I kept trying to hammer this on Lemmy, but no one listens :( Mods also seem uninterested in information hygiene.

CPU makers can’t really make system memory affordably, unfortunately. That’s why it’s separate in the first place :(

Intel has actually done this in the past, with a little eDRAM cache for their integrated graphics on some older 5000 series CPUs, like the 5775C. It topped out at 128MB.

AMD already does something similar with their X3D CPUs, albeit with SRAM… it tops out at 64MB.

They will sell you a bigger version, with IIRC 768MB of L3 memory, for many thousands of dollars.

Another issue is that CPU designs take many, many years to go from initial idea to manufacturing, along with truckloads of cash. So they couldn’t even respond to this shortage in 2026 if they wanted to.

Another is that AMD outsources their manufacturing anyway, though not Intel.

Strix Halo (AI Max CPUs) are basically that.

But they’re still DDR5 hanging off a bus, manufactured in the same place as sticks, so that wouldn’t really affect the price.

Another is to use ublacklist (if you aren’t already using ublock origin for some inexplicable reason) and add in AI/SEO slop blacklists. They will be filtered out of search.

Oh, and use the wiki redirect extension.

It doesn’t fix everything, but it’s a massive improvement for when the information you need is pretty recent.

Also, FYI the Kagi browser is great.

A prudent strategy would be a 2nd account for orders, maybe?

Yeah :(

Still though, Intel has their own fabs not really restricted by any of this. And not as easy to spin down as PCB making. So the CPU is likely to be the cheapest of anything.

You don’t need LLMs for that. An iPhone is plenty powerful enough for image recognition and text classification.

That’s sorta the funny thing about AI. There’s tons of potential, but it’s just unimplemented. Even on PC, you pretty much have to have some Nvidia GPU and fight pip setting up python repos to get anything working.

Does anyone even talk about what the “AI features” are?

Could I, liked recolor webpages? Automate ublock filters? Detect SEO/AI slop? Create a price/feature table out of a shopping page?

See, this would all be neat like auto translate is neat.

But I’m not really interested in the 7 millionth barebones chatbot UI. I’m not interested in loading a whole freaking LLM to auto name my tabs, or in some cutsie auto navigation agent experiment that still only works like 20% of the time with a 600B LLM, or a shopping chatbot that doesn’t do anything like Amazon/Perplexity.

That’s the weird thing about all this. I’m not against neat features, but “AI!” is not a feature, and everyone is right to assume it will be some spam because that’s what 99% of everything AI is. But it’s like every CEO on Earth has caught the same virus and think a product with “AI” in the name is like a holy grail, regardless of functionality.

As a silver lining, you think this could stabilize GPU prices? Or at least CPU prices?

If there’s less RAM/SSDs to build PCs with, then people will buy fewer GPUs/CPUs for them.

how are they making money??? What are you selling??

Keyword: “social media”

It’s almost like it’s coincidentally at the center of every modern grift out there.

Even diehards like Banon/MTG got pushed out over it. Never thought I’d see that.

Yep.

There’s tons of great reasons to hate AI. Tech Bros. Spam in every nook and cranny.

…But the power/water use has always been overblown. US tech is particularly sloppy about it, setting up generators and evaporators in cities, running huge training runs without proper optimization, but thats FOMO, impatience, and Tech Bro Evangelism more than a fundamental characteristic. It doesn’t have to be that way, and I’d wager they can’t sustain it long.

I dunno what you’re on about, Battlemage is great for the money and they appear to have committed to stick with Arc. And they have fab customers now.

…Yeah, Intel still has that corporate Game of Thrones going on internally. That’s not ideal. But AMD sunk much lower than that, and climbed out.

Out of curiosity, are Canadians not on Lemmy generally about that hostile to the US these days?

It’s a bit confusing looking from the outside in. It seems like there’s a lot of grassroots resentment, but on the other hand, there are some weird (conservative?) groups that seem pretty Trumpy.

Joke’s on you.

They don’t actually make any money. Not unless their a monopoly that’s captured regulators anyway.

All the billionares watched a few too many 2077 YouTube videos.

Why do you think they’re so specifically interested in space datacenters? And AR glasses? And AGI, and corpo states? Just to start.

since it’s needed to store training data.

Again, I don’t buy this. The training data isn’t actually that big, nor is training done on such a huge scale so frequently.

Aside: WTF are they using SSDs for?

LLM inference in the cloud is basically only done in VRAM. Rarely stale K/V cache is cached in RAM, but new attention architectures should minimize that. Large scale training, contrary to popular belief, is a pretty rare event most data centers and businesses are incapable of.

…So what do they do with so much flash storage!? Is it literally just FOMO server buying?

HDDs should be fine. No?

Well. I just FOMO ordered a SATA SSD. Thanks, OP.

Mostly because I got 2x64Gb sticks 2 months before they shot up to nearly 4X the price! And it doesn’t feel like this bump is going away soon.