They’re not, though.
Have you gone outside, and seen how people use their phones? The public is utterly addicted to social media.
They won.
I think Pondsmith got it right with Cyberpunk. The future of computing is corporate controlled siloes people simply cannot escape, and the “open web” and PC as we know it is basically defunct. Maybe small circles of people will congregate in Usenet-like circles. But not at scale.
I have a 3090, 7800, and 128GB ram from pre-rampocalypse. It’s not nothing, but there are definitely crazier home labs.
Deepseek V4 is native 4 bit.
I am running this quant, but the quantization loss is measurably low: https://huggingface.co/Downtown-Case/DeepSeek-V4-Flash-0731-128GB-RAM-IK-GGUF
The native MXFP4 is only a bit bigger.
which I think is only available from China right now.
Nah, 0731 weights were released. I’m running it locally right this second.
I don’t know about Deepseek API specifically, but there are tons of places to get it.
Well that would mean holding social media liable.
That would be fantastic. But it seems unlikely, sadly.
Well, the current trajectory points to:
LLM capabilities topping out.
All the tooling around them not keeping up anyway.
Actual “AGI” being distant and completely unrelated to contemporary models.
Inference costs plummeting.
The last one is critical.
Right this second, you can run Minimax H3 on a desktop for the tiny fraction of the compute/RAM OpenAI Sora took. And it’s better.
In a month, it will be ~8X faster.
You can run DeepseekV4 flash, dirt cheap, and get what Claude was less than a year ago. And it’s gonna spread to every host out there, to systems like Cerebas ASICs that don’t even need HBM.
So… Even if you’re an AI acolyte. And we go with that for the sake of argument…
We don’t actually need all that RAM for hosting generative models?
I’m very interested to see what happens to all these datacenters over the next two-three years.
They spent all this money on something that’s gonna be cheap as dirt to run, largely run locally, and that won’t need GPUs once bitnet takes off, sooo… they can’t make money off that.
What happens then?
What happens to all those Stargate RAM wafers, and excess datacenters?
I have similar anecdotes.
I don’t get that at all. Hence I don’t purely buy the “popular with bigots” angle either; I think it’s just trendy, kind of like Starbucks, and people probably aren’t thinking about their stances when they wait in line.
I’ve discovered that good honey mustard mixed with BBQ sauce is better than the Chuck-fil-a sauce anyway (as it’s just mustard + bbq + sugar).
But yeah, that’s interesting. I’ve never seen one of their food trucks before.
GPU prices have gone up and up since RTX 2000 era, basically.
My 3090 is about as expensive than it was at launch, 6 years ago.
There were crypto slumps, many moons ago, but I don’t think that’s happening again.
I’m not sure I worded this right before, but it takes ~half a decade to plan and produce a GPU.
They can’t just switch out the bus or memory on an existing design.
So they could make a HBM gaming GPU. But they would have had to have started ~2 years ago.
…That’s extremely unlikely.
It’s unlikely they’ll even plan one now.
AMD did indeed try making HBM gaming GPUs, and it turns out it’s supremely uneconomical; they immediately pivoted to the opposite extreme (narrow bus + larger cache).
However, LPDDR GPUs appear to already be in the pipe, so that will provide some GPU market relief.
If the memory wafers are largely HBM, so they can’t be used for gaming hardware.
If you’re thinking of HBM gaming GPUs, the chip design pipeline is too long. If Nvidia/AMD/Intel started making one right now, and rushed it, it would still be many years away.
…However, a lot of LPDDR5 is being made too.
Hence the rumors of AMD hedging their bets and using LPDDR for some future GPUs.
It’s engagement optimized, that’s why.
That’s the whole point of the spam; being engagement bait.
Same thing applies to any human made clickbait, and people have demonstrated they simply cannot help themselves, and click it.
That won’t work for long. Generation can be hosted in a country outside the jurisdiction.
Current services do need such a law anyway, and many already are watermarked. But I’m just saying it wouldn’t even slow down spammers once they figure it out.
Its not trivial.
Check the level1 forums, where some electronics gurus are trying to get a single adapter MI250X to even boot. The problem isn’t the adapters, but firmware that makes it difficult.
…They actually did.
I got an AMD 7950 is a crypto bust. Then a Nvidia 980 TI in the next one, dirt cheap.
Nvidia figured it out in the most recent bust though, and controlled supply fairly well.
They aren’t PC components. They’re server GPUs.
They don’t even have ROPs anymore. There was a time when the Nvidia P100/A100 could technically game, but that’s generations ago.
FYI, during the crypto busts, Nvidia bought back GPUs and basically incinerated them to keep demand for new stuff up.
…I’m pretty sure that’s what they will do again.
To be fair, people generally won’t keep up with something they aren’t interested in, much less something they are actively disinterested in. That’s reasonable.
I do have an irrational pet peeve of people making claims on something they know little about though, which is why I may have come on as hostile…
I was mistaken, I think. Apparently they used to be involved with Nebula, but that’s no longer the case.
I’ve dipped my toe in and out of modded MC, but it’s kinda unengaging without a good server to play on.
Yeah.
The Bill Gates connection is there, but the level of influence is just a conspiracy as far as I can tell.
Will it though?
I posit there’s a saturation point where there’s “enough” LLM in use, and we are not far from that.
The whole pitch from Altman and such is scaling models up. But that’s not working.