In the Dwarkesh podcast with Semi-Analysis's Dylan Patel they forecast the phone market will shrink by 50% this year because of RAM prices:
But that’s the high end of the market, which is only a few hundred million phones a year. Apple sells two or three hundred million phones annually. The bulk of the market is mid-range and low-end. It used to be that 1.4 billion smartphones were sold a year. Now we’re at about 1.1 billion. Our projections are that we might drop to 800 million this year, and down to 500 or 600 million next year.
We look at data points out of China from some of our analysts in Asia, Singapore, Hong Kong, and Taiwan. They’ve been tracking this, and they see Xiaomi and Oppo cutting low-end and mid-range smartphone volumes by half.
Yes, it’s only a $150 BOM increase on a $1,000 iPhone where Apple has some larger margin. But for smaller phones, the percentage of the BOM that goes to memory and storage is much larger. And the margins are lower, so there’s less capacity to even eat the margins. And they have also generally tended not to do long-term agreements on memory.
Why this is a big deal is that if smartphone volumes halve, that drop will happen in the low and mid-range, not the high end.
At work we just got a quote to upgrade a couple servers, original price a few years ago was ~ $150k. Essentially the same hardware, just newer, is now quoted at ~ $450k.
We decided to just keep our current hardware for now and extend a support contract for ~ normal price.
Yeah. Not true. Or send me the name of your server vendor. I’m buying.
Having issues with both price and availability on NVMe, SATA flash, starting to see some CPUs, and for a personal project high density spinning rust (24TB+).
You're right that fuel prices have risen. But usually the impact of fuel prices is mostly felt on bulkier, lower cost items first.
After all, a truck can carry a 10kg sack of rice, or a 10kg nvidia gpu. If shipping costs for 10kg rise by $15 the sack of rice has doubled in price, but the GPU is only 0.5% more expensive.
For a truck yeah, but across the ocean, it isn't quite that simple because GPUs and grains are sent in different types of ships (or different modes entirely) that aren't interchangeable.
This is driven by AI datacenter demand, not fuel prices. RAM prices have actually dropped significantly in the last couple days as the Iran war hit and the possibility that interest rates might go up and pop the AI bubble sunk in. (Though let’s see where they go after the last couple days of whipsawing.)
Helium supply issues are only going to make this worse.
I feel like for the first time in our lives we might have seen peak technology for the next few years. Everyone is going to have to make do instead of depending on ever increasing performance.
Helium is almost all captured from gas wells by cryogenically liquefying the nitrogen out of it. I guess you could do technically do that with the fab's air but it is a LOT of volume of air to liquefy and likely costs more than even inflated helium prices.
Most helium from most wells is simply vented because it is expensive to separate even with its relatively high concentration, and I imagine even the best case scenario for capturing it from a fab has abysmal concentration of helium. But because most of it is vented it also means if the capital is put down to build more helium separators on gas wells it wouldn't take long to increase supply. Short term for a year or two it can be a problem, but beyond that it is simply a cost versus demand issue. There is neither a technological nor source limitation, it is a pure capital investment limitation.
AFAIK they recapture most, but recapturing all simply isn't possible / financially feasible. And they use a lot of helium, so even if they capture most of it, the losses are still higher than the currently available supply.
Finally, good efficient code is going to get its moment to shine! Which will totally happen because it's not like 80% of the industry is vibe coding everything, right?
Even before the hikes, SBCs were $80-$100 a pop, compared to pennies for basic MCUs and maybe $5 for high-performance ones. People were clearly willing to pay 100x more just for familiarity and the ecosystem ("hats", forums, etc). I don't know if 300x is going to make more hobbyists see the light, or just result in fewer of them being able to afford the hobby?
I’ve been having a lot of fun with the Pi Pico 2W. It can host an access point, a web server, be a USB host, and of course has GPIO. And not running an OS means it’s way simpler.
There are ups and downs in the prices of components. Often people forget that during COVID prices were high for SBCs because of supply chain issues. Video cards just were not available in the UK and afterwards (every supplier had long lead times) and are still relatively expensive (at least there are now lower priced options). Raspberry Pis you couldn't get hold of and many people (Jeff Included) was using a website checking for availability which was non-existent for anything other than low end models.
I remember 15-20 years ago when hard drive prices went up through the roof because there was a flood in Thailand and it too years for prices to come down.
There is going to be supply chain issues due to the current Geopolitical situation (Helium comes out of the Gulf and that is need in chip manufacture) is also going to affect the price of components.
Eventually in a few years (as the article states) the situation will change. It just sucks at the moment.
