The RAM shortage is bringing back DDR4
theverge.com
theverge.com
There's an entire parallel ecosystem around Chinese X79 and X99 boards [1](brands like Huananzhi and Machinist). They exist almost entirely to soak up scrap-tier Xeons and decommissioned server DDR3 and DDR4 ECC RAM that get pulled from datacenters for next to nothing. You see them online in Russian/Chinese/SE Asian online communities as budget builds, so nearly all the BIOS mods and documentation are in Russian or Chinese.
I think in NA we'd tend to write this hardware off as e-waste unless you're a homelabber (though I know Kimsufi still includes DDR3 builds in their catalogue). But in emerging markets, people have been squeezing real utility out of discarded enterprise silicon for a long time out of pure economic necessity. I've been tempted to do my own build but I already seek to have collected an overwhelmingly unnecessary amount of 2012-era workstations that I'm not sure my partner would appreciate an addition to.
My "6 years ago" build was with more typical current-gen consumer-grade hardware, because at that point I was on the hook for the electricity bill. I haven't thought about what my next build will look like because it's still going strong (aside from a flakey DIMM... so my next system will have ECC).
Maybe that's still too much power for some people, but I'm happy with the tradeoffs.
At about $350 each, I had to try them. Everything else was double.
Only thing I can say so far is, they boot linux on second hand EPYC 7002 series CPU's.
Haven't put them through their paces yet. V
Oh, I remember quirk #1 ... apparently, only 1 of the case fan headers is speed controllable, the others run at full speed all the time, with no bios setting.
You can fix it with a fan hub that supports pwm so not a show stopper by any means, for those wanting to try these motherboards.
I guess its a substitute for "just so you know" or "fyi" or any other "I'm an authority on something" ending. :D
Now that everyone has bitmap emojis, not so much.
Initially it was just for home server stuff (ZFS array for storage/backups, Plex, Vaultwarden, etc) but then I wanted to experiment with a remote dev environment using LXC/LXD.
I now do all my web development (mostly back-end, Python and C#) on that machine via tmux and VS Code remote SSH. I'm sure if I was running my browser on it it might painful given the slow single-core performance, but honestly I've felt zero need to upgrade it. Once the RAM-pocalypse is over I'll probably look at doing so just for reliability's sake.
EU average is just under 29c/kWh (non-contract price) [0]
In the UK I got 27.5c/kWh (+ 58c/day fixed charge) on an 18-month contract a month ago
[0] https://ec.europa.eu/eurostat/statistics-explained/index.php...
But after all the "sales true ups" and "affordability charges" and "resource adjustments" and "fuel cost charges" and "interim rate adjustments", not to mention the city fees and sales tax, the final bill divided by my usage gave me an effective rate of 23 c/kWh.
Utilities have to argue for rate increases and get them approved by regulators, so they’re not just dropping fees on top of everything as they please from month to month.
Really nobody should assume that their service is flat rate with a slope down to $0/month if you consume 0kWh. There are fixed and marginal costs in every service like this.
So no LLMs for 3-5 months a year in most of Europe?
I believe we can generate baseload energy by cooling the planet... nearly everywhere, no fuels needed.
Afaik electricity is around 7c/kWh in China. I got 8c/kWh 2y contract a couple weeks ago here in Finland. Though have to add 4c/kWh for transfer to that (no clue if the Chinese prices I could find include that). Commercial prices are less then that (no VAT and possibly lower power tax bracket. The 8c includes all taxes/fees)
Which is why I suspect most nominal "data centers" are fronts for crypto mining, an activity that seeks out the cheapest energy.
if anybody got a tip, i'd be more than glad to listen, please.
Yeah the Xeon E5 v4 to me are really an amazing spot. These things, today, are basically free. If you look a bit you can buy stuff like four 20 cores E5 v4 for $60, making them effectively $15 each. Basically free.
And here's the trick: I don't need my Xeon to be up 24/7. 8/7 is sufficient, so I save 2/3rd of the electricity cost. I set up my network so that when I sleep, my main server doesn't need to be on. The services that really need be on 24/7 (say my DNS for when I'm sleeping and my wife/kid need Internet) are running on a Raspberry Pi (which consumes nothing).
