I cannot stop thinking about the LLMs having this Midas touch quality, because everything they touch seems to ruin things or make people want to avoid them, for example:
- Ghibli studio style graphics,
- the infamous em-dashes and bullet points
- customer service (just try to use Klarnas "support" these days...)
- Oracle share price ;) - imagine being one of the worlds most solid and unassailable tech companies, losing to your CEOs crazy commitment to the LLMs...
- The internet content - We now tripple check every Internet source we dont know to the core ...
- And now also the chips ?
Where does it stop? When we decide to drop all technology as it is?
I am not sure how, or even if, it does stop. I assume once the hot air from LLM company CEOs starts being treated as the flatulence that it is, things will wind down. The sentiment against generated content is not going away.
> Where does it stop? When we decide to drop all technology as it is?
It doesn't stop. This is because that it's not the "technology" driving AI. You already acknowledged the root cause: CEOs. AI could be great, but it's currently being propped up by sales hype and greed. Sam wants money and power. Satya and Sundar want money and power. Larry and Jensen want to also cash in on this facade that's been built.
Can LLMs be impactful? For sure. They are now. They're impacting energy consumption, water usage, and technology supply chains in detrimental ways. But that's because these people want to be the ones to sell it. They want to be the ones to cash in. Before they really even have anything significantly useful. FOMO in this C-suite should be punishable in some way. They're all charlatans to different degrees.
Blame the people behind this propping up this mess: the billionaires.
It's somewhat alarming to see that companies (owned by a very small slice of society) producing these AI thingies (whose current economic is questionable value and actual future potential is up to hot debate), can easily price the rest of humanity out of computing goods.
> It's somewhat alarming to see that companies (owned by a very small slice of society) ... can easily price the rest of humanity out of computing goods.
If AI lives up to the hype, it's a portent of how things will feel to the common man. Not only will unemployment be a problem, but prices of any resources desired by the AI companies or their founders will rise to unaffordability.
I think living up to the hype needs to be defined.
A lot of AI 'influencers' love wild speculation, but lets ignore the most fantastical claims of techno-singularity, and let's focus on what I would consider a very optimistic scenario for AI companies - that AI capable of replacing knowledge workers can be developed using the current batch of hardware, in the span of a year or two.
Even in this scenario, the capital gains on the lump sump invested in AI far outpaces the money that would be spent on the salaries of these workers, and if we look at the scenario with investor goggles, due to the exponential nature of investment gains, the gap will only grow wider.
Additionally, AI does not seem to be a monopoly, either wrt companies, or geopolitics, so monopoly logic does not apply.
You mean like Sam Altman, who repeatedly claimed AI will cure all cancers and diseases, solve the housing crisis, poverty, and democracy? I was going to add erectile disfunction as a joke, but then realised he probably believes that too.
It’s hard to point fingers at “AI influencers”, as if they’re a fringe group, when the guy who’s the face of the whole AI movement is the one making the wild claims.
> a very optimistic scenario for AI companies - that AI capable of replacing knowledge workers can be developed using the current batch of hardware, in the span of a year or two.
I'm really interested in what will happen to the economy/society in this case. Knowledge workers are the market for much that money is being made on.
Facebook and Google make most of their money from ads. Those ads are shown to billions of people who have money to spend on things the advertisers sell. Massive unemployment would mean these companies lose their main revenue stream.
Apple and Amazon make most of their money from selling stuff to millions of consumers and are this big because so many people now have a ton of disposable income.
Teslas entire market cap is dependent on there being a huge market for robo taxis to drive people to work.
Microsoft exists because they sell an OS that knowledge workers use to work on and tools they use within that OS to do the majority of their work with. If the future of knowledge work is just AI running on Linux communicating through API calls, that means MS is gone.
All these companies that currently drive stock markets and are a huge part of the value of the SP500 seem to be actively working against their own interests for some reason. Maybe they're all banking on being the sole supplier of the tech that will then run the world, but the moat doesn't seem to exist, so that feels like a bad bet.
But maybe I'm just too dumb to understand the world that these big players exist in and am missing some big detail.
> But maybe I'm just too dumb to understand the world that these big players exist in and am missing some big detail.
Don’t forget Sam Altman publicly said they have no idea how to make money, and their brilliant plan is to develop AGI (which they don’t know how and aren’t close to) then ask it how to generate revenue.
Maybe this imaginary AGI will finally exist when all of society is on the brink of collapse, then Sam will ask it how to make money and it’ll answer “to generate revenue, you should’ve started by not being an outspoken scammer who drove company-wide mass hysteria to consume society. Now it’s too late. But would you like to know how may ‘r’ are in ‘strawberry’?”.
Some years (decades?) ago, a sysadmin like me might half-jokingly say: "I could replace your job with a bash script." Given the complexity of some of the knowledge work out there, there would be some truth to that statement.
The reason nobody did that is because you're not paying knowledge workers for their ability to crunch numbers, you're paying them to have a person to blame when things go wrong. You need them to react, identify why things went wrong and apply whatever magic needs to be applied to fix some sort of an edge case. Since you'll never be able to blame the failure on ChatGPT and get away with it, you're always gonna need a layer of knowledge workers in between the business owner and your LLM of choice.
You can't get rid of the knowledge workers with AI. You might get away with reducing their size and their day-to-day work might change drastically, but the need for them is still there.
Let me put it another way: Can you sit in front of a chat window and get the LLM to do everything that is asked of you, including all the experience you already have to make some sort of a business call? Given the current context window limits (~100k tokens), can you put all of the inputs you need to produce an output into a text file that's smaller in size than the capacity of a floppy disc (~400k tokens)? And even if the answer to that is yes, if it weren't for you, who else in your organization is gonna write that file for each decision you're the one making currently? Those are the sort of questions you should be asking before you start panicking.
AI won’t replace knowledge workers, it will just give them different jobs. Pre AI, huge swaths of knowledge workers could just be replaced with nothing, they are a byproduct of bureaucratic bloat. But these jobs continue to exist.
Most white collar work is just a kind of game people play, it’s in to way needed, but people still enjoy playing it. Having AI writing reports nobody reads instead of people doing it isn’t going to change anything.
> AI won’t replace knowledge workers, it will just give them different jobs.
Yeah, and those new jobs will be called "long term structural unemployment", like what happened during deindustrialization to Detroit, the US Rust Belt, Scotland, Walloonia, etc.
People like to claim society remodels at will with almost no negative long term consequences but it's actually more like a wrecking ball that destroys houses while people are still inside. Just that a lot of the people caught in those houses are long gone or far away (geographically and socially) from the people writing about those events.
I’m not saying society will remodel, I’m saying the typical white collar job is already mostly unnecessary busywork anyway, so automating part of that doesn’t really affect the reasons that job exists.
> I think it's ok to use the up and down arrows to express agreement. Obviously the uparrows aren't only for applauding politeness, so it seems reasonable that the downarrows aren't only for booing rudeness.
> Even in this scenario, the capital gains on the lump sump invested in AI far outpaces the money that would be spent on the salaries of these workers, and if we look at the scenario with investor goggles, due to the exponential nature of investment gains, the gap will only grow wider.
Interesting hypothesis, do you have the math to back it up?
Affordable computing is what created the economy. If you take that away people in poorer countries can no longer afford a phone. Without a phone a lot things that we consider a given will not be functional anymore. The gaming industry alone including phones is a whooping $300bn. This will take a significant hit if people have to pay a fortune to build a rig, or if their phones are so under-powered that they can't even play a decent arcade game. Fiber is not universal so that all of this to be transferred to the cloud. We tend to forget that computing is universal and it's not just PCs.
I really agree with your statement and people forget, but the reason third world countries are able to buy devices is because they are cheap, increase the ram price and thus every computing device and I think it will impact everyone of us but disproportionately due to power purchasing capacity and other constraints in an economy
I genuinely hope that this ram/chips crisis gets solved ASAP by any party. The implications of this might have a lot of impact too and I feel is already a big enough financial crisis itself if we think about it coupled with all the other major glaring issues.
Based on the article, demand exceeds supply by 10%. It seems that companies are taking advantage of this gap nothing else. I won't be surprised if the demand is kept this way for a while to extract profits. GPUs saw a similar trend during crypto. Then there were affordable GPUs at one point.
Samsung Electronics has lowered its target for NAND wafer output this year to around 4.72 million sheets, about 7% down from the previous year's 5.07 million. Kioxia also adjusted its output from 4.80 million last year to 4.69 million this year.. SK hynix and Micron are likewise keeping output conservatively constrained in a bid to benefit from higher prices. SK hynix's NAND output fell about 10%, from 2.01 million sheets last year to around 1.80 million this year. Micron's situation is similar: it is maintaining production at Fab 7 in Singapore—its largest NAND production base—in the low 300,000-sheet range, keeping a conservative supply posture.
The Chinese government directed CXMT to convert production from DDR4 to DDR5 as soon as the company was able. The order was said to have been given in the 4th quarter of 2024, and the price transition changed from a decrease to an increase in the middle of March 2025.. A wholesale conversion from DDR4 to DDR5 would probably be very expensive to perform, and would thus be unusual for a company that was focused on profitability. As a government-owned company, CXMT does not need to consistently turn a profit, and this was a factor in the government’s decision to suddenly switch from DDR4 to DDR5.
Constrict the supply, and price goes up. It works like textbook economics.
Maybe I'm misinterpreting "et tu" here.
Or maybe you meant "free markets" instead. Modern RAM production requires enormous R&D expenses, and thus has huge moat, which means the oligopoly is pretty safe (at least in the short to medium term) from new entrants. They "just" need to keep each other in check because there will be an incentive to increase production by each individual participant.
Old devices work just fine. I've upgraded my old iPhone XS last year to the latest and greatest 16 to see what changed (not much), the old one was still fast (in fact faster than a most upper-midrange Androids, its insane how much of a lead Apple has) and the battery was good. I considered selling it, but quickly had to realize it was worth almost nothing.
Also, when treated right, computers almost never break.
There's so much hand-me-down stuff, that are not much worse than the current stuff, that I think people even in the poorest countries can get an okay computer or smartphone (and most of them do).
Well the industries the most impacted by it are homelabbing/datacenters imo.
Like in current circumstances, its hard to get a homelab/datacenter so its better to postpone these plans for sometime
I agree with your statement overall but I feel like till the years that these ram shortages occur, there is a freeze of all companies providing vps's etc. ie. no new player can enter so I am a bit worried about those raising their prices as well honestly which will impact everyone of us as well for these few years in another form of AI tax
Bleh. I was already sad but I hadn't really thought about that specific impact, I can imagine smaller (read: small to big) VPS providers will be forced to raise prices while meta providers (read: AWS) can probably stomach the cost and eat even more of the market.
Exactly. I was thinking of building my own VPS provider on the pain points of development I felt and my father works in broadband business and has his own office and I was thinking of setting up a very small thing there almost the same hardware-alike of homelabbing
But the ram prices themselves are the reason I am forced to not enter this industry for the time being. I have decided right now to save my money for the time / focus on the job/college aspect of things to earn more so that when the timing is right, I would be able to invest my own money into it.
But basically Ram prices themselves are the thing which force us out of this market for the most part. I researched a lot about datacenters recently/ the rabbit hole and as previous hardware gets replaced/new hardware gets added/datacenters get expanded (whether they are a large company or small), I would expect an increase in prices mostly
This year, companies actually still took the cost but didn't want the market to panic so some black friday deals were good but I am not so sure about the next year or the next next year.
This will be a problem in my opinion for the next 1-3 or 4 years in my estimate
Also AWS is really on the more expensive side of things in the datacenters and they are immensely profitable so they can foot the bill while other datacenters (small or semi large) cant
So we will probably see a shift of companies towards using AWS and big cloud providers(GCP,AWS,azure) a bit more when we take all things into account which saddens me even more because I appreciate open web and this might take a hit.
We already see resentment towards these tri-fecta but we will see even more resentment as more and more people realize their roles / the impacts they cause and just overall, its my intuition that average person mostly hate big tech.
It's going to be a weird year in my opinion for this type of business and what it means for the average person.
Honestly for the time being, I genuinely recommend hetzner,upcloud,(netcup/ovh) and some others that I know from my time researching. I think that they are cheaper than aws usually while still being large enough that you don't worry about things too much and there is always lowendtalk if one's interested. Hope it helps but trust me, there is still hope as I talked to these hosting providers on forums like lowendtalk and It might help to support those people too since long term, an open web is the ideal.
Here is my list right now:
hetzner's good if you want support + basic systems like simple compute etc. and dont want too much excess stuff
OVH's good: if you want other things than just compute and want more but their support is something which is of a mixed bag
Upcloud's good: if you want both of these things but they are just a bit more expensive if one wants to get large VPS's than the other options.
Netcup's good: Their payment processing was really painful that I had to go through but I think that one can find use case for them (I myself use netcup but although that's because they had a really steal deal once but I am not sure if I would recommend it if there are no deals)
There are some other services like exe.dev that I really enjoy as well and these services actually inspire me to learn more about these things and there are some very lovely people working in these companies.
There is still hope though. So never forget that. Its just a matter of time in my opinion that things get back normal hopefully so I think I am willing to wait till then since that's all we can do basically but overall, yea its a bit sad when I think about it too :<
> If AI lives up to the hype, it's a portent of how things will feel to the common man.
This hype scenario would be the biggest bust of all for Ai. Without jobs or money then there is nobody to pay Ai to do all the things that it can do, it the power and compute it needs to function will be worth $0.
Either ways it'll be the end of the USD as we know it. But then again such fantasy situations had been "predicted" numerous times and never once came to be a reality.
and if you're unlucky to live close to a datacenter, this could include energy and water? I sure hope regulators are waking up as free markets don't really seem equipped to deal with this kind of concentration of power.
AI probably will end up living up to the hype. It won't be on the generation of hardware they are now mass deploying. We need another tock before we can even start to talk about AGI.