TBH I am more worried about my ability to fill up the tank on my car as both Petrol and Diesel is unavailable locally. I can make do with whatever computer equipment I have.
> People are quick to forget that during COVID prices were high for SBCs because of supply chain issues.
inb4 AI has the same supply chain effects as a worldwide pandemic. I guess those AI doomers that talked about it being the end of the world had it right!
There is a saying that is often trotted out my economists "That the cure for high prices, is high prices".
There is a consumer market and business need for DRAM outside of AI. Someone will fulfil the need as there is a high incentive to. It just going to take a bit of time for this to happen. My equipment is going to be fine for another few years. So I am going to just hang tight and make do with what I got for now.
Main producers actually reduced dram output in 2026. When you have few players with very high capital cost you will end up with cartels like light bulb cartel.
Someone will come in when the price goes up enough. It will take time, but it will happen. What people are complaining about is that the time for this to happen is too long.
Oh look, there is a player coming into the market it seems:
In reality were they going to survive anyway? I would wager likely not.
Raspberry PI is the defacto standard for SBCs. Almost all the other SBCs had significant problems usually around software support and also third party support e.g. Hats, cases etc.
The price for a couple of 32GB sticks is now over $1200 after being stable at about $200 for several years until last September. That's not a blip; that's 6-fold hike and there is no sign it is slowing down any time soon.
Did you not read what I said? I couldn't even get a replacement video card at any price during the height of COVID and believe you I had the money to pay for one. I couldn't even get a Raspberry PI (any model) for about a year. They were constantly out of stock.
> That's not a blip; that's 6-fold hike and there is no sign it is slowing down any time soon.
How does that invalidate anything I said? As states in the article this will change, it will take years but it isn't forever.
I find it hard to believe that people here cannot make do with whatever hardware they already have.
I also don't believe those small SBCs would have survived long term anyway. Most people just use a Raspberry PI. It is either a MiniPC or a Raspberry PI.
Discord groups that had real-time line counts and pictures of the line at most best buys across the country (US).
The only way I got one was overpaying and a lottery system that bundled it with other hardware because they knew everyone would still buy it. It was impossible to buy online normally as you needed some kind of automated way to buy it before stock zeroed the minute it was posted.
You could pay a scalper for a gfx card, but stores had none. Now, stores have RAM at least.
> Did you not read what I said? I couldn't even get a replacement video card at any price during the height of COVID and believe you I had the money to pay for one.
You're comparing to memory sticks that went up 6x. If you were offering anywhere near 6x MSRP and you couldn't get a video card... I don't believe you.
> If you were offering 5x MSRP and you couldn't get a video card... I don't believe you.
My 1080Ti had died. I had to use a 8800GTS from the late 2000s for about a year. As that was the only GPU I had. I have no iGPU on my CPU.
There was at one time, no stock available. Not on Amazon, Not on Overclockers, Not on Scan. They had some weird lotto system taking place on most sites.
Scalpers claimed to have cards. But I wouldn't risk sending a lot of money to some random seller on ebay.
> Unless this article is massively misleading, sure it was out of stock at 1x price but it wasn't out of stock at 2-3x price.
Again I am in the UK. You could not buy any PI other than 1GB model and maybe the zero. Both of which were useless to me.
Okay, UK, maybe that changes things more than I expected. But what about ebay and the sites that replaced classified ads? And is it unreasonable for me to say that you could have bought a US listing and had it reshipped?
Edit since you added: Scalpers claimed to have cards. But I wouldn't risk sending a lot of money to some random seller on ebay.
Even with ebay's buyer protection?
Well not to be mean but I think "I refused to use ebay" invalidates your claim that you couldn't buy a card.
Those will dry up soon enough. Corporate laptop refreshes will be drawn out as they try and cost save on the increased price.
You also better hope the aliexpress dont figure out a way to get the RAM out of those things because they will start harvesting it for sure if there is money to made.
We're talking about a pi replacement. The Pi 5 is slower than a 10yo laptop. That's gives us a very vast pool of used laptops.
> You also better hope the aliexpress dont figure out a way to get the RAM
That is a real worry and I can see used machines being gutted because selling DDR3/4/5 sticks is way easier and profitable than the whole machine. Adapters for SODIMM to regular DIMM are readily available and cheap, too.
For most older laptops it's easy enough, you just open them up and take the RAM sticks out. There are SO-DIMM to DIMM adapters to fit a laptop memory stick in a DIMM socket.