And when I need the shitloads of cores / ZFS etc.: I turn the Xeon on.
Now it's still a server in my case: my main machine is a recent AMD.
> if anybody got a tip, i'd be more than glad to listen, please.
Slap 128 GB of the 256 in a $50 used Xeon server and use that has a backup / to run VMs / ZFS / etc.
The specs page (at least for the board I tried) noted the board "Accepts" DDR4 ECC RDIMMS, but nothing I tried indicated ECC was actually functioning.
In all seriousness, anyone remember when every library, business, government agency, and institution, etc was buying $20,000 cisco routers to "get on the internet" during the dot com boom/bust in the late 90's ?
Only to see them all end up on the surplus market, unused, for a couple hundred ?
We're in the AI boom/bust cycle, and its my understanding DRAM fab is specialized enough, that the supply is getting squeezed.
As for "doing it on purpose" ? It's being milked for all its worth. That's what companies do. They milk the market for what it will pay.
Also, for everyone who is trying to do local AI, on Xeon or EPYC platforms, memory bandwidth scales with module count, making them that much more in demand in terms of processing capacity / wafer area.
That's why used enterprise/industrial electronics are way less common on EU eBay compared to other places, despite the industrial base being so large (or used to being so large). You can actually see when WEEE came into force because very little used T&M equipment entered circulation in the EU after the early 2000s.
Another factor is lease agreements, where companies don't buy equipment and the lessor "recycles" (i.e. crushes) the equipment to avoid creating a used/second hand market as an additional benefit.
Which option is a profit-oriented, not-totally-profitable AI outfit going to choose?
It's very simple: some is better than none
Clothing companies do typically put items on sale first. They'll even sell "at a loss"
The EU Ban on the Destruction of Unsold Clothing: https://transition-pathways.europa.eu/textiles/opinions/eu-b...
[1] https://environment.ec.europa.eu/news/ban-destruction-unsold...
If so, on a liquidation event, I'd imagine they'd have to be sold to compensate the shareholders.
In supply constrained markets the bidders set the price floor, not the manufacturer. A manufacturer produces a supply; they tender the supply for bids. The highest bidder gets the opportunity to purchase the supply. Everyone else has to wait for the next bid, the waiting is their opportunity cost, and is priced into their bid.
Theres machine learning algorithms in the mix, and it all works perfectly smoothly, until it doesn't.
Smoothly here, being defined as the markets emotional feelings about it all, not, the price went up.
The price went up because it could, as you pointed out.
I agree completely.
There are some wrinkles to this. If CXMT's RAM was somehow inferior, then the market could sustain two different prices for two different products. But it's difficult to separate the market like this without just making unusable trash. The one separation we do have - DDR5 vs HBM - works against the consumer's favor. One layer down in the supply chain, RAM manufacturers have to buy silicon wafers to etch circuits into, and HBM takes up treble more wafer space than DDR5 per gigabyte. In fact, this is specifically the scarcity that allowed Sam Altman to engineer a RAM crisis, by buying up a bunch of wafer supply that he could then redirect to more AI-optimized HBM.
We can also infer from all of this that CXMT - while producing lots of memory - does not have the scale necessary to actually quench the inflated demand. Either because their yields are shit, or because China wants to run a business and made the same calculation as the other RAM producers that the AI bubble would pop before they could get additional equipment online to service demand.
https://www.tomshardware.com/tech-industry/data-centers-in-n...
It’s not the same as dark fiber capacity being overbuilt for demand that wasn’t there yet.
Is this really true? Other than the occasional outages, the frontier labs don't seem to be getting overloaded by API pricing customers at peak times, and with Opus 5.5 using even less tokens (which translates into less GPU time and less money spent on the same tasks), it seems actual capacity isn't literally strained right now. If current needs are met by currently available capacity, where is this demand going to come from?
Maybe somebody will use them, but not the original owners.
I can directly confirm this fact, and I'm just some random dude out here in suburbia, who can only imagine what the Indiana Jones warehouse of glowing GPUs looks like in a data center.
We're all salivating over this imagined future where an AI toilet seat greets us every morning.
Who here has already asked AI how to make your own DRAM ? I know you have. Thats a sign of peak interest in a subject or market.