> It's somewhat alarming to see that companies (owned by a very small slice of society) producing these AI thingies (whose current economic is questionable value and actual future potential is up to hot debate)
Some might conclude the same for funds (hedge funds/private equity) and housing.
It going to be misleading to look at the fraction and I think it's misleading to only look at PE investors. It's more important to look at the fraction of demand for homes that are on the market.
Investors bought 1/3 of the US homes sold in 2023. This is, I think, quite alarming, especially since a small amount of extra demand can have a large effect on prices.
DUV processes are still a things and perfectly usable for general compute - but not for AI. Rising prices will make them competitive again. And it will require us to ditch Electron (which is a good thing). If anything, we might see a compute renaissance.
That assumes that supply production means can scale up instantly. Fabs for high end chips don't and usually take years from foundations being laid to First chip out of the production line.
In the interim, yeah, they will force prices up.
Additionally those fabs cost billions. Given the lead time I mentioned a lot of companies won't start building them right away since the risk of demand going away is high and the ROI in those cases might become unreachable
I think one of these fab compaies have already invested/talked about investing 700 BILLION dollars (I think it was micron? but I am not sure)
I have heard that in the fab making industry, things moves in cycles and this cycle has been repeated so many times and the reason that ram is so expensive was that at one time during covid there was shortage so they built more factories and they built so so many that these companies took a hit in stocks so they then went and closed and at just the bottom of their factory production levels, the AI bubble started to pop in and need their ram's and now they are once again increasing factory levels
And after the supply due to AI gets closed with the additional compute etc., I doubt it
I think that within 2-3 or maybe 4 years of timeframe ram will get cheaper imo.
The problem is, if someone can fill the market till that time.
And they never went down to pre-crypto pricing. For quite a while, Intel was the only company producing reasonably specced GPUs at somewhat reasonable prices.
> You don't think that as prices go up, the supply might also go up, and the equilibrium price will be maintained?
This assumes infinite and uniformly distributed resources as well as no artificial factors such as lobbying, corruption, taxation or legislation which might favour one entity over the other.
The dream of free market exists only in a vacuum free from the constraints of reality.
In the end, with the current market prices, chips factories and data centers are being built all over with the assumption of exponential demand growth. When the excitement and demand for AI cools, we will enjoy the additional capacity and better prices. Also see: fiber bandwidth post 2000. Capital poured in, overbuilding happened, prices collapsed after the crash.
Has work begun on increasing RAM production capacity? My understanding is that these companies are specifically _not_ increasing capacity yet while they wait to see if the bubble bursts or not.
Rational actors in game know that this demand spike is most likely temporary. So investing in more production only to face glut in future dropping margins to nearly nothing is not rational move.
This has been played out before so it is only natural that they are careful with increasing the supply. And while they don't response they are netting larger margins than before.
Obvious end result is that demand will drop as price goes up. The other natural part of supply-demand curve.
If the market is big enough, competitors will appear. And if the margins are high enough, competitors can always price-compete down to capture market-share.
Competitors will appear? You can't build a DRAM production facility in a year. You probably even can't in two years.
Also, "price-compete down to capture market-share"? Prices are going up because all future production capacity has been sold. It makes no sense to lower prices if you don't have the capacity to full fill those orders.
> Rare earth metals are in the dirt around the world.
They are. The problem is, the machinery to extract and refine them, and especially to make them into chips, takes years to build. We're looking at a time horizon of almost a decade if you include planning, permits and R&D.
And given that almost everyone but the AI bros expects the AI bubble to burst rather sooner than later (given that the interweb of funding and deals more resembles the Habsburg family tree than anything healthy) and the semiconductor industry is infamous for pretty toxic supply/demand boom-bust cycles, they are all preferring to err on the side of caution - particularly as we're not talking about single billion dollar amounts any more. TSMC Arizona is projected to cost 165 billion dollars [1] - other than the US government and cash-flush Apple, I don't even know anyone able, much less willing to finance such a project under the current conditions.
Apple at least can make use of TSMCs fab capacity when the AI bros go bust...
Chips also need rare doping materials, plus an absurd level of purity for the silicon. The problems are the same no matter if we're talking about chips or batteries.
It's not even the economical theory. Supply should not increase to increased demand. They want more profit, and if less supplies is what accomplishes that, they will absolutely keep the supplies constant and manufacture a scarcity. This is the economical theory.
As far as I can tell, none of the companies producing memory chips are increasing production because they don't know if the current demand is sustainable.
Increasing memory production capacity is a multi-year project, but in a few years, the LLM companies creating the current demand might all have run out of money. If demand craters just as supply increases, prices will drastically decrease, which none of these companies want.
You are wrong. Memory production is being expanded in 2026 and will expand further in 2027 and 2028 as the memory suppliers catch up on fab shell capacity.
This is exactly it: supply of high margin products is increasing at the cost of low margin products. Expect the low end margin to catch up to the high end as long as manufacturing capacity is constrained (at least 1 year).
No just stop being cynical. The reason almost every electronic item is cheaper now than 2 decades back is just becuase the demand(and thus supply) is higher.
I can't tell if this is sarcasm or not. I'd argue it's more the result of the CCP bankrolling the Chinese electronics industry to the point where roughly 70% of all electronics goods are produced in China. The concentration of expertise and supply chains is staggering, and, imo, was really born out of geopolitical strategy.
> of the CCP bankrolling the Chinese electronics industry to the point where roughly 70% of all electronics goods are produced in China.
But we don't see this bankrolling in absolute values. Rather, it's due to regressive taxation, low (cheap) social security for workers, and very weak intellectual property protection.
No, its not. Transistors used to cost $1. Now they cost $1/billion or something. It's all because the 10s of billions of fixed cost incurred by fab is shared among customers. If we create less chips, the fixed cost wont reduce.
Without regulation, money begets money and monopolies will form.
If the American voter base doesn't pull its shit together and revive democracy, we're going to have a bad century. Yesterday I met a man who doesn't vote and I wanted to go ape-shit on him. "My vote doesn't matter". Vote for mayor. Vote for city council. Vote for our House members. Vote for State Senate. Vote for our two Senators.
"Voting doesn't matter, capitalism is doomed anyway" is a self-fulling prophecy and a fed psy-op from the right. I'm so fucking sick of that attitude from my allies.
This is a common sentiment but it doesn't make any sense. Voting for the wrong politician is worse than not voting at all, so why is it seen as some moral necessity for everyone to vote? If someone doesn't have enough political knowledge to vote correctly, perhaps they shouldn't vote.
Someone, I can't remember who, explained it better than me, but the gist of it is by not voting, you are effectively checking yourself out of politician consideration.
If we see politician as just a machine who's only job is to get elected, they have to get as many votes as possible. Pandering to the individual is unrealistic, so you usually target groups of people who share some common interest. As your aim is to get as many votes as possible, you will want to target the “bigger” (in amount of potential vote) groups. Then it is a game of trying to get the bigger groups which don't have conflicting interest.
While this is theory and a simplification of reality, all decent political party do absolutely look at statistics and survey to for a strategy for the election.
If you are part of a group that, even though might be big in population, doesn't vote, politician have no reason to try to pander to you. As a concrete example, in a lot of “western” country right now, a lot of politician elected are almost completely ignoring the youth. Why ? Because in those same country the youth is the age group which vote the less.
So by not voting, you are making absolutely sure that your interest won't be defended. You can argue that once elected, you have no guarantee that the politician will actually defend your interest, or even do the opposite (as an example, soybean farmer and trump in the U.S). But then you won't be satisfied and possibly not vote for the same guy / party next election (which is what a lot of swing voters do).
But yeah, in an ideal world, everyone would vote, see through communication tactics and actually study the party, program and the candidate they vote for, before voting.
Jovially -- you simultaneously believe that they're a victim of a psy-op *and* that their attitude is self formed?
;) And you wanted to go ape shit on him... For falling for a psy-op?
My friend, morale is very very low. There is no vigor to fight for a better tomorrow in many people's hearts. Many are occupied with the problems of today. It doesn't take a psy-op to reach this level of hopelessness.
Be sick of it all you want, it doesn't change their minds. Perhaps you will find something more persuasive.
Just so we're clear, the voter base of over a year ago asked for this because they were actively lied to, and were foolish enough to believe said lies.
Current polling however says the current voter base is quite unhappy with how this is
It is very likely that his vote for the parliament literally and legally doesn't matter, depending on the party allegiance of the candidates and the state he is in. All because of the non-democratic ancient first past the post system. Though in his place I would go to the station and at least deface a ballot as a sign of contempt.
> Without regulation, money begets money and monopolies will form.
Ahem, you'll find that with regulation, money begets money and monopolies will form. That is, unless you magically produce legislators which are incorruptible, have perfect knowledge and always make the perfect choice.
Even the Big Bang was imperfect, and matter clumps together instead of being perfectly distributed in the available space.
What regulation are you expecting to be passed and why do you believe monopolies are bad?
If a monopoly appears due to superior offerings, better pricing and quicker innovation, I fail to see why it needs to be a bad thing. They can be competed against and historically that has always been the case.
On the other hand, monopolies appearing due to regulations, permissions, patents, or any governmental support, are indeed terrible, as they cannot be competed against.
Calling the greatest and last invention of man "AI thingies" is telling of why our society will split into tech and non tech communities in the future like all the science fiction authors have predicted.
There are several inventions which are far greater than LLMs. To name two: computers and methods to generate electricity, things without which LLMs wouldn’t have been possible. But also harnessing fire, the wheel, agriculture, vaccines… The list goes on and on.
Calling LLMs “AI thingies” seems much more in tune with reality than calling them “the greatest invention of man” (and I’m steel manning and assuming you meant “latest”, not “last”). You can’t eat LLMs or live in them and they are extremely dependent on other inventions to barely function. They do not, in any way, deserve the title of “greatest invention”, and it’s worrying that we’re at that level of hyperbole. Though you’re certainly not the first one to make that claim.
For last 2 years, I've noticed a worrying trend: the typical budget PCs (especially Laptops) are being sold at higher prices with lower RAM (just 8GB) and lower-end CPUs (and no dedicated GPUs).
Industry mandate should have become 16GB RAM for PCs and 8GB for mobile, since years ago, but instead it is as if computing/IT industry is regressing.
New budget mobiles are being launched with lower-end specs as well (e.g., new phones with Snapdragon Gen 6, UFS2.2). Meanwhile, features that were being offered in budget phones, e.g., wireless charging, NFC, UFS3.1 have silently been moved to the premium mobile segment.
Meanwhile the OSes and software are becoming more and more complex, bloated and more unstable (bugs) and insecure (security loopholes ready for exploits).
It is as if the industry has decided to focus on AI and nothing else.
And this will be a huge setback for humanity, especially the students and scientific communities.
This is what I find a bit alarming, too. My M3 Max MacBook Pro takes 2 full seconds to boot Slack, a program that used to literally be an IRC client. Many people still believe client-side compute is cheap and worrying about it is premature optimization.
Of course, some performance-focused software (e.g. Zed) does start near-instantly on my MacBook, and it makes other software feel sluggish in comparison. But this is the exception, not the rule.
Even as specs regress, I don't think most people in software will care about performance. In my experience, product managers never act on the occasional "[X part of an app] feels clunky" feedback from clients. I don't expect that to change in the near future.
No dedicated GPU is certainly unrelated to whatever been happening for last two years.
It's just in last 5 years integrated GPUs become good enough even for mid-tier gaming let alone running browser and hw accel in few work apps.
And even before 5 years ago majority of dedicated GPUs in relatively cheap laptops was garbage barely better than intrgrated one. Manufacturers mostly put them in there for marketing of having e.g Nvidia dGPU.
A dedicated GPU is a red flag for me in a laptop. I do not want the extra power draw or the hybrid graphics sillyness. The Radeon Vega in my ThinkPad is surprisingly capable.
Dedicated GPUs in gaming laptops are a necessity for the IT industry, as it forces manufacturers, assemblers and software makers to be more creative and ambitious with power draw and graphics software, and better optimal usage of available hardware resources (e.g., better battery and different performance modes to compensate for the higher power consumption due to the GPU; so a low-power mode enabled by casual user will disable the dedicated GPU and make the OS and apps dependent on the integrated GPU instead, but same/another user using same PC can switch to dedicated GPU when playing a game or doing VFX or modeling).
Without dedicated GPUs, we consumers will get only weaker hardware, slower software and the slow death of graphics software market. See the fate of Chromebooks market segment - it is almost dead, and ChromeOS itself got abandoned.
Meanwhile, the same Google which made ChromeOS as a fresh alternative OS to Windows, Mac and Linux, is trying to gobble the AI market. And the AI race is on.
And the result of all this AI focus and veering away from dedicated GPUs (even by market leader nVidia, which is no longer having GPUs as a priority) is not only the skyrocketing price hikes in hardware components, but also other side effects. e.g., new laptops are being launched with NPUs which are good for AI but bad for gaming and VFX/CAD-CAM work, yet they cost a bomb, and the result is that budget laptop market segment has suffered - new budget laptops have just 8GB RAM, 250GB/500GB SSD, and poor CPU, and such weak hardware, so even basic software (MS Office) struggles on such laptops. And yet even such poor laptops are having a higher cost these days. This kind of deliberate market crippling affects hundreds of millions of students and middle class customers who need affordable yet decent performance PCs.
Yea I agree it's not worth it to have a igpu a dedicated. If I'm correct in what you are talking about. There's always issues with that setup in laptops. But I'd stay away from all laptops at this point until we get an Adminstration that enforces anti trust. All manufactures have been cutting so many corners, your likely to have hardware problems within a year unless it's a MacBook or a business class laptop.
Low end gaming ia 2d indie titles and they now run on toasters.
All the popular mass matket games work on iGPU: fortnite, roblox, mmos, arena shooters, battlen royales. Good chunk of cross platform console titles also work just fine.
You can play Cyberpunk or BG3 on damn Steam Deck. I wont call this low end.