Taking a big, complex, already well optimised program like Chrome or the linux kernel and optimising the memory footprint is hard. But 90% of programs are just crappy web apps that nobody has even bothered to optimise at all. (Sometimes wrapped in electron or something.)
If you go look, you often discover that 90% of the requests are useless, or at least could be combined. That 60% of bandwidth is used up by 3 high res images which get displayed at 30x30 pixels. That CPU performance is dominated by some rubbish code that populates an array of a million items every call, then looks up 1 element then throws the whole thing away, only to regenerate the exact same list again a few microseconds later.
We have plenty of RAM. In absolute terms, 8gb of ram in the macbook neo is 8 billion bytes. 64 billion ones and zeros. You don't need rocket science to make a CRUD app that runs well with that much ram.
Computers don't get slower over time. If we were merely as lazy with computing resources as programmers 10 years ago, most programs would scream on modern hardware.
You can still write JS or TypeScript code that tries its best to keep memory use under check. JavaScript was around in the late 90s when the memory footprint of software was at least an order of magnitude lower, so it's absolutely doable.
You don't have to go that deep. 99% of the time our analytics or risk management teams have some really memory inefficient Python and they want me to write them one of our "magic C things" it turns out to be fixable by replacing their in-memory iterations with a generator.
Most people don't have the chance to do that, but hopefully we can see some other languages get first class access on the web. At least there is the whole WASM project.
Just rewrite your biggest memory hogs in Rust, it routinely slashes RAM footprint and demand for RAM throughput. The effect is even bigger than the typical reduction in CPU use. You can even ask AI to help you with the task, it will use a lot less RAM for it than the rewrite will save down the road.
Why is that less realistic than saying 'rewrite in rust, make sure there are no memory leaks'?
My point, which I should have been clearer with, is that we aren't at a state where you can just one shot a rewrite of a complex application into another language and expect some sort of free savings. Once we are at that state, and it's good enough to pull it off, why wouldn't the AI be able to pull it off in C as well?
You don't have to trust the AI to do it with Rust, you just have to ensure certain conventions are followed and you can formally prove you're 'safe' from certain classes of issue, no AI magic dice-roll.
A lot of people are very excited by the idea that now language capabilities (and almost every other technical nuance) somehow don't matter but much like gravity they will continue to assert themselves whether you believe in them or not.
So far humans have proven unable to write large apps in C without those issues, given their work is the training basis for LLMs this creates two problems, one being that they don't 'know' what a safe app looks like either and any humans reviewing the outputted code will be unable to validate that either.
LLMs are great at C, probably because C is historically the most popular language in the world, by far. It only declined slightly very recently. But there's insane amount of code written in it.
Based on a FIDO2 spec I used it to write a reasonably compliant security token implementation that runs on top of Linux USB gadget subsystem (xcept for attestation, because that's completely useless anyway). It also extracted tests from an messy proprietary electron based compliance testsuite that FIDO alliance uses and rewrote them in clean and much more understandable C without a shitton of dependencies that electron mess uses. Without any dependencies but openssl libcrypto, for that matter.
In like 4 hours. (and most of that was me copy pasting things around to feed it reasonable chunks of information, feature by feature)
It also wrote a real-time passive DTLS-SRTP decryptor in C in like 1 hour total based on just the DTLS-SRTP RFC and a sample code of how I write suckless things in C.
I mean people can believe whatever they want. But I believe LLMs can write a reasonably fine C.
I believe that coding LLMs are particularly nice for people who are into C and suckless.
am I crazy for thinking that the 16GB Pi 5 is just there to absorb money from people who purchase the most expensive version of things? Like really nobody needs that much RAM on a Pi?
I am running a bunch of stuff on my 8GB Pi and I've run out of memory to put more stuff on. I use it as a low power server running a bunch of Docker containers. Some of these require at least 200mb and some use 2G of memory.
I was going to buy a small nuc and load it up on memory but I've acquired an old Mac Mini with 16GB of ram, which will do.
Yes, you are crazy for thinking that. The extra ram is useful for small LLMs and also running lots of dock containers. The very low power consumption makes it ideal for a low end home server.
I use the 16GB SKU to host a bunch of containers and some light debugging tools, and the power usage that sips at idle will probably pay for the whole board over my previous home server, within about 5 years.
Docker is about containerization/sandboxing, you don't need to duplicate the OS. You can run your app as the init process for the sandbox with nothing else running in the background.
