the omission.

And I was needlessly nitpicking. I mostly found it amusing to read a few sentences about not having a GPU, when almost every piece of software that renders anything has come to depend on GPUs.

> And there you go, it's over for open source AI. Our supplies will dwindle and we'll be far behind the 'licensed and regulated' data centers that are allowed these 'munitions'.

This is the endgame for the rhetoric OpenAI and its associates are espousing. They're positioning OpenAI et al. to be the Lockheed Martin of AI.

To be fair, it's never been easier to get access to thousands of GPUs in the cloud. It might be expensive, but that is an entirely different kind of barrier. Just a decade ago, it used to be that the only way to get access to thousands of GPUs was to get access to a supercomputer at a national lab. Now anybody with enough money can rent thousands of GPUs (with good interconnects too!) in the cloud. There's certainly a limitation on it from a money perspective, but access to the computational resources themselves is not a problem.

"The Cloud" == "Somebody else's computers"

If your government passes a law saying that GPUs can only be purchased or rented with a license, as OP was suggesting, all of that capacity disappears with the snap of a finger.

The GPU capacity of another country will be happy to fill the gap and export your money.

"Dear government, yes, I was spending my money accessing an illegal minution, please don't throw me too far under the jail"

I'd like you to stop and think about this for a minute... which country is going to be exporting powerful GPUs? If the US blocked GPU resources, do you think China will suddenly start allowing everyone to have them? No, they'd jump right on board with their own limitations so they didn't have to worry about internal issues.

Yes I've thought about it.

I'm talking about cloud services. Have you ever heard of anyone engaging in the use of illegal cloud services? I have.

Governments: "Whoever buys over X GPU hours must register with a state identification"

Amazon: "Can do"

Government: "Any projects that pool GPU resources will also be against the law and you must prevent it"

Amazon: "Ouch"

Also Amazon: "Hey bitch, we built and run your .gov cloud infra in your NSA/CIA staffed Mormon Data Centers"

Look at me ; Who is the surveillance now

(Yes the intels use mormons a lot...)

(Why do you think we have the largest private hedge fund [aside from apple, and those other guys])

There are many other foreign chip makers capable of making GPUs on their own, it is just their product is not as competitive as NVIDIAs in the consumer market. But that doesn't matter if we're talking about potential global human extinction level threat. Governments will fund it.

I cant imagine that such policy will be popular because GPUs are useful for many important things besides AI.

This "require licensing to own and/or operate GPUs beyond hobbyist capabilities" policy is exactly what Sam Altman asked Congress for last week, and what a bipartisan congressional committee seemed to agree with.

And there's OpenAI's moat.

They can also offer to imprison people who won't turn theirs in, and publicly announce that they got all relevant online sales records through a secret court warrant.

"Sorry, I sold my GPU to a buyer in a non-compliant state."

Hardware doesn't have to be physically present in my house in order for me to use it to run my code.

You do realize we're not training high end LLMs with even a few GPUs, but massive piles of them, right?

Of course.

But a ban won't happen anytime soon - it will be in the future. We are seeing improvement in efficiency, and GPUs are getting faster. We have various methods for distribution of computing function. All of these trends mean that by the time the axe falls it will be far more feasible for smaller bands of geographically distributed hackers to put something together that defeats the regulatory regimes that are being proposed.

And even if we aren't training models ourselves, execution of the large community models of the future will still benefit from GPU hardware.

Reply on news.ycombinator.com