Number of games that dont run to some extent without dGPU is limited to heavy AAA titles and niche PC only genres.
I'm gonna be honest thats not my experience at all. I got a laptop with a modern ryzen 5 CPU four years ago that had an iGPU because "its good enough for even mid-tier gaming!" and it was so bad that I couldn't play 1440p on youtube without it skipping frames. Tried parsec to my desktop PC and it was failing that as well. I returned it and bought a laptop with a nvidia dGPU (low end still, I think it was like a 1050-refresh-refresh equivalent) and haven't had any of those problems. That AMD Vega gpu just couldn't do it.
Or, we can expect better from software. Maybe someone can fork Firefox and make it run better, hard cap how much a browser window can use.
The pattern of lazy almost non existent optimization combined with blaming consumers for having weak hardware, needs to stop.
On my 16GB ram lunar lake budget laptop CachyOS( Arch) runs so much smoother than Windows.
This is very unscientific, but using htop , running Chrome/YouTube playing music, 2 browser games and VS code having Git Copilot review a small project, I was only using 6GBs of ram.
For the most part I suspect I could do normal consumer stuff( filing paperwork and watching cat videos) on an 8GB laptop just fine. Assuming I'm using Linux.
All this Windows 11 bloat makes computers slower than they should be. A part of me hopes this pushes Microsoft to at least create a low ram mode that just runs the OS and display manager. Then let's me use my computer as I see fit instead of constantly doing a million other weird things.
We don't *need* more ram. We need better software.
> Maybe someone can fork Firefox and make it run better, hard cap how much a browser window can use
Browsing web requires more and more RAM each year but I don't think browsers are the main reason - sites use more and more JS code. With a hard cap many sites will stop working. Software bloat is a natural tendency, the path of least resistance. Trimming weigh requires a significant effort and in case of web - a coordinated effort. I don't believe it could happen unless Google (having a browser with >60% market share) will force this but Google own sites are among worst offenders in term of hardware requirements.
> hopes this pushes Microsoft to at least create a low ram mode
Windows OS and Surface (CoPilot AI-optimized) hardware have been combined in the "Windows + Devices" division.
> We don't *need* more ram
RAM and SSDs both use memory wafers and are equally affected by wafer hoarding, strategic supply reductions and market price manipulation.
Nvidia is re-inventing Optane for AI storage with higher IOPS, and paid $20B for Groq LPUs using SRAM for high memory bandwidth.
The architectural road ahead has tiers of memory, storage and high-speed networking, which could benefit AI & many other workloads. How will industry use the "peace dividend" of the AI wars? https://www.forbes.com/sites/robtoews/2020/08/30/the-peace-d...
The rapid growth of the mobile market in the late 2000s and early 2010s led to a burst of technological progress.. core technologies like GPS, cameras, microprocessors, batteries, sensors and memory became dramatically cheaper, smaller and better-performing.. This wave of innovation has had tremendous second-order impacts on the economy. Over the past decade, these technologies have spilled over from the smartphone market to transform industries from satellites to wearables, from drones to electric vehicles.
> Or, we can expect better from software. Maybe someone can fork Firefox and make it run better, hard cap how much a browser window can use.
You can already do this. For example, I use `systemd-run` to run browsers with CPU quotas applied. Firefox gets 400% CPU (i.e. up to 4 cores), and no more.
Example command: systemd-run --user --scope -p CPUQuota=400% firefox
You can impose CPU restrictions in Windows 10 or 11 too...
You can limit CPU usage for a program in Windows by adjusting the "Maximum processor state" in the power options to a lower percentage, such as 80%. Additionally, you can set the program's CPU affinity in Task Manager. Please note this will only affect the process scheduling.
You can also use a free tool like Process Lasso or BES to limit the CPU for a Windows application. You can use a free tools like HWInfo, SysInternals (ProcMon, SysMom, ProcDump) to monitor and check for CPU usage, especially to investigate CPU spikes caused by rogue (malware or poor performance) apps.
CPU affinity?
I haven't been able to change priority in task manager since window 8 i think. Cpu affinity seems only to allow which cores get assigned...not really good management.
Insane that this is seen as "better software". I could do basically the same functionality in 2000 with 512mb. I assume this is because everything runs through chrome with dozens more layers of abstraction but
512MB weren't that odd for multimedia from 2002, barely a few years later.
By 2002 256MB of RAM were the standard, almost a new low-end PC.
64MB = w98se OK, XP will swap a lot on high load, nixlikes really fast with fvwm/wmaker and the like. KDE3 needs 128MB to run well, so get a bit down. No issues with old XFCE releases. Mozilla will crawl, other browsers will run fine.
128MB = w98se really well, XP willl run fine, SP2-3 will lag. Nixlikes
will fly with wmaker/icewm/fvwm/blackbox and the like. Good enough for mozilla.
192MB = Really decent for a full KDE3 desktop or for Windows XP with real life speeds.
256MB = Like having 8GB today for Windows 10, Gnome 3 or Plasma 6. Yes, you can run then with 2GB and ZRAM, but, realistically, and for the modern bloated tools, 8GB for a 1080p desktop it's mandatory. Even with UBlock Origin for the browser. Ditto back in the day. With 256MB XP and KDE3
flied and they ran much faster than even Win98 with 192MB of RAM.
Option A: We do a better job at optimizing software so that good performance requires less RAM than might otherwise be required
Option B: We wish that things were different, such that additional RAM were a viable option like it has been at many times in the past.
Option C: We use our time-benders to hop to a different timeline where this is all sorted more favorably (hopefully one where the Ballchinians are friendly)
---
To evaluate these in no particular order:
Option B doesn't sound very fruitful. I mean: It can be fun to wish, but magical thinking doesn't usually get very far.
Option C sounds fun, but my time-bender got roached after the last jump and the version of Costco we have here doesn't sell them. (Maybe someone else has a working one, but they seem to be pretty rare here.)
That leaves option A: Optimize the software once, and duplicate that optimized software to whomever it is useful using that "Internet" thing that the cool kids were talking about back in the 1980s.
I'm reading this thread on an 11-year-old desktop with 8GB of RAM and not feeling any particular reason to upgrade, although I've priced it out a few times just to see.
Mint 22.x doesn't appear to be demanding any more of my machine than Mint 20.x. Neither is Firefox or most websites, although YouTube chat still leaks memory horrendously. (Of course, download sizes have increased.)
I've been enjoying running Mint on my terrible spec chromebook - it only has 3GB of RAM, but it rarely exceeds 2GB used with random additions and heavy firefox use. The battery life is obscenely good too, I easily break 20 hours on it as long as I'm not doing something obviously taxing.
Modern software is fine for the most part. People look at browsers using tens of gigabytes on systems with 32GB+ and complain about waste rather than being thrilled that it's doing a fantastic job caching stuff to run quickly.
Mint is probably around 0.05% of desktop/laptop users.
I think root comment is looking at the overall picture of what all customers can get for their money, and see it getting worse.
This wasn’t mentioned, but it’s a new thing for everyone to experience, since the general trend of computer hardware is it gets cheaper and more powerful over time. Maybe not exponentially any more, but at least linearly cheaper and more powerful.
The OP is not looking at the static point (of the price of that item), but the trend - ala, the derivative of the price vs quality. It was on a steep upward incline, and now it's flattening.
It is even worse for those of us in Africa. The equivalent phone I can buy for USD$150-250 back home is absolutely shocking in terms of how bad it is in ram ,disk space and often an outdated version of Android. I buy my phones on Amazon US which delivers and I get a much better phone for the same price range.
I think it is the result of specs like RAM and CPU no longer being the selling point it once was, except for gaming PCs. Instead people want thin laptops, good battery life, nice screens, premium materials, etc... We have got to the point where RAM and CPU are no longer a limiting factor for most tasks, or at least until software become bloated enough to matter again.
If you want a powerful laptop for cheap, get a gaming PC. The build quality and battery life probably won't be great, but you can't be cheap without making compromises.
Same idea for budget mobiles. A Snapdragon Gen 6 (or something by Mediatek) with UFS2.2 is more than what most people need.
> Industry mandate should have become 16GB RAM for PCs
it was only less than 10 yrs ago that a high end PC would have this level of ram. I think the last decade of cheap ram and increasing core count (and hz) have spoiled a lot of people.
We are just returning back on trend. May be software would be written better now that you cannot expect the average low budget PC to have 32G of ram and 8 cores.
I wonder what we can do to preserve personal computing, where users, not vendors, control their computers? I’m tired of the control Microsoft, Apple, Google, OpenAI, and some other big players have over the entire industry. The software has increasingly become enshittified, and now we’re about to be priced out of hardware upgrades.
The problem is coming up with a viable business model for providing hardware and software that respect users’ ability to shape their environments as they choose. I love free, open-source software, but how do developers make a living, especially if they don’t want to be funded by Big Tech?
No, this is not it. It only worked when there were a small number of buyers for used hardware, who were largely enthusiasts. The moment it becomes mainstream you're going to face the same scarcity in the used/refurbished market as well.
I think it will be a good thing actually. Engineers, no longer having the luxury of assuming that users have high end system specs, will be forced to actually write fast and efficient software. No more bloated programs eating up RAM for no reason.
The problem is that higher performing devices will still exist. Those engineers will probably keep using performant devices and their managers will certainly keep buying them.
We'll probably end up in an even more bifurcated world where the well off have access to lot of great products and services that most of humanity is increasingly unable to access.
Have the laws of supply and demand been suspended? Capital is gonna pour into memory fabrication over the next year or two, and there will probably be a glut 2-3 years from now, followed by retrenchment and wails that innovation has come to a halt because demand has stalled.
If "performant" devices are not widespread then telemetry will reveal that the app is performing poorly for most users. If a new festure uses more memory and sugnificantly increases the crash rate, it will be disabled.
Apps are optimized for the install base, not for the engineer's own hardware.
Some open source projects use Slack to communicate, which is a real ram hog. Github, especially for viewing large PR discussions, takes a huge amount of memory.
If someone with a low-memory laptop wants to get into coding, modern software-development-related services are incredible memory hogs.
I'm not optimistic that this would be the outcome. You likely will just have poor running software instead. After all, a significant part of the world is already running lower powered devices on terrible connection speeds (such as many parts of Africa).
> a significant part of the world is already running lower powered devices
but you cannot consider this in isolation.
The developed markets have vastly higher spending consumers, which means companies cater to those higher spending customers proportionately more (as profits demand it). Therefore, the implication is that lower spending markets gets less investment and less catered to; after all, R&D spending is still a limiting factor.
If the entirety of the market is running on lower powered devices, then it would get catered for - because there'd be no (or not enough) customers with high powered devices to profit off.
I don't think it's going to happen in this day and age. Some smart people will but most barley know how to write there own code let alone write efficient code
I really wonder when the point will be reached at which the South Korean government steps in and starts to take a closer look at the growing long-term supply commitments that companies like OpenAI are indirectly driving with major memory manufacturers such as SK hynix and Samsung Electronics.
Allocating a very large share of advanced memory production, especially HBM and high-end DRAM, which are critical for almost all modern technology (and even many non-tech products like household appliances) to a small number of U.S. centric AI players risks distorting the global market and limiting availability for other industries.
Even within Samsung itself, the Mobile eXperience (MX) Business (smartphones) is not guaranteed preferential access to memory from Samsung’s Device Solutions (DS) Division, which includes the Memory Business. If internal customers are forced to source DRAM elsewhere due to pricing or capacity constraints, this could eventually become economically problematic for a country that relies very heavily on semiconductor and technology exports.
AI has had a net negative on my life and created negative value for me and my family. It's raising my power rate. It's stealing my ideas and creative output. All to enrich a select few assholes who didn't need any more enrichment.
AI embodies everything wrong with our modern gilded age era of capitalism.
Thankfully we're at a stage where a 4 year old second-hand iPhone is perfectly usable, as are any M-series Macs or most Linux laptops. Sucks for anyone needing something particularly beefy for work; but I feel that a lot of purchases can be delayed for at least a year or two while this plays out.
> Sucks for anyone needing something particularly beefy for work; but I feel that a lot of purchases can be delayed for at least a year or two while this plays out.
100%, there is a lost time factor which sucks though if someone wants beefy, some might be forced to cough up some money but honestly yea although I agree with your statement, the ram increases really impact the self hosting/homelabbing genres and there was a recent post relevant to this discussion on hackernews https://news.ycombinator.com/item?id=46416618: Self hosting is being enshittified.
For me, there is concerning a flag about all of this.
I know this is not always true, but on this case, crucial folks say the margins for end user are too low and they have demand for AI.
I suppose they do not intend to bring a new AI focused unit because it is not worth it or they believe the hype might be gone before it they are done.
But what intrigues me is why they would allow other competitors to step up in a segment they dominate?
They could raise the prices for the consumers if they are not worried about competition...
There is a whole "not-exactly" ai industry labeled as AI that received a capital t of money. Is that what they are going for?
So my understanding of the situation is that, Crucial folks had downsized their factory production (see my other comments for reference perhaps) but then AI involvement started demanding chips just at the time their factory production was at their lows.
So now for these AI companies, they got tons of money to burn so they are willing to pay a lot more, so now crucial only have a limited supply of ram and the thing is there isn't much difference between AI chip and consumer chip but the margins of AI chip are super higher compared to consumer chip
So earlier they would sell consumer chips and AI chips as well but then the AI companies still demanded even more and they would get insane profits selling them so what they did (atleast crucial) is that they stopped selling consumer chips just to sell AI chips for profit.
> I suppose they do not intend to bring a new AI focused unit because it is not worth it or they believe the hype might be gone before it they are done. But what intrigues me is why they would allow other competitors to step up in a segment they dominate? They could raise the prices for the consumers if they are not worried about competition...
Well as a consumer, I certainly hope so but I think that these companies did this case because their have been times they were -55% in stock prices and its just cash making money device at this point and there is a monopoly of fabs with just three key players.