I think that on linux docker is not nearly as resource intensive as on Mac. Not sure of the actual (for example) memory pressures due to things like not sharing shared libs between processes, granted
No you are not crazy. It's silly to try to use a raspberry pi 5 16GB (or equivalent priced product) as a desktop workstation with a GUI on it when much better actual x86-64 based workstations exist. Ones with real amounts of PCI-E lanes of I/O, NVME SSD interfaces on motherboard, multiple SATA3 interfaces on motherboard, etc. In very small form factors same as you'd see in any $bigcorp office cubicle.
It’s an incredibly lopsided machine. The Pi 5 is decently powerful, but you really really should not be attempting to use one as a desktop replacement. While theoretically possible you are so much better off with a $50 used SFF PC.
The SBC markets been on life support for a long time. Youtubers making videos about them don't seem to grasp that and keep pumping out reviews and projects like its still 2019. The pi specifically has plummeted in popularity and for most use cases they just aren't a cost effective option when second hand micro pcs are dirt cheap and vastly more capable.
I don't think comparing new Pis to used micro pcs is fair. Compare a _used_ Pi with a used micro pc. If you have any geek friends, it's probably not hard to find a used Pi for free.
The extreme DRAM market has had an unexpected side effect of triggering a lot of panic buying. I know several people who delayed PC upgrades for years but then panic bought new systems in this market. The trigger was seeing all of the "It's only going to get worse" and "This is the end of personal computing" headlines.
They're already regretting spending so much now that prices have started to tick downward.
I keep telling everyone: If you don't have a pressing need to buy right now, please wait 6 months and check again.
wasn't "panic" buy but I built a new comp early 2025, cuz at worst case would be complete supply crash and at best case it was going to be more expensive.
Def don't regret doing that, though I regret not springing for the extra RAM.
That's actually a reasonable response to market volatility and illiquidity. It's not just high prices, but prices that still fail to be representative of the actual market stance despite the rises.
Yep. I just bought a Pi CM5 for my son, for his ClockworkPi uConsole. CAD $200 for the 8GB module. I bought a whole Pi5 16GB not long ago for under CAD $200.
I will not be buying any more SBC's at this price point. I wonder if Raspberry PI will survive.
Got my RPi 5 16GB quite a while ago for around $160 and already thought that was expensive... It’s still powerful enough for almost everything I throw at it, honestly a bit overkill in most scenarios.
With prices steadily going up, for me it's starting to feel more sensible to repurpose the RAM sticks I've collected from old PC builds / laptops and just throw together small amd64 boxes instead of buying more RPis.
I wonder if there are low power Intel or AMD boards that accept DDR3. So many sticks of 2 / 4 / 8GB DDR3 inside laptops going into recycling or landfills which would do perfectly fine for low power purposes. Hell, performance for standard workloads scales with access times, not bandwidth, and DDR3 sits nicely at CAS8 1600MHz and CAS10 2133MHz..
The main cost input is presumably ram. They are passing it through.
If everything on the board but the ram costs $30, and ram is going from $10/gb to $20/gb, then they have to change the price $50 -> $70 to break even on the 2gb board, and $190 -> $350 for the 16gb board.
In other words, the raspi is now priced like a stick of ram with a bonus computer attached because ram is massively more expensive than the rest of the computer.
Bought a couple of 32gb SBCs before this all hit the fan. And also built a SSD NAS before the wave hit.
So timed that all pretty great. What worries me is my desktop is up for a full new buy somewhere around early '28. That could be a train wreck depending on how taiwan situation goes
> So timed that all pretty great. What worries me is my desktop is up for a full new buy somewhere around early '28
That's a very specific date / timeline. How do you decide to do a full new buy? I ask because I own a desktop that I built 15 years ago which I was flirting with replacing completely last year, but unfortunately I didn't pull the trigger ... oops :(
My old rig is still going strong. The motherboard can only take up to 32GB DDR3 though. CPU is an Intel i7-4790k which is still very fair today if you are not running a resource hog OS (looking at you Windows). Overall it is completely serviceable for my needs. Being honest with myself the only reason I wanted to upgrade was for nerd cred but I don't game much anymore and don't do any ML tasks that require lots of local compute.
Startup costs measured in the billions, with no guarantee of success, and a long payback time horizon in a market that almost everyone thinks is - in one way or another - a bubble.
Oh yeah, the market is also getting intense scrutiny from powerful geopolitical entities that are quite explicit that they don't believe in fair play or consistent, stable rules.
After discovering Dell Alienware clearance and graphics card availability in those Alienware computers, I haven't felt the need to build a computer for the last five years.