I'm not, but I can certainly believe the surveillance industry vectoralists are salivating over a future where their AI toilet seats greet each of their data subjects every morning.
Today, the "old" corporations like google, meta, amazon are also spending a lot of money.
Not according to this article[1], which says:
"I believe we are now in an inevitable overbuild situation, one with no neat, tidy Dot-Com Bubble-style exit story. Demand for NVIDIA GPUs — and those from Broadcom, AMD and other semiconductor companies — is driven by speculative capital believing that the AI industry will become magnitudes larger than it is today, largely driven by the fact that everybody believes there’s far more demand for compute capacity than actually exists. ... NVIDIA has created a remarkable illusion perpetuated by the media — that GPU sales are a direct measurement of the actual demand for AI compute, rather than a measurement of how a few companies are willing to invest in an idea two years in advance, using circular financing as a means of creating the sense that you must buy these GPUs now, or you’ll miss out on the future. ... At the very least, hyperscalers are going to be burdened with brutal depreciation charges or onerous write-offs for years to come, whether their capacity turns into revenue or not. ... I don’t see how 90%+ of NVIDIA’s sales ever end up generating a single dollar of revenue, and considering the amount of project financing-backed data center debt deals, there’s very little that exists to protect investors if AI compute demand never arrives."
The man saw the market demand for saying "AI bubble is going to pop and AI tech is going to wither away and die", and went to meet it on the supply side - truth be damned.
So far, AI companies still keep getting bottlenecked on compute, AI utilization increases - driven by, among other things, increased price-performance of AI making it viable in more and more roles. And the demand for both AI inference and AI hardware in general shows no signs of stopping.
HN discussion: https://news.ycombinator.com/item?id=49526069
I read Dan Luu's ... well, sort of aspirational takedown.
I don't buy it at all.
It has a massive unstated assumption: that what the big LLM vendors are saying is true.
It is not true. The whole pile of them are conmen.
The other hyperscalers are even further delayed. CoreWeave has some VeraRubin, but not in volume (as I understand).
A lot of its current use is just basic questions that could be handled by local llms just as well, yet they are being processed on GB200s on the other side of the world.
After the crash no one will need to build a new data center for 20 years, and there will be plentiful power connections
It’s what everyone does, from company to individual. When you sell your used car or list your house, you price it according to what the market will pay. When companies sell a product, they list it for what the market will pay.
The conspiracy theories and accusations of collusion were flying everywhere until CXMT came online. It felt like every comment section and headline was predicting that CXMT’s arrival was going to crush the cartel and smash the “price fixing” when a new competitor arrived.
Then CXMT memory arrived at nearly the same price as everything else because, to the surprise of nobody with basic economic understanding, they were selling their products at market rate too.
The high price is being driven by demand combined with a supply chain that takes years to increase output. Just prior to this demand explosion, RAM and SSDs were so cheap that producers couldn’t justify building out more capacity. It would have been a bad choice at the time unless they had a crystal ball that could have predicted the future.
I’m willing to spend a lot to drive down the price. As a show of good faith, I promise not to actually pay any more than 50% of what you can get on the used market.
Obviously, spending additional dollars on purchasing RAM will drive the price of RAM up.
Just as obviously, spending additional dollars on manufacturing RAM will drive the price of RAM down.
This particular market is pretty clearly not functioning well, and there are textbook Econ 101 reasons why a market may not function well. Of those textbook reasons, I suspect the most obviously applicable one here is large barriers to entry.
Econ 101 would be understanding that you can’t snap your fingers and have a new DRAM factory online overnight, and that this isn’t an indicator that the market is not functioning.
You’re defining a market “functioning well” as meaning the price is low for you and all demand is satisfied immediately without raising prices.
That’s not really what is meant by functioning markets. A functioning market does allow the price to increase as this is what spurs investment into new capacity and new developments.
It also allows the DRAM to go to the most profitable activities. This is actually what the whole Georgism and Land Value Tax people advocate for with property taxes, which bizarrely many people think they like: The idea is to raise the price of something (in that case, your property tax) to ensure it’s not being “wasted” on an economic activity that isn’t optimal. The most optimal economic activity is the one that can afford to pay the most, which is the one that captures the resource.