So the answer to your question is "money" and "more money" short term. Their stock prices are already up I think and a company really loves short term rising stock prices
> They could raise the prices for the consumers if they are not worried about competition
Well, would you increase the prices 3-4x? Because supposedly thats how much the AI chips from what I've heard are... And due to this, the second hand market itself is selling these at a close-enough mark.
I don't know but I hope that new players come in the market, I didn't know that this ram industry was such monopolistic with there being only 3 key players and how that became a chokehold for the whole world economy in a way
> Well, would you increase the prices 3-4x?
(Text below is quite long, let me say it here, you made great points in your answer!)
It seems they could. They not only single handed caused it to double or more without trying :/
Not sure if it would trigger other sorts of regulatory issues though
It is my impression that there is a fabricated scarcity of all goods.
That's a common practice in cloth retailers. In the 90s they thought for brand name and market share.
They noticed it was silly because they could sell half for double of the price and as this means less logistics, it also meant higher margins.
It is not a lunch free approach. Selling less means that you delegate at least the bottom portion of your clients to the market, and if there are options, they might just be gone.
That's exactly what happened with Chevrolet, Ford, etc. they stopped investing and when a new competitor appeared, even if it was more marketing than product, they lots rivers of money and barely can keep the fight on (except for maybe making a puppet tell others that there is no such thing as climate change, but that's something else)
Technology space right now looks like it. We already see major brands stagnation allegedly because they did all that is possible and it will take some time until some nouvelle approach to appear.
As a consumer, I want to believe this won't take long to settle but I'm afraid money is going elsewhere
Building a dedicated AI-focused consumer line is risky: long development cycles, uncertain demand, and the chance that today's hype cools before the product ships
I understand the issue with all the devices. But what about the rest of the things that depend on these electronics, especially DRAMs? Automotive, Aircraft, Marine vessels, ATC, Shipping coordination, traffic signalling, rail signalling, industrial control systems, public utility (power, water, sewage, etc) control systems, transmission grid control systems, HVAC and environment control systems, weather monitoring networks, disaster altering and management systems, ticketing systems, e-commerce backbones, scheduling and rostering systems, network backbones, entertainment media distribution systems, defense systems, and I don't know what else. Don't they all require DRAMs? What will happen to all of them?
Industrial microcontrollers and power electronics use older process nodes, mostly >=45nm. These customers aren’t competing for wafers from the same fabs as bleeding edge memory and TPUs.
Okay, but what about the rest? The ones that aren't embedded in someway and use industrial grade PCs/control stations? Or ones with large buffers like network routers? I'm also wondering about the supply of the alternate nodes and older technologies. Will the manufactures keep those lines running? Was it micron that abandoned the entire retail market in favor of supplying the hyperscalers?
> The ones that aren't embedded in someway and use industrial grade PCs/control stations? Or ones with large buffers like network routers?
Not sure if they require DDR5 but the AI crisis just caused the prices of DDR5 to rise but the market supply of DDR4 thus grew and that's why they got more expensive too
> I'm also wondering about the supply of the alternate nodes and older technologies.
I suppose these might be chinese companies but there might be some european/american companies (not sure) but if things continue, there is gonna be a strain on them in demand and they might increase their prices too
> Was it micron that abandoned the entire retail market in favor of supplying the hyperscalers?
That might be the case only for the infotainment system, but there’s usually many other ECUs in an EV. The ADAS ECUs are carrying similar amounts as an iPhone or the infotainment system. Telematics is also usually also a relatively complex one, but more towards lower sized amounts.
Then you have around 3-5 other midsized ECUs with relatively high memory sizes, or at least enough to require MMUs and to run more complex operating systems supporting typical AUTOSAR stacks.
And then you have all the small size ECUs controlling all small individual actuators.
But also all complex sensors like radars, cameras, lidars carry some amounts of relevant memory.
I still think your point is valid, though. There’s no difference in orders of magnitude when it comes to expensive RAM compared to an iPhone. But there’s cars also carried lots of low-speed, automotive grade memory in all the ECUs distributed throughout the vehicle.
Okay, accepted. But are you sure that the supply won't be a problem as well? I mean, even if these products choose a different process nodes compared to the hyperscalers, will the DRAM manufactures even keep those nodes running in favor of these industries?
What will probably happen is that the reselling market/2nd market of these might probably rise
> will the DRAM manufactures even keep those nodes running in favor of these industries?
Some will, Some might not, In my opinion, the longevity of these brands will only depend if they allow buying ram for the average person/consumer brands so I guess we might see new competition perhaps or give more marketshare to all the other fab companies beyond the main three of these industries.
I am sure that some company will 100% align with the consumers but the problem to me feels that they wouldn't be able to supply enough production to consumers in the first place so prices still might rise.
And those prices most likely will be paid by you in one form or another but it would be interesting to see how long the companies who buy dram from these providers or build datacenters or anything ram intensive will hold their price up, perhaps they might eat the loss short term similar to what we saw some companies do during trump tarrifs.
We’ve been able to hold the same price we had at launch because we had buffered enough component inventory before prices reached their latest highs. We will need to increase pricing to cover supplier cost increases though, as we recently did on DDR5 modules.
Note that the memory is on the board for Ryzen AI Max, not on the package (as it is for Intel’s Lunar Lake and Apple’s M-series processors) or on die (which would be SRAM). As noted in another comment, whether the memory is on the board, on a module, or on the processor package, they are all still coming from the same extremely constrained three memory die suppliers, so costs are going up for all of them.
Longer contracts are riskier. The benefit of having cheaper RAM when prices spike is not strong enough to outweigh the downside of paying too much for RAM when prices drop or stay the same. If you’re paying a perpetual premium on the spot price to hedge, then your competitors will have pricing power over you and will slowly drive you out of the market. The payoff when the market turns in your favor just won’t be big enough and you might not survive as a business long enough to see it. There’s also counterparty risk, if you hit a big enough jackpot your upside is capped by what would make the supplier insolvent.
All your competitors are in the same boat, so consumers won’t have options. It’s much better to minimize the risk of blowing up by sticking as closely to spot at possible. That’s the whole idea of lean. Consumers and governments were mad about supply chains during the pandemic, but companies survived because they were lean.
In a sense this is the opposite risk profile of futures contracts in trading/portfolio management, even though they share some superficial similarities. Manufacturing businesses are fundamentally different from trading.
They certainly have contracts in place that cover goods already sold. They do a ton of preorders which is great since they get paid before they have to pay their suppliers. Just like airlines trade energy futures because they’ve sold the tickets long before they have to buy the jet fuel.
the risk is that such longer contracts would then lock you into a higher cost component for longer, if the price drops. Longer contracts only look good in hindsight if ram prices increased (unexpectedly).
If you’re Apple, maybe that works, in this case we’re seeing 400% increases in price, instead of your RAM you’ll be delivered a note to pay up or you’ll get your money back with interest and termination fees and the supplier is still net positive.
> Question: are SoCs with on die memory be effected by this?
SoCs with on-die memory (which is, these days, exclusively SRAM, since I don't think IBM's eDRAM process for mixing DRAM with logic is still in production) will not be effected. SiPs with on-package DRAM, including Apple's A and M series SiPs and Qualcomm's Snapdragon, will be effected -- they use the same DRAM dice as everyone else.
The aforementioned Ryzen AI chip is exactly what you describe, with 128 GB on-package LPDDR5X. I have two of them.
To answer the original question: the Framework Desktop is indeed still at the (pretty inflated) price, but for example the Bosgame mini PC with the same chip has gone up in price.
Apple secured at least a year-worth supply of memory (not in actual chips but in prices).
The bigger the company = longer the contract.
However it will eventually catch up even to Apple.
It is not prices alone due to demand but the manufacturing redirection from something like lpddr in iphones to hbm and what have you for servers and gpu
To be honest, it starts to look more and more like a single company (we all know which one), is just buying up all DRAM capacities to keep others out of the (AI) game.
I have a feeling every single supplier of DRAM is going to be far more interested in long-term contracts with Apple than with (for example) OpenAI, since there's basically zero possibility Apple goes kaput and reneges on their contracts to buy RAM.
I would think so because fab capacity is constrained, and if you make an on-die SoC with less memory, it uses fewer transistors, so you can fit more on a wafer.
Outside say video and image editing and maybe lossless audio. Why is this much ram even needed in most use cases? And I mean actually thinking about using it. Computer code unless you are actually doing whole Linux kernel, is just text. So lot of projects probably would fit in cache. Maybe software companies should be billed for user's resources too...
Electron apps hog memory. The vast, vast majority of computer users are Windows users. Using 8 GB of memory without really "using" it, is trivial. Chrome + some Microsoft office apps will spend that much.
I have multiple apps using 300 GB+ PostgreSQL databases. For some queries, high RAM is required. I enable symmetrical NVMe swaps, too. Average Joe with gaming needs wouldn't need more than 64 GB for a long time. But for the database, as the data grows, RAM requirements also grow. I doubt my situation is relatable to many.
One might think so but the AI companies actually lose a lot and I mean a lot of money in these deals.
Even if they might sell their inference and everything, they still wouldn't be that much profitable.
So like, the key point is that a ram company can supply openAI ram and get some really high quick bucks which would be even more than if they were to create thier own datacenters,run open source models in them, provide inference in say open router.
Now you might ask: Is openAI or these AI companies mad for burning so much money?
No, because if you have the capacity to make e.g. ram chips it makes more economic sense to sell them for ai bucks. Serving the other market segment is an opportunity cost unless you're selling to them at same prices. In the long run though if enough players emerge the price will eventually come down just due to oversupply.
Not quite. Making specialized DRAM chips for AI hardware needs, requires high tech components. Making low(er) end DRAM chips for consumer needs might be easier to get started with.
I am pretty sure, in the next year we will see a wave of low end ram components coming out of china.
I believe we're in the middle of a "perfect storm" and AI is not the only one to blame:
- Right now A LOT of PCs are getting out of date because Windows 11 wants some new hardware that's missing in older PCs.
- At the same time, smartphones were starting to use and need more memory.
- Modern cars (mostly electric) have much more electronics inside than older ones... they're now distributed systems with several CPUs working together along a bus.
- The cloud needs to upgrade and newer servers have much more memory than older ones with almost the same foorprint (wich means you need to lease less datacenter space).
- And GPUs for AI are in demand and need RAM.
But only AI is to blame although we're living in a perfect storm for RAM demand.
Software has gotten bad over the last decade. Electron apps were the start but these days everything seems to be so bloated, right from the operating systems to browsers.
There was a time when apple was hesitant to add more ram to its iPhones and app developers would have to work hard to make apps efficient. Last few years have shown Apple going from 6gb to 12gb so easily for their 'AI' while I consistently see the quality of apps deteriorating on the App Store. iOS 26 and macOS 26 are so aggressive towards memory swapping that loading settings can take time on devices with 6gb ram (absurd). I wonder what else they have added that apps need purging so frequently. 6gb iphone and 8gb M1 felt incredibly fast for the couple of years. Now apparently they are slow like they are really old.
Windows 11 and Chrome are a completely different story. Windows 10 ran just fine on my 8th gen pc for years. Windows 11 is very slow and chrome is a bad experience. Firefox doesn't make it better.
I also find that gnome and cosmic de are not exactly great at memory. A bare minimum desktop still takes up 1.5-1.6gb ram on a 1080p display and with some tabs open, terminal and vscode (again electron) I easily hit 8gb. Sway is better in this regard. I find alacrity sway and Firefox together make it a good experience.
I wonder where we are heading on personal computer software. The processors have gotten really fast and storage and memory even more so, but the software still feels slow and glitchy. If this is the industry's idea of justifying new hardware each year we are probably investing in the wrong people.
Firefox is set to allocate memory until a certain absolute limit or memory pressure is reached. It will eat memory whether you have 4GB of RAM of 40GB.
Set this to something you find reasonable: `browser.low_commit_space_threshold_percent`
And make sure tab unloading is enabled.
Also, you can achieve the same thing with cgroups by giving Firefox a slice of memory it can grow into.
Fair. I installed a MIDI composition app recently that was 1.2 GB! Now, it does have some internal synthesis that like uses samples, but only a limited selection of sounds so I think 95% of the bulk is from Electron.
I for one don't remember software ever being very good. Windows XP crashed all the time and needed frequent formats due to all the viruses (all antivirus software) that made it unusable. Win 2k was even worse. Old MacOS's also crashed frequently. Old Linux's had barely any hardware support. In fact while I agree that software is more bloated than ever, it also seems more stable than ever thinking back to what was before.
This is where government can jump in and take advantage of the less supply by opening large semi conductor factories quickly and helping people in avoiding this bottlneck.
For a while CPUs kept getting significantly faster every year. Then ram kept getting bigger. Then NVMe came along and brought 200-1000x IOPS.
So writing optimized software was niche and overshadowed by the huge gains of hardware. People and corporations didn't care. They preferred fast feature delivery. Even when optimizing techniques like multi-tenant servers, we ended up having heavy containers wasting RAM and resources. And most apps switched to web frameworks like Electron where each one has its own huge web browser. Most people didn't care.
I hope this shortage has a silver lining of untangling the mess and refactoring the sea of bloat. But I know it's more likely some trick will be found to just patch it up a bit and only reduce the pain up to the new equilibrium level. For example something idiotic like Electron sharing resources across multiple apps and taking huge security risks. Corporations love to play false dichotomies to save pennies.
It is true that (most) people don't care, but...
... if you're beign charged by RAM (CPU is essentially free) ...
... AND you develop an application that's also your source of income
... THEN memory usage matters.
It feels like a weird tension: we worry about AI alignment but also want everyone to have unrestricted local AI hardware. Local compute means no guardrails, fine-tune for whatever you want.
Maybe the market pricing people out is accidentally doing what regulation couldn't? Concentrating AI where there's at least some oversight and accountability. Not sure if that's good or bad to be honest.