We're somehow in a race between LLMs curing cancer, destroying the planet by "You're right to be mad, I shouldn't have issued those launch codes, it's even in my Claude.md file, I'm sorry," and rendering modern technological civilization uneconomical. I know this is statistically the best time in history to live, but lord, I could use a vacation.
It's terrible. Fake money is fueling the exhaustion of real resources in search of questionable outcomes ("AGI"). Imagine if all of these money were invested in curing cancer.
Let's also imagine an alternative reality where some reasonable percentage of the $2.5T in current year AI spending was instead invested in the "general intelligence" researchers we already have for the same purpose. I think it's a pretty reasonable expectation that 1) they'd probably make more progress and 2) that money would help a lot more people in the process (through jobs and economic activity).
Obviously there's no "evidence". Why would you even think we need AGI? But I'm happy to hear your reasoning if you were one of the few/only? people who imagined that software that could predict the next word could do what it now is doing.
I've already seen at least one person who was pretty sure that the preprint paper they co-authored with AI (read: AI wrote for them) was going to cure cancer and make them billions of dollars.
There was only one problem. The paper jumped straight from "this paper will show how our new treatment could cures cancer forever" to "as you can see, these results clearly show that our treatment cures cancer" - with neither any actual results nor any specifics on the treatment. And I don't just mean that the paper didn't go into details; writing the paper was the full extent of their "research".
AI was used fundamentally for COVID vaccine development. AI is used for research in all modern drugs. It’s a certainty if cancer gets cured AI will have played a fundamental role since it’s already fundamental to precursors.
Unless you're really using the GPIO pins or other weird I/O, I really fail to see the purpose in having an 8GB or 16GB RAM Raspberry Pi (at a much higher price than it used to be) as a desktop workstation with a GUI on it.
The idea of putting sixteen gigs of RAM in a raspberry pi is nuts. The legit thing you want to use a raspberry pi (or a competitor) for as an embedded headless thing with no KB/mouse/display attached should run fine in 2GB of RAM or less, assuming an ordinary debian-based OS environment.
I would much rather have a used, ex-corporate/ex-lease, small form factor or ultra small form factor x86-64 desktop PC (Dell, HP, Lenovo, whatever) with 16GB of RAM in it and an SSD on a SATA3 or NVME interface. Whatever is the "best" SFF that you can buy via huge eBay used equipment dealers on any given month.
Despite being many years old, whatever you can buy on ebay for 200 bucks (at least before the recent RAM fiasco) with some recent-ish quad core core i5/i7 or Ryzen in it will run circles around a raspberry pi 5.
Is there anything (technically) preventing SBC manufacturers adding SODIMM slots?
I was expecting the Milk V Titan to avoid this memory nonsense since it has two unpopulated DDR4 slots, but it has fallen off the radar like several other SBCs.
SODIMMs are huge compared to a BGA memory package which is a problem if your goal is to minimize your board size (e.g. I don't think there's a reasonable way to fit it into a Raspberry Pi form factor without something weird and expensive like a mezzanine connector). Routing the signals is also somewhat more annoying because they all come out of one edge of the connector compared to a BGA package which has them fan out in every direction, giving more space for length matching traces, etc. You'll likely need additional PCB layers compared to a BGA chip.
The title should say: "Collusion of large corporations promoting LLMs with RAM manufacturers is killing the hobbyist SBC market (and bankrupting anybody trying to get a PC or laptop)".
Because we all know that DRAM prices have spiked since production is going to those infernal chatbot training data centers. Same as a lot of the electricity in some parts of the world, BTW.
Can you elaborate on the collusion aspect? Is the implication that OpenAI and Anthropic are coordinating their purchases in such a way that they target the hobbyist market? What’s the collusion angle here?
Only works so long as you eventually pay up... well unless the manufacturers make too much this way. That said are there some Chinese manufacturers that aren't part of the cabal and could undercut them?
Except that it doesn't work like that. If you buy DRAM and don't do anything genuinely worthwhile with it, you'll ultimately dump it all right back onto the market, and everyone knows that. The biggest worry is that it's actually OpenAI and their direct competition starving the rest of the market because they predict AI research and the like to be a highly valued use for the stuff, compared to building gaming PC battlestations or whatever the highest-valued use was before. Many observers think that this will also happen with GPUs and cutting-edge digital logic more generally.
> We all know that DRAM prices have spiked since production is going to those infernal chatbot training data centers
I know it's very fashionable here to talk about capitalism as some hand-washes-hand big corp organized scam, but if you put that ideology aside for a moment, you contradicted yourself here, I think.