It’s not fun when your use case isn’t the one that can pay the most, but that’s literally how markets work.
Wasn't there this whole legal thing a couple of years ago involving most of the dram manufacturers?
You can quibble over my usage of “functioning,” but all I’m referring to are the textbook characteristics of perfect competition. I’m not claiming that someone is to blame if any of these conditions don’t hold. Of course they never hold perfectly in practice!
I’m not trying to blame anyone for the fact that it takes a lot of money and time to get a new DRAM factory online. I’m just saying that this fact is bad for essentially everyone except for the small number of current DRAM producers.
> You’re defining a market “functioning well” as meaning the price is low for you and all demand is satisfied immediately without raising prices.
Well, yes, that’s pretty much the only reason to advocate for competitive markets, isn’t it? If you don’t want allocative efficiency and productive efficiency, then what do you want?
A calculated move. If they're right and the AI demand is a fluke, they don't overinvest in expensive manufacturing infra, they don't get saddled with more capacity than they need, and there is no memory glut eating their margins in the future. If they're wrong and the AI demand isn't a fluke, they get to pad their margins big time, and the supply bottleneck isn't their problem. As long as they shake hands and bet on "fluke" in sync, the downside is very limited for all of them.
They shook hands, they bet in sync, and they were wrong. Now the supply bottleneck is everyone's problem.
Potentially even theirs - if Chinese vendors can use it to wedge themselves into the market when it's at its most lucrative, and invest in growth. Which they almost certainly will.
Let's not rewrite history: Sam Altman kicked this off by trying to corner the memory market to get an edge over his competition...in return for only a year's supply of memory. The resulting supply squeeze has been very profitable to the RAM manufactures, and will last many multiples of OpenAI's initial contracts, so they can't be too mad at him for bot disclosing his other wafer deals during negotiations. They can afford to gift him a thank-you yacht named the "Adverse Supply Shock"
>Only to see them all end up on the surplus market, unused, for a couple hundred ?
You realize to people that aren't you, this looks like holding two competing ideas at once.
And the comparison makes zero logic. Your library, government, employer are on the internet, who gives a fuck what dump their first modem went into
Lots of business owners here too that build their own clusters
2/ If it's a boom and bust cycle as he said, where's the bust? It's not coming. Humanity will use exponentially more memory and computing, as we've seen. That's not going away.
Let's have interesting conversations about the next steps for memory, is it quantum? what's up next? is humanity really beholden to a monopoly of private chip makers
Why is so hard to get self-hosted bare metal which uses all of the modules? Most server providers go for 1 stick and call it a day....
ffs
This boom is only superficially like the dot com bubble.
There are absolutely parallels.
But, a GPU that can help me write emails, code, articles, etc. has intrinsic value that a Cisco router simply didn't provide.
I don't believe GPUs are a new asset as Jensen Huang claims. (His take seems very bubble-ish.) But, I feel these elevated prices are going to last until production catches up.
If you're writing emails, you're gonna need to send them through a router or two before someone reads them.
> I don't believe GPUs are a new asset as Jensen Huang claims.
He sells GPU's, so taken in context, he's promoting the financing of GPU's, by anyone who is listening to his sales pitch. He travels in CEO circles, so, he's talking to other CEO's, and the banking industry.
The phrasing of it also directly pushes back against the articles accusing the AI industry of circularly financing itself to inflate everyones balance sheet.
And the other thing this idea of it being an asset that you invest in, like a house presumably, is that it primes all of us for 2 future realities: price increases and minimal performance gains ( "its useful for longer so finance it" ).
As to economic value proposition to any individual or group, it boils down to how they are going to use it to recoup the expense.
Don't just take the sales guys word for it, ever.
There have been many videos lately, on whats happening in the automotive market. The industry has been adding more and more tech, to the point where its now financially extractive via mandatory dealer interaction aka fees, and people are becomming fed up with it all, wanting something simpler for less money.
Translate this to the big fat GPU card industry, and how long is it before AMD/Intel add just enough AI ops into every CPU that noone cares anymore and they start asking themselves "Why am I buying 96GB of RAM and a GPU, and 96GB RAM and a CPU ? I'm buying ram twice ! It's a conspiracy !"