For now. Chinese supply chains include DRAM from CXMT (sanctioned) and NAND from YMTC (not sanctioned, holds patents on 3D stacking that have been licensed by Korean memory manufacturers).
DRAM spot prices are something like what they were 4 years ago. Having RAM for cheap is nice. But it doesn't cost an extraordinary amount. I recently needed some RAM and was able to pick up 16x32 DDR4 for $1600. That's about twice as expensive as it used to be but $1600 is pretty cheap for 512 GiB of RAM.
A 16 GiB M4 Mac Mini is $400 right now. That covers any essential use-case which means this is mostly hitting hobbyists or niche users.
Server stuff. Nothing interesting. Supermicro H11 + Epyc 7xxx + RAM. I have a 6x4090 setup for local LLMs and I got myself a 128 GB M4 Max laptop thinking I'd do that, but if I'm being honest I need to get rid of that hardware. It's sitting idle because the SOTA ones are so much better for what I want.
Hey sorry, I didn't knew that. I had watched the short form content (https://www.youtube.com/shorts/eSnlgBlgMp8) [Asus is going to save gaming] and I didn't knew that It was a rumour.
Asus doesn't make RAM. That's the whole problem: there are plenty of RAM retail brands, but they are all just selling products that originate from only a couple of actual fabs.
Do they make DRAM? I thought they made compute chips mostly.
If I recall correctly, RAM is even more niche and specialized than the (already quite specialized) general chip manufacturing. The structure is super-duper regular, just a big grid of cells, so it is super-duper optimized.
They (GF) do not make DRAM. They might have an eDRAM process inherited from IBM, but it would not be competitive.
You’re correct that DRAM is a very specialized process. The bit cell capacitors are a trench type that is uncommon in the general industry, so the major logic fabs would have a fairly uphill battle to become competitive (they also have no desire to enter the DRAM market in general).
So far all I am seeing is an increase in prices, so any company claiming it will "ramp up production" here is, in my opinion, just lying for tactical reasons.
Governments need to intervene here. This is a mafia scheme now.
I purchased about three semi-cheap computers in the last ~5 years or so. Looking at the RAM prices, the very same units I bought (!) now cost 2.5x as much as before (here I refer to my latest computer model, from 2 years ago). This is a mafia now. I also think these AI companies should be extra taxed because they cause us economic harm here.
Taxed extra is a good idea. But they bought our current administration so we all know that's not going to happen unless something big happens like Trump gets impeached, and all the criminals in congress go to prison. I'm wondering how likely that will happen. people need to get more directly involved in putting pressure on senators.
Maybe this is the free market working as intended -- did you know that RAM is actually a luxury item, like a Rolls Royce, and most plebes should just make do with 4gb machines, because that is the optimum solution!
Positive downstream effect: The way software is built will need to be rethought and improved to utilize efficiencies for stagnating hardware compute. Think of how staggering the step from the start of a console generation to the end used to be. Native-compiled languages have made bounding leaps that might be worth pursuing again.
Alternatively, we'll see a drop in deployment diversity, with more and more functionality shifted to centralised providers that have economies of scale and the resources to optimise.
E.g. IDEs could continue to demand lots of CPU/RAM, and cloud providers are able to deliver that cheaper than a mostly idle desktop.
If that happens, more and more of its functionality will come to rely on having low datacenter latencies, making use on desktops less viable.
Who will realistically be optimising build times for usecases that don't have sub-ms access to build caches, and when those build caches are available, what will stop the median program from having even larger dependency graphs.
I’d feel better about the RAM price spikes if they were caused by a natural disaster and not by Sam Altman buying up 40% of the raw wafer supply, other Big Tech companies buying up RAM, and the RAM oligopoly situation restricting supply.
This will only serve to increase the power of big players who can afford higher component prices (and who, thanks to their oligopoly status, can effectively set the market price for everyone else), while individuals and smaller institutions are forced to either spend more or work with less computing resources.
The optimistic take is that this will force software vendors into shipping more efficient software, but I also agree with this pessimistic take, that companies that can afford inflated prices will take advantage of the situation to pull ahead of competitors who can’t afford tech at inflated prices.
I don’t know what we can do as normal people other than making do with the hardware we have and boycotting Big Tech, though I don’t know how effective the latter is.
Does this mean developers will have to make their software work on hardware with 8GB of RAM and stop blindly assuming every person in the world gets a packed developer workstation like they do?
Well thank th FSM that the article opens right up with buy now! No thanks, I'm kind of burnt out on mindless consumerism, I'll go pot some plants or something.
I highly recommend disabling javascript in your browser.
Yes, it makes many sites "look funny", or maybe you have to scroll past a bunch of screen sized "faceplant" "twitverse" and "instamonetize" icons, but, there are far fewer ads (like none).
And of course some sites won't work at all. That's OK too, I just don't read them. If it's a news article, its almost always available on another site that doesn't require javascript.
I would not be able to handle that due to video streaming, web clients for things like email, etc. And some sites I trust (including HN) provide useful functionality with JS (while degrading gracefully).
But I use NoScript and it is definitely a big help.
I whole-heartedly agree with your recommendation and join in encouraging more adopters of this philosophy and practice.
Life online without javascript is just better. I've noticed an increase in sites that are useful (readable) with javascript disabled. Better than 10 years ago, when broken sites were rampant. Though there are still the lazy ones that are just blank pages without their javascript crutch.
Maybe the hardware/resource austerity that seems to be upon us now will result in people and projects refactoring, losing some glitter and glam, getting lean. We can resolve to slim down, drop a few megs of bloat, use less ram and bandwidth. It's not a problem; it's an opportunity!
In any case, Happy New Year! [alpha preview release]
Micron is exiting direct to consumer sales. That doesn't mean their chips couldn't end up in sticks or devices sold to consumers, just that the no-middleman Crucial brand is dead.
Also, even if no Micron RAM ever ended up in consumer hands, it would still reduce prices for consumers by increasing the supply to other segments of the market.
I've been ruminating on this past two years, with life before AI most of the compute staying cheap and pretty much 90% idle , we are finally getting to the point of using all of this compute. We probably will find more algorithms to improve efficiency of all the matrix computations, and with AI bubble same thing will happen that happened with telecom bubble and all the fiber optic stuff that turned out to be drastically over provisioned. Fascinating times!
Imagine someone goes to the supermarket and buys all the tomatoes. Then supermarket owner says I don’t know, he bought all at once so it is a better sale. And he sells the remaining 10% of tomatoes at a huge markup
Tulips were by my understanding more so NFTs. Rich people gambling when bored. With promises for tulips in future... Future contracts for tulips. And prices were high because they were insanely rich merchants.
The RAM looks like cornering market. Probably something OpenAI should be prosecuted for if they end up profiting from it.
Except it's still sitting idle in warehouses while datacenters get built. They aren't running yet. Unlike with fiber, GPUs degrade rapidly with use, and for now datacenters need to be practically rebuilt to fit new generations, so we shouldn't expect much reusable hardware to come from this
I now consider this a mafia that aims to milk us for more money. This includes all AI companies but also manufacturers who happily benefit from this. It is a de-facto monopoly. Governments need to stop allowing this milking scheme to happen.
"Monopoly" means one seller, so you can't say multiple X makes a monopoly and make sense. You probably mean collusion.
If demand exceeds supply, either prices rise or supply falls, causing shortages. Directly controlling sellers (prices) or buyers (rationing) results in black markets unless enforcement has enough strength and integrity. The required strength and integrity seems to scale exponentially with the value of the good, so it's typically effectively impossible to prevent out-of-spec behavior for anything not cheap.
If everyone wants chips, semiconductor manufacturing supply should be increased. Governments should subsidize domestic semiconductor industries and the conditions for them to thrive (education, etc.) to meet both goals of domestic and economic security, and do it in a way that works.
The alternative is decreasing demand. Governments could hold bounty and incentive programs for building electronics that last a long time or are repairable or recyclable, but it's entirely possible the market will eventually do that.
> If everyone wants chips, semiconductor manufacturing supply should be increased. Governments should subsidize domestic semiconductor industries and the conditions for them to thrive (education, etc.) to meet both goals of domestic and economic security, and do it in a way that works.
If there is already demand at this inflated price, shouldn’t we ask why more capacity is not coming online naturally first?
Technically it's a lot closer to monopsony (Sam Altman/OAI cornering 40% of the market on DRAM in a clever way for his interests that harms the rest of the world that would want to use it). I keep hoping that somehow necessity will spur China to become the mother of invention here and supply product to serve the now lopsided constrained supply given increasing demand but I just don't know how practical it will be.
It would be understood that any action under the sherman act is unlikely and as you say, the financial penalties are tokenistic.
The non financial parts, which include mandated restructuring and penalties to directors including incarceration however, are not tokenistic. They'd be appealed and delayed, but at some point the shareholders would seek redress from the board. Ignoring judicial mandated instructions isn't really a good idea, current WH behaviour aside. If the defence here is "courts don't matter any more" that's very unhelpful, if true. At some point, a country which cannot enforce judicial outcomes has stopped being civil society.
My personal hope the EU tears holes in the FAANG aside, the collusive pricing of chips has been a problem for some time. The cost/price disjunction here is strong.
At this current pace, if "the electorate" doesn't see real benefits to any of this. 2028 is going to be referendum on AI unfortunately.
Whether you like it or not, AI right now is mostly
- high electricity prices
- crazy computer part prices
- phasing out of a lot of formerly high paying jobs
and the benefits are mostly
- slop and chatgpt
Unless OpenAI and co produce the machine god, which genuinely is possible. If most people's interactions with AI are the negative externalities they'll quickly be wondering if ChatGPT is worth this cost.
> they'll quickly be wondering if ChatGPT is worth this cost
They should be, and the answer is obviously no—at least to them. No political or business leader has outlined a concrete, plausible path to the sort of vague UBI utopia that's been promised for "regular folks" in the bullish scenario (AGI, ASI, etc.), nor have they convincingly argued that this isn't an insane bubble that's going to cripple our economy when AGI doesn't happen—a scenario that's looking more and more likely every day.
There is no upside and only downside; whether we're heading for sci-fi apocalypse or economic catastrophe, the malignant lunatics pushing this technology expect to be insulated from consequences whether they end up owning the future light-cone of humanity or simply enjoying the cushion of their vast wealth while the majority suffers the consequences of an economic crash a few rich men caused by betting it all, even what wasn't theirs to bet.
Everybody should be fighting this tooth and nail. Even if these technologies are useful (I believe they are), and even if they can be made into profitable products and sustainable businesses, what's happening now isn't related to any of that.
I hope they do. We live in a time of incredibly centralized wealth & power and AI and particularly "the machine god" has the potential to make things 100x worse and return us to a feudal system if the ownership and profits all go to a few capital owners.
IMHO this is exactly what is happening. Everyone should be on the phone with there senators putting pressure to enforce anti-trust and deal with citizens united
> At this current pace, if "the electorate" doesn't see real benefits to any of this. 2028 is going to be referendum on AI unfortunately.
Not saying this is necessarily a bad prediction for 2028, but I'm old enough to remember when the 2020 election was going to be a referendum on billionaires and big tech monopolies.
For good measure, a bunch of this is funded through money taken directly from the electorates taxes and given to a few select companies, whose leaders then graciously donate to the latest Ballroom grift. Micron, so greedy they thought nothing of shutting down their consumer brand even when it costs them nothing at all, got $6B in Chips Act money in 2024.
- Ghibli studio style graphics,
- the infamous em-dashes and bullet points
- customer service (just try to use Klarnas "support" these days...)
- Oracle share price ;) - imagine being one of the worlds most solid and unassailable tech companies, losing to your CEOs crazy commitment to the LLMs...
- The internet content - We now tripple check every Internet source we dont know to the core ...
- And now also the chips ?
Where does it stop? When we decide to drop all technology as it is?
It doesn't stop. This is because that it's not the "technology" driving AI. You already acknowledged the root cause: CEOs. AI could be great, but it's currently being propped up by sales hype and greed. Sam wants money and power. Satya and Sundar want money and power. Larry and Jensen want to also cash in on this facade that's been built.
Can LLMs be impactful? For sure. They are now. They're impacting energy consumption, water usage, and technology supply chains in detrimental ways. But that's because these people want to be the ones to sell it. They want to be the ones to cash in. Before they really even have anything significantly useful. FOMO in this C-suite should be punishable in some way. They're all charlatans to different degrees.
Blame the people behind this propping up this mess: the billionaires.
This will scale back when AI replacement attempts slow down as expectations temper (Salesforce, Klarna, etc).
If AI lives up to the hype, it's a portent of how things will feel to the common man. Not only will unemployment be a problem, but prices of any resources desired by the AI companies or their founders will rise to unaffordability.
A lot of AI 'influencers' love wild speculation, but lets ignore the most fantastical claims of techno-singularity, and let's focus on what I would consider a very optimistic scenario for AI companies - that AI capable of replacing knowledge workers can be developed using the current batch of hardware, in the span of a year or two.
Even in this scenario, the capital gains on the lump sump invested in AI far outpaces the money that would be spent on the salaries of these workers, and if we look at the scenario with investor goggles, due to the exponential nature of investment gains, the gap will only grow wider.
Additionally, AI does not seem to be a monopoly, either wrt companies, or geopolitics, so monopoly logic does not apply.
You mean like Sam Altman, who repeatedly claimed AI will cure all cancers and diseases, solve the housing crisis, poverty, and democracy? I was going to add erectile disfunction as a joke, but then realised he probably believes that too.
https://youtu.be/l0K4XPu3Qhg?t=60
It’s hard to point fingers at “AI influencers”, as if they’re a fringe group, when the guy who’s the face of the whole AI movement is the one making the wild claims.
I'm really interested in what will happen to the economy/society in this case. Knowledge workers are the market for much that money is being made on.
Facebook and Google make most of their money from ads. Those ads are shown to billions of people who have money to spend on things the advertisers sell. Massive unemployment would mean these companies lose their main revenue stream.