I personally don't like conspiracy-theory-thinking. If I was a DRAM manufacturer and had to choose between servicing a single customer, who orders hundreds of millions worth of my product, or service a very large number of customers who order tiny amounts of the product a piece, then of course I would focus on the large client, because they are easier to service for the expected profit margin. I wouldn't even need to think about advertisement, sales, all that jazz. Looking at it from that perspective, it seems pretty logical to me that a spike in demand from datacenter operators would rise prices dramatically. I struggle to see room for collusion / conspiracy here.
A couple of issues, first there is a history of price collusion (see DRAM price fixing scandal on Wikipedia) and while it may be "logical" from a seller point of view to prefer large orders, this upsets a lot of people and used to be illegal in the United States (it may still be illegal, but it's not enforced)
Not everyone earns tech bro salaries and can sustain a thousand cuts. Many hobbiests are scraping and saving money to acquire hardware. For some it very well msy be the end of their world.
We are talking about brand new latest gen hardware here. People with low budgets are always scraping and saving for deals and don’t need to buy something brand new from a pricey brand name like raspberry pi.
You can still jump on eBay and buy all kinds of dirt cheap used pieces of hardware.
My buddy just bought a used ThinkPad T14 with 32GB of RAM and 1TB of storage for about $500. You can get by with a whole lot less.
In this context, I will also present the idea that Rasperry Pi has represented quite poor cost value for many years now.
Have you looked at how expensive international shipping is? eBay covers just a few countries, the rest of us can't buy there because we'll be paying 10 times the cost of hardware to get it over here.
I already moaned about this recently, but to briefly reiterate: the only hardware that's becoming available for most people in my region are Frankenstein desktops built from heavily used 10+ year old Xeons running on suspicious motherboards made by obscure Chinese manufacturers you've never heard of. This is pushing ever more people towards smartphones and away from actual computers.
But at least we got the bullshit machine in return, that's something, I guess.
> Have you looked at how expensive international shipping is?
It really shocks me how bad shipping has gotten. It's nearly unaffordable to buy things on eBay from the US as a Canadian due to shipping costs, so I can only imagine just how bad it is for people from other countries.
Yes, 90%+ of sellers refuse to ship here (and we're not even under any sanctions and/or political pressure of any sort). I hear about these magical 100$ Thinkpads all the time; I'm yet to see anything cheaper than 300$ (add another 100$+ for shipping).
In the Dwarkesh podcast with Semi-Analysis's Dylan Patel they forecast the phone market will shrink by 50% this year because of RAM prices:
But that’s the high end of the market, which is only a few hundred million phones a year. Apple sells two or three hundred million phones annually. The bulk of the market is mid-range and low-end. It used to be that 1.4 billion smartphones were sold a year. Now we’re at about 1.1 billion. Our projections are that we might drop to 800 million this year, and down to 500 or 600 million next year.
We look at data points out of China from some of our analysts in Asia, Singapore, Hong Kong, and Taiwan. They’ve been tracking this, and they see Xiaomi and Oppo cutting low-end and mid-range smartphone volumes by half.
Yes, it’s only a $150 BOM increase on a $1,000 iPhone where Apple has some larger margin. But for smaller phones, the percentage of the BOM that goes to memory and storage is much larger. And the margins are lower, so there’s less capacity to even eat the margins. And they have also generally tended not to do long-term agreements on memory.
Why this is a big deal is that if smartphone volumes halve, that drop will happen in the low and mid-range, not the high end.
DRAM pricing is killing the everything market.
We just had a vendor uplift our quote 50% per unit for some machines because of a mix of memory + supply chain issues.
At work we just got a quote to upgrade a couple servers, original price a few years ago was ~ $150k. Essentially the same hardware, just newer, is now quoted at ~ $450k.
We decided to just keep our current hardware for now and extend a support contract for ~ normal price.
That’s strange, there aren’t wider market supply chain issues outside of DRAM. Maybe your vendor is just throwing excuses around.
Yeah. Not true. Or send me the name of your server vendor. I’m buying.
Having issues with both price and availability on NVMe, SATA flash, starting to see some CPUs, and for a personal project high density spinning rust (24TB+).
>That’s strange, there aren’t wider market supply chain issues outside of DRAM.
GPUs, ram, ssds, hdds, hell even CPUs are starting to climb in price. It's an everything shortage and it's only getting worse.
A workstation that two years ago cost $3,000 was $10,000 last month and $10,500 this month. There are parts which aren't available at any price.
Fuel price rises = logistics price rises.
You're right that fuel prices have risen. But usually the impact of fuel prices is mostly felt on bulkier, lower cost items first.