Unified architectures seem to be inevitable, GPU's would no longer need seperate ram, then shrinkflation is gonna hit your video card, and you're gonna pay GPU+RAM prices for just a GPU. The upside to this is that you could fit more GPU's in your box and they won't need 2.5 slots each. For now, they are hoping the DGX Spark type boxes fill the market demand, and allow them to sell higher priced products. Unified architectures would dent demand for those.
Just my 2 cents on it all. Enjoy the ride.
That’s what the cable company wants you to believe. We sent their damned boxes back. They wanted us renting a modem that never hangs up, then buying a router just so our own computers could talk to each other. Mine used to call yours directly over the telephone!
Now there’s a computer inside the modem, another pretending to be a router, and a mess of switches, gateways, and subnet masks to do what two modems and a telephone number used to accomplish.
They took the party line, put it on a television channel, called it broadband, and convinced everybody to rent the equipment.
In my day, when somebody tried to break into your computer, you could hear the bastard dialing. And when you were done with the Internet, you hung up. Now they want you connected twenty-four hours a day.
Just wait until somebody breaks into your television through that cable. You’ll wish you’d kept your rabbit ears.
I'm sorry, what?! Communications is a valuable thing. How is all that AI slop going to hit the Internet if there's no routers?
Which it to say, if models eventually do useful work, demand for substrate will continue to outpace production. I don’t envy the fabs that have to forecast how much memory the world will desire in 2030
If ML + robots can be produced for fewer resources than it takes to feed and house a human, then no reasonable amount of human labour will be sufficient to pay for robots+compute.
We may have a bust, with broad overcapacity and cheap, distributed ML compute. We may have the paperclip problem, where datacenters and fabs are taking over all reasonable resources and humans squeezed out. We may have everything in between.
I've heard 2040 (note this is not 2030), 2029, and 2027 for 'back to normal', 'new production online', and 'potentially a lower demand will be seen and still not enough supply', respectively. All hearsay from me reading news headlines.
Global inflation and f&&&ing Dump will impact all this ofc.
Something like 3200mhz DDR4 is still perfectly fine for 90% of consumer use. You really don't need more to look at cat pictures on facebook.
That's the plain truth: most sites/servers simply serve very very very little and have not anywhere near the need for 256 GB of RAM, let alone 512 GB.
These old servers aren't sufficient for the Amazon/Apple/Microsoft/Meta/Google/Oracle of this world, but for 99.9% of the rest of the world, they're more than plenty.
And the world and technology shall keep on marching: we'll have DDR6 and as soon as that happens, suddenly all the hyperscalers shall consider DDR5 to be oh-so-lame and dumped by the truckloads in landfills.
We may have two more painful years ahead for RAM but we'll then have plenty of used enterprise/datacenters gear on the market.
I still use it occasionally and it's really the old CPU and worn out SSD which slow it down.
Sorry.
EDIT: and $120 for the Kingston ones, albeit LP variant.
I think the reason why those processors are produced are because it rather makes sense to upgrade your old rig with them. Seems to be the only way actually to increase CPU sales beyond laptops ans smart phones.
I don't see more DDR4 ram being produced. Chinese brands are already starting to produce DDR5 and soon HBM. 'Bringing back DDR4' is IMHO a too bold claim.
[1]: https://www.electronicdesign.com/technologies/embedded/artic...
[2]: https://www.reuters.com/technology/nvidia-supplier-sk-hynix-...
[3]: https://arxiv.org/abs/2606.24616
[4]: https://www.apollo.com/insights-news/insights/daily-spark/No...
Remember too that these models only really became useful less than a year ago with Opus 4.6 being an inflection point. It takes a long time for the economy to reorganize itself so it's not surprising we're not immediately seeing the payoff in economic data.
I can't imagine it can even feel productive to anyone, unless you use agents to automate re-rolling your slop machines, but then you're just going to end up with an unmaintainable ball of mud. Either way, it's definitely net negative.