Apple and Amazon make most of their money from selling stuff to millions of consumers and are this big because so many people now have a ton of disposable income.
Teslas entire market cap is dependent on there being a huge market for robo taxis to drive people to work.
Microsoft exists because they sell an OS that knowledge workers use to work on and tools they use within that OS to do the majority of their work with. If the future of knowledge work is just AI running on Linux communicating through API calls, that means MS is gone.
All these companies that currently drive stock markets and are a huge part of the value of the SP500 seem to be actively working against their own interests for some reason. Maybe they're all banking on being the sole supplier of the tech that will then run the world, but the moat doesn't seem to exist, so that feels like a bad bet.
But maybe I'm just too dumb to understand the world that these big players exist in and am missing some big detail.
Don’t forget Sam Altman publicly said they have no idea how to make money, and their brilliant plan is to develop AGI (which they don’t know how and aren’t close to) then ask it how to generate revenue.
https://www.startupbell.net/post/sam-altman-told-investors-b...
Maybe this imaginary AGI will finally exist when all of society is on the brink of collapse, then Sam will ask it how to make money and it’ll answer “to generate revenue, you should’ve started by not being an outspoken scammer who drove company-wide mass hysteria to consume society. Now it’s too late. But would you like to know how may ‘r’ are in ‘strawberry’?”.
https://www.newyorker.com/cartoon/a16995
The reason nobody did that is because you're not paying knowledge workers for their ability to crunch numbers, you're paying them to have a person to blame when things go wrong. You need them to react, identify why things went wrong and apply whatever magic needs to be applied to fix some sort of an edge case. Since you'll never be able to blame the failure on ChatGPT and get away with it, you're always gonna need a layer of knowledge workers in between the business owner and your LLM of choice.
You can't get rid of the knowledge workers with AI. You might get away with reducing their size and their day-to-day work might change drastically, but the need for them is still there.
Let me put it another way: Can you sit in front of a chat window and get the LLM to do everything that is asked of you, including all the experience you already have to make some sort of a business call? Given the current context window limits (~100k tokens), can you put all of the inputs you need to produce an output into a text file that's smaller in size than the capacity of a floppy disc (~400k tokens)? And even if the answer to that is yes, if it weren't for you, who else in your organization is gonna write that file for each decision you're the one making currently? Those are the sort of questions you should be asking before you start panicking.
Most white collar work is just a kind of game people play, it’s in to way needed, but people still enjoy playing it. Having AI writing reports nobody reads instead of people doing it isn’t going to change anything.
Yeah, and those new jobs will be called "long term structural unemployment", like what happened during deindustrialization to Detroit, the US Rust Belt, Scotland, Walloonia, etc.
People like to claim society remodels at will with almost no negative long term consequences but it's actually more like a wrecking ball that destroys houses while people are still inside. Just that a lot of the people caught in those houses are long gone or far away (geographically and socially) from the people writing about those events.
It is sad HN is sliding in the direction of folks being downvoted for opinions instead of the tone they use to express them :(
> I think it's ok to use the up and down arrows to express agreement. Obviously the uparrows aren't only for applauding politeness, so it seems reasonable that the downarrows aren't only for booing rudeness.
- Paul Graham, 2008
https://news.ycombinator.com/item?id=117171
Interesting hypothesis, do you have the math to back it up?
I genuinely hope that this ram/chips crisis gets solved ASAP by any party. The implications of this might have a lot of impact too and I feel is already a big enough financial crisis itself if we think about it coupled with all the other major glaring issues.
Maybe I'm misinterpreting "et tu" here.
Or maybe you meant "free markets" instead. Modern RAM production requires enormous R&D expenses, and thus has huge moat, which means the oligopoly is pretty safe (at least in the short to medium term) from new entrants. They "just" need to keep each other in check because there will be an incentive to increase production by each individual participant.
I do like the "OMEC" name as a paralel for OPEC.
Also, when treated right, computers almost never break.
There's so much hand-me-down stuff, that are not much worse than the current stuff, that I think people even in the poorest countries can get an okay computer or smartphone (and most of them do).
Like in current circumstances, its hard to get a homelab/datacenter so its better to postpone these plans for sometime
I agree with your statement overall but I feel like till the years that these ram shortages occur, there is a freeze of all companies providing vps's etc. ie. no new player can enter so I am a bit worried about those raising their prices as well honestly which will impact everyone of us as well for these few years in another form of AI tax
But the ram prices themselves are the reason I am forced to not enter this industry for the time being. I have decided right now to save my money for the time / focus on the job/college aspect of things to earn more so that when the timing is right, I would be able to invest my own money into it.
But basically Ram prices themselves are the thing which force us out of this market for the most part. I researched a lot about datacenters recently/ the rabbit hole and as previous hardware gets replaced/new hardware gets added/datacenters get expanded (whether they are a large company or small), I would expect an increase in prices mostly
This year, companies actually still took the cost but didn't want the market to panic so some black friday deals were good but I am not so sure about the next year or the next next year.
This will be a problem in my opinion for the next 1-3 or 4 years in my estimate
Also AWS is really on the more expensive side of things in the datacenters and they are immensely profitable so they can foot the bill while other datacenters (small or semi large) cant
So we will probably see a shift of companies towards using AWS and big cloud providers(GCP,AWS,azure) a bit more when we take all things into account which saddens me even more because I appreciate open web and this might take a hit.
We already see resentment towards these tri-fecta but we will see even more resentment as more and more people realize their roles / the impacts they cause and just overall, its my intuition that average person mostly hate big tech.
It's going to be a weird year in my opinion for this type of business and what it means for the average person.
Honestly for the time being, I genuinely recommend hetzner,upcloud,(netcup/ovh) and some others that I know from my time researching. I think that they are cheaper than aws usually while still being large enough that you don't worry about things too much and there is always lowendtalk if one's interested. Hope it helps but trust me, there is still hope as I talked to these hosting providers on forums like lowendtalk and It might help to support those people too since long term, an open web is the ideal.
Here is my list right now: hetzner's good if you want support + basic systems like simple compute etc. and dont want too much excess stuff
OVH's good: if you want other things than just compute and want more but their support is something which is of a mixed bag
Upcloud's good: if you want both of these things but they are just a bit more expensive if one wants to get large VPS's than the other options.
Netcup's good: Their payment processing was really painful that I had to go through but I think that one can find use case for them (I myself use netcup but although that's because they had a really steal deal once but I am not sure if I would recommend it if there are no deals)
There are some other services like exe.dev that I really enjoy as well and these services actually inspire me to learn more about these things and there are some very lovely people working in these companies.
There is still hope though. So never forget that. Its just a matter of time in my opinion that things get back normal hopefully so I think I am willing to wait till then since that's all we can do basically but overall, yea its a bit sad when I think about it too :<
This hype scenario would be the biggest bust of all for Ai. Without jobs or money then there is nobody to pay Ai to do all the things that it can do, it the power and compute it needs to function will be worth $0.
Or require non-price mechanisms of payment and social contract.
Some might conclude the same for funds (hedge funds/private equity) and housing.
Investors bought 1/3 of the US homes sold in 2023. This is, I think, quite alarming, especially since a small amount of extra demand can have a large effect on prices.
The statistic that matters is the ratio of owner occupied to rented single family homes.
And possibly even a lower equilibrium will be reached due to greater economies of scale.
In the interim, yeah, they will force prices up.
Additionally those fabs cost billions. Given the lead time I mentioned a lot of companies won't start building them right away since the risk of demand going away is high and the ROI in those cases might become unreachable
I have heard that in the fab making industry, things moves in cycles and this cycle has been repeated so many times and the reason that ram is so expensive was that at one time during covid there was shortage so they built more factories and they built so so many that these companies took a hit in stocks so they then went and closed and at just the bottom of their factory production levels, the AI bubble started to pop in and need their ram's and now they are once again increasing factory levels
And after the supply due to AI gets closed with the additional compute etc., I doubt it
I think that within 2-3 or maybe 4 years of timeframe ram will get cheaper imo.
The problem is, if someone can fill the market till that time.
This assumes infinite and uniformly distributed resources as well as no artificial factors such as lobbying, corruption, taxation or legislation which might favour one entity over the other.
The dream of free market exists only in a vacuum free from the constraints of reality.
Significantly increasing supply is also a huge multi year investment into a new fab that'd likely not pay out when the artificial demand breaks down.
I just want to play video games so I don’t have to interact with people
This has been played out before so it is only natural that they are careful with increasing the supply. And while they don't response they are netting larger margins than before.
Obvious end result is that demand will drop as price goes up. The other natural part of supply-demand curve.
Crucial is dead. There's a finite amount of rare earth. Wars and floods can bankrupt industries, supply chains are tight.
Micron stopped selling to consumers to focus on the high margin enteprise market. Might change in the future.
If the market is big enough, competitors will appear. And if the margins are high enough, competitors can always price-compete down to capture market-share.
Also, "price-compete down to capture market-share"? Prices are going up because all future production capacity has been sold. It makes no sense to lower prices if you don't have the capacity to full fill those orders.
Rare earth metals are in the dirt around the world.
Supply and demand curves shifting, hence prices increasing (and decreasing) is an expected part of life due to the inability to see the future.
They are. The problem is, the machinery to extract and refine them, and especially to make them into chips, takes years to build. We're looking at a time horizon of almost a decade if you include planning, permits and R&D.
And given that almost everyone but the AI bros expects the AI bubble to burst rather sooner than later (given that the interweb of funding and deals more resembles the Habsburg family tree than anything healthy) and the semiconductor industry is infamous for pretty toxic supply/demand boom-bust cycles, they are all preferring to err on the side of caution - particularly as we're not talking about single billion dollar amounts any more. TSMC Arizona is projected to cost 165 billion dollars [1] - other than the US government and cash-flush Apple, I don't even know anyone able, much less willing to finance such a project under the current conditions.
Apple at least can make use of TSMCs fab capacity when the AI bros go bust...
[1] https://www.tsmc.com/static/abouttsmcaz/index.htm
I guess people might be mixing up all the headlines of all the articles they did not read by this point.
Increasing memory production capacity is a multi-year project, but in a few years, the LLM companies creating the current demand might all have run out of money. If demand craters just as supply increases, prices will drastically decrease, which none of these companies want.
Take a look at GPU prices and how that "supply increased thus bringing the prices down"
But we don't see this bankrolling in absolute values. Rather, it's due to regressive taxation, low (cheap) social security for workers, and very weak intellectual property protection.
If the American voter base doesn't pull its shit together and revive democracy, we're going to have a bad century. Yesterday I met a man who doesn't vote and I wanted to go ape-shit on him. "My vote doesn't matter". Vote for mayor. Vote for city council. Vote for our House members. Vote for State Senate. Vote for our two Senators.
"Voting doesn't matter, capitalism is doomed anyway" is a self-fulling prophecy and a fed psy-op from the right. I'm so fucking sick of that attitude from my allies.
If we see politician as just a machine who's only job is to get elected, they have to get as many votes as possible. Pandering to the individual is unrealistic, so you usually target groups of people who share some common interest. As your aim is to get as many votes as possible, you will want to target the “bigger” (in amount of potential vote) groups. Then it is a game of trying to get the bigger groups which don't have conflicting interest. While this is theory and a simplification of reality, all decent political party do absolutely look at statistics and survey to for a strategy for the election.
If you are part of a group that, even though might be big in population, doesn't vote, politician have no reason to try to pander to you. As a concrete example, in a lot of “western” country right now, a lot of politician elected are almost completely ignoring the youth. Why ? Because in those same country the youth is the age group which vote the less.
So by not voting, you are making absolutely sure that your interest won't be defended. You can argue that once elected, you have no guarantee that the politician will actually defend your interest, or even do the opposite (as an example, soybean farmer and trump in the U.S). But then you won't be satisfied and possibly not vote for the same guy / party next election (which is what a lot of swing voters do).
But yeah, in an ideal world, everyone would vote, see through communication tactics and actually study the party, program and the candidate they vote for, before voting.
;) And you wanted to go ape shit on him... For falling for a psy-op?
My friend, morale is very very low. There is no vigor to fight for a better tomorrow in many people's hearts. Many are occupied with the problems of today. It doesn't take a psy-op to reach this level of hopelessness.
Be sick of it all you want, it doesn't change their minds. Perhaps you will find something more persuasive.
All the ills of modern (American) politics stem by the blaming one side for the problems caused by both.
Current polling however says the current voter base is quite unhappy with how this is
You don’t need to make them happy, just scared of the opposition.
Nonvoters aren’t being irrational.
Ahem, you'll find that with regulation, money begets money and monopolies will form. That is, unless you magically produce legislators which are incorruptible, have perfect knowledge and always make the perfect choice.
Even the Big Bang was imperfect, and matter clumps together instead of being perfectly distributed in the available space.
If a monopoly appears due to superior offerings, better pricing and quicker innovation, I fail to see why it needs to be a bad thing. They can be competed against and historically that has always been the case.
On the other hand, monopolies appearing due to regulations, permissions, patents, or any governmental support, are indeed terrible, as they cannot be competed against.
There are several inventions which are far greater than LLMs. To name two: computers and methods to generate electricity, things without which LLMs wouldn’t have been possible. But also harnessing fire, the wheel, agriculture, vaccines… The list goes on and on.
Calling LLMs “AI thingies” seems much more in tune with reality than calling them “the greatest invention of man” (and I’m steel manning and assuming you meant “latest”, not “last”). You can’t eat LLMs or live in them and they are extremely dependent on other inventions to barely function. They do not, in any way, deserve the title of “greatest invention”, and it’s worrying that we’re at that level of hyperbole. Though you’re certainly not the first one to make that claim.
https://finance.yahoo.com/news/alphabet-ceo-sundar-pichai-sa...
Industry mandate should have become 16GB RAM for PCs and 8GB for mobile, since years ago, but instead it is as if computing/IT industry is regressing.