After all, a truck can carry a 10kg sack of rice, or a 10kg nvidia gpu. If shipping costs for 10kg rise by $15 the sack of rice has doubled in price, but the GPU is only 0.5% more expensive.
For a truck yeah, but across the ocean, it isn't quite that simple because GPUs and grains are sent in different types of ships (or different modes entirely) that aren't interchangeable.
This is driven by AI datacenter demand, not fuel prices. RAM prices have actually dropped significantly in the last couple days as the Iran war hit and the possibility that interest rates might go up and pop the AI bubble sunk in. (Though let’s see where they go after the last couple days of whipsawing.)
DRAM is up more than that 50% though.
Flash has supply (and price) problems too.
I assume this is sarcasm.
This isn't true: NAND flash prices are up too, though not nearly as dramatically. But the war means that fuel and shipping prices are way up as well.
They’re throwing something around.
SSDs and HDDs are being squeezed as well.
Don't forget SDCards
"Memory card prices have TRIPLED in the last few months: when will this madness stop?!" https://www.digitalcameraworld.com/cameras/memory-cards/memo...
Sony stopped making their cards entirely, which stinks because I'd settled on their pro cards for all my camera bodies.
We just had a vendor tell us none of the HDDs we were looking for were available unless we also committed to a full NAS offering.
uh, prthspd you've heard of the third world war started in Iran?
Helium supply issues are only going to make this worse.
I feel like for the first time in our lives we might have seen peak technology for the next few years. Everyone is going to have to make do instead of depending on ever increasing performance.
I expect my 5 year old desktop will last a lot longer, but start worrying about the bathtub curve.
Ultra clean rooms with massive air handling systems can't recapture all their helium?
Or is this just a temporary thing based on where processing is located?
Helium is almost all captured from gas wells by cryogenically liquefying the nitrogen out of it. I guess you could do technically do that with the fab's air but it is a LOT of volume of air to liquefy and likely costs more than even inflated helium prices.
Most helium from most wells is simply vented because it is expensive to separate even with its relatively high concentration, and I imagine even the best case scenario for capturing it from a fab has abysmal concentration of helium. But because most of it is vented it also means if the capital is put down to build more helium separators on gas wells it wouldn't take long to increase supply. Short term for a year or two it can be a problem, but beyond that it is simply a cost versus demand issue. There is neither a technological nor source limitation, it is a pure capital investment limitation.
AFAIK they recapture most, but recapturing all simply isn't possible / financially feasible. And they use a lot of helium, so even if they capture most of it, the losses are still higher than the currently available supply.
Finally, good efficient code is going to get its moment to shine! Which will totally happen because it's not like 80% of the industry is vibe coding everything, right?
OTOH things which belong on microcontrollers are now being pushed back to microcontrollers for cost reasons, so there is a win to be found there.
Even before the hikes, SBCs were $80-$100 a pop, compared to pennies for basic MCUs and maybe $5 for high-performance ones. People were clearly willing to pay 100x more just for familiarity and the ecosystem ("hats", forums, etc). I don't know if 300x is going to make more hobbyists see the light, or just result in fewer of them being able to afford the hobby?
I’ve been having a lot of fun with the Pi Pico 2W. It can host an access point, a web server, be a USB host, and of course has GPIO. And not running an OS means it’s way simpler.
There are ups and downs in the prices of components. Often people forget that during COVID prices were high for SBCs because of supply chain issues. Video cards just were not available in the UK and afterwards (every supplier had long lead times) and are still relatively expensive (at least there are now lower priced options). Raspberry Pis you couldn't get hold of and many people (Jeff Included) was using a website checking for availability which was non-existent for anything other than low end models.
I remember 15-20 years ago when hard drive prices went up through the roof because there was a flood in Thailand and it too years for prices to come down.
There is going to be supply chain issues due to the current Geopolitical situation (Helium comes out of the Gulf and that is need in chip manufacture) is also going to affect the price of components.
Eventually in a few years (as the article states) the situation will change. It just sucks at the moment.
TBH I am more worried about my ability to fill up the tank on my car as both Petrol and Diesel is unavailable locally. I can make do with whatever computer equipment I have.
> People are quick to forget that during COVID prices were high for SBCs because of supply chain issues.
inb4 AI has the same supply chain effects as a worldwide pandemic. I guess those AI doomers that talked about it being the end of the world had it right!
Doomers IMO are just click baiting.
There is a saying that is often trotted out my economists "That the cure for high prices, is high prices".