Kind of same outcome as if you just generated lots and lots of random code (perhaps at AST level; but don't tell your Claude, that'd trick it into making itself obsolete tech!) and compiled it in parallel till it passed tests, but this time around, the computer also tells you that you hit the nail on its head, that you're totally right <s>and that you're such a good puppy. agent uses tool `clicker`</s>
Update your priors.
Also, is “update your priors” some slang for “tell your parents”…? In that case, I can only say older folks are especially vulnerable to slop and I can’t imagine my parents would benefit from that tech.
Anyway, legitimate uses for chatbots are extremely niche, to the point I can’t really name even one such use…
> I can’t see how it’s productive to sustain a forever-junior?
90% of everything was already slop. The simple matter is that in many fields, slop is enough to clear the bar, especially when that slop makes the unit tests light up in green.
You know, what Microsoft has been trying with Palladium (can't risk general purpose computers! People might commit, oh the horror, copyright infringement, with those!)
Imho, as hardware keeps getting better, it seems to me this is the only thing that can make cloud investments worth it in the long term (which is only 2-3 years away) for hyperscalers. It's the only explanation. If OpenAI/SpaceX/Musk/... really thought AI would take over all jobs, they wouldn't be investing in compute, they'd be buying up farms, land, harbors, mines, bankrupt factories, fucking tax prep businesses ...
Make compute only possible within Goog/Amazon/Microsoft/... walled gardens ... that's the play. And this AI is a cute technology that conveniently needs lots of compute, but it's not even the actual core of the play. A very simple very old style monopoly grab, you know, the business plan people got convicted for trying in ancient Persia, 5000 years ago.
Unless prices get categorically higher, I don't think that's likely. Like 10x more than it is now.
Companies were buying computers for all their employees in the 90's, where a mediocre laptop could pretty easily cost two grand in 1990s money. My mom just bought a middle-of-the-road laptop for about a thousand dollars in today-money, and that was a consumer buying it; companies can generally spend more. "No choice but to use the cloud" feels like it's way too strong of a conclusion to draw.
Are you human? Then you must verify it from a "trusted computer", preferably with your driver's license. Gets governments and corporations as much information as they want, along with a guarantee that you can't "steal" their content and can't avoid their ads.
I have no doubt that Google and Microsoft will increasingly do things to force us to provide more data so they can do awful things to us.
You might see interest in thin-client plays for the same reason. They'd like to replace the local desktop with throwaway $50 terminals because it's easier to deploy on a standardized VM image and not have to replace dead SSDs and CPU fans.
Workstation tasks that require low-latency and high performance seem to stay local because the performance is simply not there.
Excuse my ignorance, do you have more details on this?
The Persian empire did not exist until round about 550 BCE.
Thus, the upthread statement,
> the business plan people got convicted for trying in ancient Persia, 5000 years ago.
Is somewhat nonsensical: 5000 years ago, Persia did not exist. I suppose we could be talking about whatever existed in that locale of the world around 3000 BCE. But that's what they're saying: "what do you mean, 'in ancient Persia, 5000 years ago', Persia did not exist then".
Prices are up a bit but there are plenty of "free-programmable computers" from a dozen or more manufacturers on the market. That's not going away.
I take your point that AI is pushing up the price of computers, at least temporarily, but you're leaning way into conspiracy and cynicism in your original post.
"AI" is really 2 companies, by extension funded/owned mostly by FANG, and is not pushing up the price of computers, it's buying ALL DDR5 capacity, processors, memory, and of course GPUs, including the fab capacity to build them in the first place. I don't know if you've recently tried to convince your boss recently to buy a new machine, but ...
Somehow it feels like we got a pretty good deal, in retrospect. (My server has been since replaced, and I can tell you, 256GB of DDR5 ECC was more than that! There's been a further doubling or two of prices since I bought, though. I think I could maybe sell half the RAM and finance the entire server…)
No end in sight.
Soon Zen 6 should become available (well, or not so soon). With 7200 MTs memory controller? I haven't seen such JEDEC modules at all. Are they possible without the additional clock driver (like on 6400 CUDIMMs)?
Speed has indeed decreased for the modules you can actually buy, while historically you would get faster and faster modules at lower prices in the years after the introduction is a new DDR standard.
Not much cheaper than DDR5.