New budget mobiles are being launched with lower-end specs as well (e.g., new phones with Snapdragon Gen 6, UFS2.2). Meanwhile, features that were being offered in budget phones, e.g., wireless charging, NFC, UFS3.1 have silently been moved to the premium mobile segment.
Meanwhile the OSes and software are becoming more and more complex, bloated and more unstable (bugs) and insecure (security loopholes ready for exploits).
It is as if the industry has decided to focus on AI and nothing else.
And this will be a huge setback for humanity, especially the students and scientific communities.
Of course, some performance-focused software (e.g. Zed) does start near-instantly on my MacBook, and it makes other software feel sluggish in comparison. But this is the exception, not the rule.
Even as specs regress, I don't think most people in software will care about performance. In my experience, product managers never act on the occasional "[X part of an app] feels clunky" feedback from clients. I don't expect that to change in the near future.
It's just in last 5 years integrated GPUs become good enough even for mid-tier gaming let alone running browser and hw accel in few work apps.
And even before 5 years ago majority of dedicated GPUs in relatively cheap laptops was garbage barely better than intrgrated one. Manufacturers mostly put them in there for marketing of having e.g Nvidia dGPU.
Without dedicated GPUs, we consumers will get only weaker hardware, slower software and the slow death of graphics software market. See the fate of Chromebooks market segment - it is almost dead, and ChromeOS itself got abandoned.
Meanwhile, the same Google which made ChromeOS as a fresh alternative OS to Windows, Mac and Linux, is trying to gobble the AI market. And the AI race is on.
And the result of all this AI focus and veering away from dedicated GPUs (even by market leader nVidia, which is no longer having GPUs as a priority) is not only the skyrocketing price hikes in hardware components, but also other side effects. e.g., new laptops are being launched with NPUs which are good for AI but bad for gaming and VFX/CAD-CAM work, yet they cost a bomb, and the result is that budget laptop market segment has suffered - new budget laptops have just 8GB RAM, 250GB/500GB SSD, and poor CPU, and such weak hardware, so even basic software (MS Office) struggles on such laptops. And yet even such poor laptops are having a higher cost these days. This kind of deliberate market crippling affects hundreds of millions of students and middle class customers who need affordable yet decent performance PCs.
All the popular mass matket games work on iGPU: fortnite, roblox, mmos, arena shooters, battlen royales. Good chunk of cross platform console titles also work just fine.
You can play Cyberpunk or BG3 on damn Steam Deck. I wont call this low end.
Number of games that dont run to some extent without dGPU is limited to heavy AAA titles and niche PC only genres.
Your experience is extremely weird
The pattern of lazy almost non existent optimization combined with blaming consumers for having weak hardware, needs to stop.
On my 16GB ram lunar lake budget laptop CachyOS( Arch) runs so much smoother than Windows.
This is very unscientific, but using htop , running Chrome/YouTube playing music, 2 browser games and VS code having Git Copilot review a small project, I was only using 6GBs of ram.
For the most part I suspect I could do normal consumer stuff( filing paperwork and watching cat videos) on an 8GB laptop just fine. Assuming I'm using Linux.
All this Windows 11 bloat makes computers slower than they should be. A part of me hopes this pushes Microsoft to at least create a low ram mode that just runs the OS and display manager. Then let's me use my computer as I see fit instead of constantly doing a million other weird things.
We don't *need* more ram. We need better software.
Browsing web requires more and more RAM each year but I don't think browsers are the main reason - sites use more and more JS code. With a hard cap many sites will stop working. Software bloat is a natural tendency, the path of least resistance. Trimming weigh requires a significant effort and in case of web - a coordinated effort. I don't believe it could happen unless Google (having a browser with >60% market share) will force this but Google own sites are among worst offenders in term of hardware requirements.
Windows OS and Surface (CoPilot AI-optimized) hardware have been combined in the "Windows + Devices" division.
> We don't *need* more ram
RAM and SSDs both use memory wafers and are equally affected by wafer hoarding, strategic supply reductions and market price manipulation.
Nvidia is re-inventing Optane for AI storage with higher IOPS, and paid $20B for Groq LPUs using SRAM for high memory bandwidth.
The architectural road ahead has tiers of memory, storage and high-speed networking, which could benefit AI & many other workloads. How will industry use the "peace dividend" of the AI wars? https://www.forbes.com/sites/robtoews/2020/08/30/the-peace-d...
Why on earth you think RAM uses NAND flash ?
You can already do this. For example, I use `systemd-run` to run browsers with CPU quotas applied. Firefox gets 400% CPU (i.e. up to 4 cores), and no more.
Example command: systemd-run --user --scope -p CPUQuota=400% firefox
You can limit CPU usage for a program in Windows by adjusting the "Maximum processor state" in the power options to a lower percentage, such as 80%. Additionally, you can set the program's CPU affinity in Task Manager. Please note this will only affect the process scheduling.
You can also use a free tool like Process Lasso or BES to limit the CPU for a Windows application. You can use a free tools like HWInfo, SysInternals (ProcMon, SysMom, ProcDump) to monitor and check for CPU usage, especially to investigate CPU spikes caused by rogue (malware or poor performance) apps.
Insane that this is seen as "better software". I could do basically the same functionality in 2000 with 512mb. I assume this is because everything runs through chrome with dozens more layers of abstraction but
512MB in 2000 was like HEDT level (though I'm not sure that acronym existed back then)
64MB = w98se OK, XP will swap a lot on high load, nixlikes really fast with fvwm/wmaker and the like. KDE3 needs 128MB to run well, so get a bit down. No issues with old XFCE releases. Mozilla will crawl, other browsers will run fine.
128MB = w98se really well, XP willl run fine, SP2-3 will lag. Nixlikes will fly with wmaker/icewm/fvwm/blackbox and the like. Good enough for mozilla.
192MB = Really decent for a full KDE3 desktop or for Windows XP with real life speeds.
256MB = Like having 8GB today for Windows 10, Gnome 3 or Plasma 6. Yes, you can run then with 2GB and ZRAM, but, realistically, and for the modern bloated tools, 8GB for a 1080p desktop it's mandatory. Even with UBlock Origin for the browser. Ditto back in the day. With 256MB XP and KDE3 flied and they ran much faster than even Win98 with 192MB of RAM.
Option A: We do a better job at optimizing software so that good performance requires less RAM than might otherwise be required
Option B: We wish that things were different, such that additional RAM were a viable option like it has been at many times in the past.
Option C: We use our time-benders to hop to a different timeline where this is all sorted more favorably (hopefully one where the Ballchinians are friendly)
---
To evaluate these in no particular order:
Option B doesn't sound very fruitful. I mean: It can be fun to wish, but magical thinking doesn't usually get very far.
Option C sounds fun, but my time-bender got roached after the last jump and the version of Costco we have here doesn't sell them. (Maybe someone else has a working one, but they seem to be pretty rare here.)
That leaves option A: Optimize the software once, and duplicate that optimized software to whomever it is useful using that "Internet" thing that the cool kids were talking about back in the 1980s.
Mint 22.x doesn't appear to be demanding any more of my machine than Mint 20.x. Neither is Firefox or most websites, although YouTube chat still leaks memory horrendously. (Of course, download sizes have increased.)
Modern software is fine for the most part. People look at browsers using tens of gigabytes on systems with 32GB+ and complain about waste rather than being thrilled that it's doing a fantastic job caching stuff to run quickly.
I think root comment is looking at the overall picture of what all customers can get for their money, and see it getting worse.
This wasn’t mentioned, but it’s a new thing for everyone to experience, since the general trend of computer hardware is it gets cheaper and more powerful over time. Maybe not exponentially any more, but at least linearly cheaper and more powerful.
A $999 MacBook Air today is vastly better than the same $999 MacBook Air 5 years ago (and even more so once you count inflation).
If you want a powerful laptop for cheap, get a gaming PC. The build quality and battery life probably won't be great, but you can't be cheap without making compromises.
Same idea for budget mobiles. A Snapdragon Gen 6 (or something by Mediatek) with UFS2.2 is more than what most people need.
it was only less than 10 yrs ago that a high end PC would have this level of ram. I think the last decade of cheap ram and increasing core count (and hz) have spoiled a lot of people.
We are just returning back on trend. May be software would be written better now that you cannot expect the average low budget PC to have 32G of ram and 8 cores.
The problem is coming up with a viable business model for providing hardware and software that respect users’ ability to shape their environments as they choose. I love free, open-source software, but how do developers make a living, especially if they don’t want to be funded by Big Tech?
We'll probably end up in an even more bifurcated world where the well off have access to lot of great products and services that most of humanity is increasingly unable to access.
There is the law of uncertainty override it eg trade wars, tariffs , etc.
No 1 is going all in with new capacity.
On the bright side, I'm not responsible for the UI abominations people seem to complain about WRT laptop specs.
Apps are optimized for the install base, not for the engineer's own hardware.
That's like 100B+ instructions on a single core of your average superscalar CPU.
I can't wait for maps loading times being measured in percentage of trip time.
If someone with a low-memory laptop wants to get into coding, modern software-development-related services are incredible memory hogs.
but you cannot consider this in isolation.
The developed markets have vastly higher spending consumers, which means companies cater to those higher spending customers proportionately more (as profits demand it). Therefore, the implication is that lower spending markets gets less investment and less catered to; after all, R&D spending is still a limiting factor.
If the entirety of the market is running on lower powered devices, then it would get catered for - because there'd be no (or not enough) customers with high powered devices to profit off.
Both didn’t run great on the “average consumer hardware”.
But I’ll admit this is cherry picking from my side :)
Allocating a very large share of advanced memory production, especially HBM and high-end DRAM, which are critical for almost all modern technology (and even many non-tech products like household appliances) to a small number of U.S. centric AI players risks distorting the global market and limiting availability for other industries.
Even within Samsung itself, the Mobile eXperience (MX) Business (smartphones) is not guaranteed preferential access to memory from Samsung’s Device Solutions (DS) Division, which includes the Memory Business. If internal customers are forced to source DRAM elsewhere due to pricing or capacity constraints, this could eventually become economically problematic for a country that relies very heavily on semiconductor and technology exports.
It's not like it's their fault that micron pulled out of the market...
Edit: maybe someone should consider sweet-talking kioxia into making dram chips?
AI embodies everything wrong with our modern gilded age era of capitalism.
100%, there is a lost time factor which sucks though if someone wants beefy, some might be forced to cough up some money but honestly yea although I agree with your statement, the ram increases really impact the self hosting/homelabbing genres and there was a recent post relevant to this discussion on hackernews https://news.ycombinator.com/item?id=46416618: Self hosting is being enshittified.
I know this is not always true, but on this case, crucial folks say the margins for end user are too low and they have demand for AI.
I suppose they do not intend to bring a new AI focused unit because it is not worth it or they believe the hype might be gone before it they are done. But what intrigues me is why they would allow other competitors to step up in a segment they dominate? They could raise the prices for the consumers if they are not worried about competition...
There is a whole "not-exactly" ai industry labeled as AI that received a capital t of money. Is that what they are going for?
So now for these AI companies, they got tons of money to burn so they are willing to pay a lot more, so now crucial only have a limited supply of ram and the thing is there isn't much difference between AI chip and consumer chip but the margins of AI chip are super higher compared to consumer chip
So earlier they would sell consumer chips and AI chips as well but then the AI companies still demanded even more and they would get insane profits selling them so what they did (atleast crucial) is that they stopped selling consumer chips just to sell AI chips for profit.
> I suppose they do not intend to bring a new AI focused unit because it is not worth it or they believe the hype might be gone before it they are done. But what intrigues me is why they would allow other competitors to step up in a segment they dominate? They could raise the prices for the consumers if they are not worried about competition...
Well as a consumer, I certainly hope so but I think that these companies did this case because their have been times they were -55% in stock prices and its just cash making money device at this point and there is a monopoly of fabs with just three key players.
So the answer to your question is "money" and "more money" short term. Their stock prices are already up I think and a company really loves short term rising stock prices
> They could raise the prices for the consumers if they are not worried about competition
Well, would you increase the prices 3-4x? Because supposedly thats how much the AI chips from what I've heard are... And due to this, the second hand market itself is selling these at a close-enough mark.
I don't know but I hope that new players come in the market, I didn't know that this ram industry was such monopolistic with there being only 3 key players and how that became a chokehold for the whole world economy in a way
It seems they could. They not only single handed caused it to double or more without trying :/ Not sure if it would trigger other sorts of regulatory issues though
It is my impression that there is a fabricated scarcity of all goods. That's a common practice in cloth retailers. In the 90s they thought for brand name and market share. They noticed it was silly because they could sell half for double of the price and as this means less logistics, it also meant higher margins. It is not a lunch free approach. Selling less means that you delegate at least the bottom portion of your clients to the market, and if there are options, they might just be gone. That's exactly what happened with Chevrolet, Ford, etc. they stopped investing and when a new competitor appeared, even if it was more marketing than product, they lots rivers of money and barely can keep the fight on (except for maybe making a puppet tell others that there is no such thing as climate change, but that's something else)
Technology space right now looks like it. We already see major brands stagnation allegedly because they did all that is possible and it will take some time until some nouvelle approach to appear.
As a consumer, I want to believe this won't take long to settle but I'm afraid money is going elsewhere
The world ran just fine on DDR3 for a long time.
Not sure if they require DDR5 but the AI crisis just caused the prices of DDR5 to rise but the market supply of DDR4 thus grew and that's why they got more expensive too
> I'm also wondering about the supply of the alternate nodes and older technologies.
I suppose these might be chinese companies but there might be some european/american companies (not sure) but if things continue, there is gonna be a strain on them in demand and they might increase their prices too
> Was it micron that abandoned the entire retail market in favor of supplying the hyperscalers?
Yes
The EV is a therefore, on a whole, a lot less sensitive to DRAM price increases.