There is definitely some exaggeration around timelines (2030 instead of 2027) and the industry has a cartel but most of it is real demand. From AI companies that don't have the money to pay for it and probably won't ever receive actual shipments but they locked down contracts ahead of time.
I hope so too, but I'm afraid that if China ever blockades Taiwan, then current RAM/GPU prices will look like an incredible bargain.
Now that's a pretty brave assumption!
- cheap memory: code is bloated and inefficient. Why do I need 2GB to fire up a text editor? Is that cross platform runtime really the best basis to run my app on? People need to feel the sting of a malloc saying no sometimes.
- cheap storage. It's so cheap that there is zero technical brake on storing everything. With the concomitant abuses of privacy. A widget website needs name, address, CC, not 134 pieces of information about my GPU and screen resolution along with my contacts list.
- cheap compute. we've all seen the Rube Goldberg Kubernetes/ factory contraption and it's only possible when the marginal cost of spinning up a new pod, sidecar, etc is minimal. When compute is cheap people outsource their architecture thinking to someone else's decisions.
- cheap network. Nothing new here: are apps sending a ton of gunk for no real purpose eg JS. Also related to compute complexity as systems span multiple nodes
There is a Q about how much RAM these phones will need and the average selling price.
People are seriously addicted to doom scrolling. But what are they willing to pay for a phone?
It's quite preposterous to think that an old 256GB RAM DDR3-using dual xeon server (128GB per NUMA node) that I had sitting around on a shelf gathering dust significantly appreciated in value over the past 12 months. But it's factually correct, looking at what comparables sell for for on eBay.
During every mining crunch that made GPUs very expensive, we sometimes went backwards on parts of new builds. A used GPU from years ago was often a more-viable choice than a new model in any kind of condition.
edit to add: A lot of folks in 3D printing like to have a small computer driving the machine, especially with tinkerer-friendly hobbyist printers of yore. That often meant some manner of Raspberry Pi because it was inexpensive-enough, small, and a known quantity that was simple to deal with in an appliance-like way.
During the covid-era chip shortages, Pi hardware was expensive, hard to find, or often both. So folks started driving their printers with things like old cell phones or cheap Android TV boxes instead.
I eventually upgraded to an RX 570 in 2018, and hit the same problem 5 years later.
So as far as loe to mid range cards are concerned you can wait 5 years and still have to pay more to replace it for an upgrade.
Also the enshittification of software might count? You used to own Photoshop for life, now you're expected to pay a subscription for life. And legal onslaught, too. Encryption and consumer VPNs are on the chopping block in much of the world at the rate things are going. The end goal of governments and corporations alike is to remove the ability of consumers to own hardware they control. We are rapidly moving towards a future where you will only be able to access the internet and government services with a "trusted device", which will be a thin client with a TPM, biometric verification, telemetry of everything you do on it, and which will require paying an ongoing subscription to use.
$699 in 2015 (when the 1080 was new) is ~$975 today, adjusted for inflation.
$975 is pretty close to the launch MSRP of a 5080, which is "just" $999. (We can't actually buy them for $999, of course, but we couldn't usually get 1080 cards at MSRP during the mining frenzy a decade ago either.)
Whereas today, with the 5000 series in the market for quite a while already, a 4080 is still well over MSRP
You could say that there was Titan but in practice they were not seriously considered. At least for gaming.
Businesses are competing like crazy to make algorithms that are in high demand. AI could theoretically let people and create new things that they would never be able to otherwise.
It is entitled to thing that people wanting GPUs and new PCs for gaming and hobbies is a more valid use. And if someone values it, all they have to do is pay the market price.
I get why they're upset (it's not fun to be crowded out) but it's a fundamentally selfish position.
These old workstation platforms seem to be perfectly fine for hosting GPUs for local llms.
My $20 xeons 2696v2 and 2670 are tool old for real offloading, and sata ssd drives are likely a bit slow for ssd streaming :)
30 refineries were closed in Europe in the past 15 years about, and that is just a small example of a pathological situation.
Cost of hiring an oil tanker has ballooned massively (like 40 fold) since America decided to bomb its supply chain, so these long supply chains are massively expensive.
The Republican Party has decided that running a world of global free trade was no longer desired. America wont pay the price for another decade or two, but it will.
Europe can easily be more dominant than the Americans if we chose it.
I think in a few hundred years we'll have learnt what a disgusting, poisonous system it is.
Among the many other things they won't have is an economic model.
As you like … but blaming capitalism is not going to bring industry back, and it won't stop China from using slave labour or India from using their rivers as landfills.
And I can already see the shitwank bean counters salivating over this prospect. Imagine the possibilities, we can stuff as much DRM as we want, charge as much as possible monthly, arbitrarily restrict the terms of usage in the future. What will the user do? Run the same heavy duty software on their Intel N150 + 4GB RAM shitbox? Blah.
LPDDR5 is slower but maybe a small cluster could be useful?
The amount of waste produced due to phones not receiving new batteries and software updates was utterly insane, given what goes into making them.
It's terrible if you're a labourer in India who buys phones under 10k INR (~100 USD) when your salary is 20k INR per month.
Unfortunately there are way more in the latter category than the former, and most of them won't know what ram is. They'll just be paying more and getting less going forward for a few years.
(Source, the salaries and phones I've seen of the staff at office. And we're in a major region you've likely heard of, not some tiny rural village.)
That too sounds like an actually good development all things considered. For the planet, anyway. And with the big caveat of that cloud tether, though just for scaffolding.
It would be good if we would use less resources.
But we are not.
It's just that some people don't get the newer resources.
and if it is for your child, read them books.
CXMT is planning expansion for the ram capacity, they’ll do that faster than the west, I beleive they said end of year ish? (Could be wrong).
once it comes online and they increase their yields, this whole thing changes.
Also new datacenters mean increased supply so cheaper tokens now that parameter size is a non-sequitor.
I’m legit curious how long China holds back… but I doubt it’ll be long.
Could totally be one of those situations where the current cartel outsource all their production for MoW outside the US to them and then just pocket the investment money lol
However … without the US datacenter buildout ramping continuously … the ram will eventually make its way from warehouses to the market lol. We really need a way to use server dram in consumer boards… but idk if that will happen.
Although it would take a ton of engineering and probably a new chip … a custom motherboard would that did that would be legit haha :) would solve a lot of issues.
My S21 Ultra phone is an e-waste because of broken screen. But it has 12GB of LPDDR5 and 256GB of decent flash memory. I don't have knowledge and tool to recyle/reuse those priceless parts.
I wonder, how many TB(or PB) of RAM/SSD being thrown to garbage dump every month.
I mean it somehow still feels cheep, if I watch the prices for new hardware.
But it took me days to create this upper mentioned system from all the different sellers and different sources on eBay. There is no "cheap" one-click solution with included guarantee anymore.
Home server also runs ddr4 and does ok, but that has been hitting more cpu bottlenecks now rather than memory.
That said its kinda sad personal users are getting tossed back a generation due to current generation ram being almost all gobbled up for AI.
I avoid dedicated GPUs which have always been a money-pit. I find modern video games dull and boring, and this is not offset by them being very very pretty. Gaming for me peaked with Quake II and there's been nothing of real interest since Portal and Untitled Goose Game.
A year and a half ago I bought 2 new 8G DDR4 for 47 USD. I was going to use it to upgrade my W541 from 16G to 32G. The W541 has 4 slots. But seems the one I have maxs out at 16.
Now I have another reason t hang on to the RAM :)
"Somehow, DDR4 has returned."
__
Besides that, wasn't DDR4 good enough for everything we did, anyway?
At least I did not notice all that much of a difference when migrating to my current DDR5 system other than the CPU being massively more performant. Though I am sure that the memory did play some role there.
Clanker claims that the felt absolute latency between decent DDR4 and DDR5 should be similar, while bandwidth might be lower, apparently? Something something CAS latency.
If you have some unused extra memory it will be used as disk cache, which also helps in the responsiveness of a desktop. Waiting for the disk is orders of magnitude slower than reading from memory cache.
The RAM shortage is bringing back old compute hardware
If my DDR4 RAM go kaput I'm screwed just as I would be on DDR5.
Also, the article makes no sense. Why would I want to pay more for DDR4 now? They should tax those AI companies that are responsible for this.