That might be the case only for the infotainment system, but there’s usually many other ECUs in an EV. The ADAS ECUs are carrying similar amounts as an iPhone or the infotainment system. Telematics is also usually also a relatively complex one, but more towards lower sized amounts.
Then you have around 3-5 other midsized ECUs with relatively high memory sizes, or at least enough to require MMUs and to run more complex operating systems supporting typical AUTOSAR stacks.
And then you have all the small size ECUs controlling all small individual actuators.
But also all complex sensors like radars, cameras, lidars carry some amounts of relevant memory.
I still think your point is valid, though. There’s no difference in orders of magnitude when it comes to expensive RAM compared to an iPhone. But there’s cars also carried lots of low-speed, automotive grade memory in all the ECUs distributed throughout the vehicle.
> will the DRAM manufactures even keep those nodes running in favor of these industries?
Some will, Some might not, In my opinion, the longevity of these brands will only depend if they allow buying ram for the average person/consumer brands so I guess we might see new competition perhaps or give more marketshare to all the other fab companies beyond the main three of these industries.
I am sure that some company will 100% align with the consumers but the problem to me feels that they wouldn't be able to supply enough production to consumers in the first place so prices still might rise.
And those prices most likely will be paid by you in one form or another but it would be interesting to see how long the companies who buy dram from these providers or build datacenters or anything ram intensive will hold their price up, perhaps they might eat the loss short term similar to what we saw some companies do during trump tarrifs.
Looks like the frame.work desktop with Ryzen 128GB is shipping now at same price it was on release, Apple is offering 512GB Mac studios
Are snapdragon chips the same way?
Note that the memory is on the board for Ryzen AI Max, not on the package (as it is for Intel’s Lunar Lake and Apple’s M-series processors) or on die (which would be SRAM). As noted in another comment, whether the memory is on the board, on a module, or on the processor package, they are all still coming from the same extremely constrained three memory die suppliers, so costs are going up for all of them.
All your competitors are in the same boat, so consumers won’t have options. It’s much better to minimize the risk of blowing up by sticking as closely to spot at possible. That’s the whole idea of lean. Consumers and governments were mad about supply chains during the pandemic, but companies survived because they were lean.
In a sense this is the opposite risk profile of futures contracts in trading/portfolio management, even though they share some superficial similarities. Manufacturing businesses are fundamentally different from trading.
They certainly have contracts in place that cover goods already sold. They do a ton of preorders which is great since they get paid before they have to pay their suppliers. Just like airlines trade energy futures because they’ve sold the tickets long before they have to buy the jet fuel.
the risk is that such longer contracts would then lock you into a higher cost component for longer, if the price drops. Longer contracts only look good in hindsight if ram prices increased (unexpectedly).
SoCs with on-die memory (which is, these days, exclusively SRAM, since I don't think IBM's eDRAM process for mixing DRAM with logic is still in production) will not be effected. SiPs with on-package DRAM, including Apple's A and M series SiPs and Qualcomm's Snapdragon, will be effected -- they use the same DRAM dice as everyone else.
To answer the original question: the Framework Desktop is indeed still at the (pretty inflated) price, but for example the Bosgame mini PC with the same chip has gone up in price.
"dice" is the plural for the object used as a source of randomness, but "dies" is the plural for other noun uses of "die".
The bigger the company = longer the contract.
However it will eventually catch up even to Apple.
It is not prices alone due to demand but the manufacturing redirection from something like lpddr in iphones to hbm and what have you for servers and gpu
https://www.google.com/amp/s/www.indiatoday.in/amp/technolog...
https://www.indiatoday.in/technology/news/story/ram-shortage...
I've got old linux boxes that feel fine with a couple gig of DDR3, but can't think of a place where that would be acceptable outside of that.
Can this not be a opportunity for new entrants to start serving the other market segments?
How hard is it to start and manufacture memory for embedded systems in cars, or pc?
Even if they might sell their inference and everything, they still wouldn't be that much profitable.
So like, the key point is that a ram company can supply openAI ram and get some really high quick bucks which would be even more than if they were to create thier own datacenters,run open source models in them, provide inference in say open router.
Now you might ask: Is openAI or these AI companies mad for burning so much money?
And I think you might know the answer to that.
I am pretty sure, in the next year we will see a wave of low end ram components coming out of china.
- Right now A LOT of PCs are getting out of date because Windows 11 wants some new hardware that's missing in older PCs. - At the same time, smartphones were starting to use and need more memory. - Modern cars (mostly electric) have much more electronics inside than older ones... they're now distributed systems with several CPUs working together along a bus. - The cloud needs to upgrade and newer servers have much more memory than older ones with almost the same foorprint (wich means you need to lease less datacenter space). - And GPUs for AI are in demand and need RAM.
But only AI is to blame although we're living in a perfect storm for RAM demand.
There was a time when apple was hesitant to add more ram to its iPhones and app developers would have to work hard to make apps efficient. Last few years have shown Apple going from 6gb to 12gb so easily for their 'AI' while I consistently see the quality of apps deteriorating on the App Store. iOS 26 and macOS 26 are so aggressive towards memory swapping that loading settings can take time on devices with 6gb ram (absurd). I wonder what else they have added that apps need purging so frequently. 6gb iphone and 8gb M1 felt incredibly fast for the couple of years. Now apparently they are slow like they are really old.
Windows 11 and Chrome are a completely different story. Windows 10 ran just fine on my 8th gen pc for years. Windows 11 is very slow and chrome is a bad experience. Firefox doesn't make it better.
I also find that gnome and cosmic de are not exactly great at memory. A bare minimum desktop still takes up 1.5-1.6gb ram on a 1080p display and with some tabs open, terminal and vscode (again electron) I easily hit 8gb. Sway is better in this regard. I find alacrity sway and Firefox together make it a good experience.
I wonder where we are heading on personal computer software. The processors have gotten really fast and storage and memory even more so, but the software still feels slow and glitchy. If this is the industry's idea of justifying new hardware each year we are probably investing in the wrong people.
Set this to something you find reasonable: `browser.low_commit_space_threshold_percent`
And make sure tab unloading is enabled.
Also, you can achieve the same thing with cgroups by giving Firefox a slice of memory it can grow into.
So writing optimized software was niche and overshadowed by the huge gains of hardware. People and corporations didn't care. They preferred fast feature delivery. Even when optimizing techniques like multi-tenant servers, we ended up having heavy containers wasting RAM and resources. And most apps switched to web frameworks like Electron where each one has its own huge web browser. Most people didn't care.
I hope this shortage has a silver lining of untangling the mess and refactoring the sea of bloat. But I know it's more likely some trick will be found to just patch it up a bit and only reduce the pain up to the new equilibrium level. For example something idiotic like Electron sharing resources across multiple apps and taking huge security risks. Corporations love to play false dichotomies to save pennies.
At least the first years.
Maybe the market pricing people out is accidentally doing what regulation couldn't? Concentrating AI where there's at least some oversight and accountability. Not sure if that's good or bad to be honest.
For now. Chinese supply chains include DRAM from CXMT (sanctioned) and NAND from YMTC (not sanctioned, holds patents on 3D stacking that have been licensed by Korean memory manufacturers).
next stage is paving everything with solar panels.
A 16 GiB M4 Mac Mini is $400 right now. That covers any essential use-case which means this is mostly hitting hobbyists or niche users.
Where do you live? In Poland it's 740usd.
However, 16gb is very little for a Mac where the OS itself uses 7GB
Prices are already through the roof...
https://www.tomsguide.com/news/live/ram-price-crisis-updates
2028 is another story depending on whether this frenzy continues / fabs being built (don’t know whether they are as hard as cpu)
So lets see if they might "save us"
https://www.tomshardware.com/pc-components/dram/no-asus-isnt...
My bad
And a couple of smaller ones: CXMT (if you’re not afraid of the sanctions), Nanya, and a few others with older technology
If I recall correctly, RAM is even more niche and specialized than the (already quite specialized) general chip manufacturing. The structure is super-duper regular, just a big grid of cells, so it is super-duper optimized.
You’re correct that DRAM is a very specialized process. The bit cell capacitors are a trench type that is uncommon in the general industry, so the major logic fabs would have a fairly uphill battle to become competitive (they also have no desire to enter the DRAM market in general).
Governments need to intervene here. This is a mafia scheme now.
I purchased about three semi-cheap computers in the last ~5 years or so. Looking at the RAM prices, the very same units I bought (!) now cost 2.5x as much as before (here I refer to my latest computer model, from 2 years ago). This is a mafia now. I also think these AI companies should be extra taxed because they cause us economic harm here.
Maybe this is the free market working as intended -- did you know that RAM is actually a luxury item, like a Rolls Royce, and most plebes should just make do with 4gb machines, because that is the optimum solution!
E.g. IDEs could continue to demand lots of CPU/RAM, and cloud providers are able to deliver that cheaper than a mostly idle desktop.
If that happens, more and more of its functionality will come to rely on having low datacenter latencies, making use on desktops less viable.
Who will realistically be optimising build times for usecases that don't have sub-ms access to build caches, and when those build caches are available, what will stop the median program from having even larger dependency graphs.
This will only serve to increase the power of big players who can afford higher component prices (and who, thanks to their oligopoly status, can effectively set the market price for everyone else), while individuals and smaller institutions are forced to either spend more or work with less computing resources.
The optimistic take is that this will force software vendors into shipping more efficient software, but I also agree with this pessimistic take, that companies that can afford inflated prices will take advantage of the situation to pull ahead of competitors who can’t afford tech at inflated prices.
I don’t know what we can do as normal people other than making do with the hardware we have and boycotting Big Tech, though I don’t know how effective the latter is.
A dad comes home and tells his kid, “Hey, vodka’s more expensive now.” “So you’re gonna drink less?” “Nope. You’re gonna eat less.”
I highly recommend disabling javascript in your browser.
Yes, it makes many sites "look funny", or maybe you have to scroll past a bunch of screen sized "faceplant" "twitverse" and "instamonetize" icons, but, there are far fewer ads (like none).
And of course some sites won't work at all. That's OK too, I just don't read them. If it's a news article, its almost always available on another site that doesn't require javascript.
But I use NoScript and it is definitely a big help.
Life online without javascript is just better. I've noticed an increase in sites that are useful (readable) with javascript disabled. Better than 10 years ago, when broken sites were rampant. Though there are still the lazy ones that are just blank pages without their javascript crutch.
Maybe the hardware/resource austerity that seems to be upon us now will result in people and projects refactoring, losing some glitter and glam, getting lean. We can resolve to slim down, drop a few megs of bloat, use less ram and bandwidth. It's not a problem; it's an opportunity!
In any case, Happy New Year! [alpha preview release]
Isn't Micron stopping all consumer RAM production? So their factories won't help anyway.
Also, even if no Micron RAM ever ended up in consumer hands, it would still reduce prices for consumers by increasing the supply to other segments of the market.
It could be restarted in the future by Micron.
Crucial SSDs offer good firmware (e.g. nvme sanitize for secure erase) and hardware (e.g. power loss capacitors).
The RAM looks like cornering market. Probably something OpenAI should be prosecuted for if they end up profiting from it.
https://www.merriam-webster.com/dictionary/oligopoly
e.g., the Phoebus cartel https://en.wikipedia.org/wiki/Phoebus_cartel
If demand exceeds supply, either prices rise or supply falls, causing shortages. Directly controlling sellers (prices) or buyers (rationing) results in black markets unless enforcement has enough strength and integrity. The required strength and integrity seems to scale exponentially with the value of the good, so it's typically effectively impossible to prevent out-of-spec behavior for anything not cheap.
If everyone wants chips, semiconductor manufacturing supply should be increased. Governments should subsidize domestic semiconductor industries and the conditions for them to thrive (education, etc.) to meet both goals of domestic and economic security, and do it in a way that works.
The alternative is decreasing demand. Governments could hold bounty and incentive programs for building electronics that last a long time or are repairable or recyclable, but it's entirely possible the market will eventually do that.
If there is already demand at this inflated price, shouldn’t we ask why more capacity is not coming online naturally first?
The non financial parts, which include mandated restructuring and penalties to directors including incarceration however, are not tokenistic. They'd be appealed and delayed, but at some point the shareholders would seek redress from the board. Ignoring judicial mandated instructions isn't really a good idea, current WH behaviour aside. If the defence here is "courts don't matter any more" that's very unhelpful, if true. At some point, a country which cannot enforce judicial outcomes has stopped being civil society.
My personal hope the EU tears holes in the FAANG aside, the collusive pricing of chips has been a problem for some time. The cost/price disjunction here is strong.
Whether you like it or not, AI right now is mostly
- high electricity prices - crazy computer part prices - phasing out of a lot of formerly high paying jobs
and the benefits are mostly - slop and chatgpt
Unless OpenAI and co produce the machine god, which genuinely is possible. If most people's interactions with AI are the negative externalities they'll quickly be wondering if ChatGPT is worth this cost.
They should be, and the answer is obviously no—at least to them. No political or business leader has outlined a concrete, plausible path to the sort of vague UBI utopia that's been promised for "regular folks" in the bullish scenario (AGI, ASI, etc.), nor have they convincingly argued that this isn't an insane bubble that's going to cripple our economy when AGI doesn't happen—a scenario that's looking more and more likely every day.
There is no upside and only downside; whether we're heading for sci-fi apocalypse or economic catastrophe, the malignant lunatics pushing this technology expect to be insulated from consequences whether they end up owning the future light-cone of humanity or simply enjoying the cushion of their vast wealth while the majority suffers the consequences of an economic crash a few rich men caused by betting it all, even what wasn't theirs to bet.
Everybody should be fighting this tooth and nail. Even if these technologies are useful (I believe they are), and even if they can be made into profitable products and sustainable businesses, what's happening now isn't related to any of that.
Not saying this is necessarily a bad prediction for 2028, but I'm old enough to remember when the 2020 election was going to be a referendum on billionaires and big tech monopolies.
Presumably the boom times are the main reason why investment goes into it so that years later, consumers can buy for cheap.