There is a consumer market and business need for DRAM outside of AI. Someone will fulfil the need as there is a high incentive to. It just going to take a bit of time for this to happen. My equipment is going to be fine for another few years. So I am going to just hang tight and make do with what I got for now.
Main producers actually reduced dram output in 2026. When you have few players with very high capital cost you will end up with cartels like light bulb cartel.
Someone will come in when the price goes up enough. It will take time, but it will happen. What people are complaining about is that the time for this to happen is too long.
Oh look, there is a player coming into the market it seems:
https://economy.ac/news/2026/02/202602288291#:~:text=If%20eq...
EDIT: In fact many other chinese companies are now expanding into DRAM because of the high prices. Which confirms exactly what I said.
> Someone will fulfil the need as there is a high incentive to
And those uses which fall short of the new threshold, e.g. hobbyist SBCs, slowly fall away.
In reality were they going to survive anyway? I would wager likely not.
Raspberry PI is the defacto standard for SBCs. Almost all the other SBCs had significant problems usually around software support and also third party support e.g. Hats, cases etc.
So are AI evangelists to be fair.
It is almost as if two or more things can be true at the same time.
This time is different. https://ca.pcpartpicker.com/trends/price/memory/#ram.ddr5.60...
The price for a couple of 32GB sticks is now over $1200 after being stable at about $200 for several years until last September. That's not a blip; that's 6-fold hike and there is no sign it is slowing down any time soon.
Did you not read what I said? I couldn't even get a replacement video card at any price during the height of COVID and believe you I had the money to pay for one. I couldn't even get a Raspberry PI (any model) for about a year. They were constantly out of stock.
> That's not a blip; that's 6-fold hike and there is no sign it is slowing down any time soon.
How does that invalidate anything I said? As states in the article this will change, it will take years but it isn't forever.
I find it hard to believe that people here cannot make do with whatever hardware they already have.
I also don't believe those small SBCs would have survived long term anyway. Most people just use a Raspberry PI. It is either a MiniPC or a Raspberry PI.
Ya I mean gfx card was pretty bad during Covid.
Discord groups that had real-time line counts and pictures of the line at most best buys across the country (US).
The only way I got one was overpaying and a lottery system that bundled it with other hardware because they knew everyone would still buy it. It was impossible to buy online normally as you needed some kind of automated way to buy it before stock zeroed the minute it was posted.
You could pay a scalper for a gfx card, but stores had none. Now, stores have RAM at least.
> Did you not read what I said? I couldn't even get a replacement video card at any price during the height of COVID and believe you I had the money to pay for one.
You're comparing to memory sticks that went up 6x. If you were offering anywhere near 6x MSRP and you couldn't get a video card... I don't believe you.
https://www.pcmag.com/news/scalpers-have-sold-50000-nvidia-r...
https://www.pcmag.com/news/read-it-and-weep-heres-how-bad-nv...
These show GPUs available for 1.5-2.5x price, which fits what I remember.
> I couldn't even get a Raspberry PI (any model) for about a year.
https://picockpit.com/raspberry-pi/why-are-raspberry-pi-pric...
I didn't look into Pi prices a whole lot, but this suggests they were continuously available for 2-3x price.
I am in the UK. Not the US!
> If you were offering 5x MSRP and you couldn't get a video card... I don't believe you.
My 1080Ti had died. I had to use a 8800GTS from the late 2000s for about a year. As that was the only GPU I had. I have no iGPU on my CPU.
There was at one time, no stock available. Not on Amazon, Not on Overclockers, Not on Scan. They had some weird lotto system taking place on most sites.
Scalpers claimed to have cards. But I wouldn't risk sending a lot of money to some random seller on ebay.
> Unless this article is massively misleading, sure it was out of stock at 1x price but it wasn't out of stock at 2-3x price.
Again I am in the UK. You could not buy any PI other than 1GB model and maybe the zero. Both of which were useless to me.
Okay, UK, maybe that changes things more than I expected. But what about ebay and the sites that replaced classified ads? And is it unreasonable for me to say that you could have bought a US listing and had it reshipped?
Edit since you added: Scalpers claimed to have cards. But I wouldn't risk sending a lot of money to some random seller on ebay.
Even with ebay's buyer protection?
Well not to be mean but I think "I refused to use ebay" invalidates your claim that you couldn't buy a card.
> Even with ebay's buyer protection?
I've had problems with it before. I'd rather not going through the hassle and/or risk in the first place.
> Well not to be mean but I think "I refused to use ebay" invalidates your claim that you couldn't buy a card.
No it doesn't. I would be foolish to trust some overpriced (or underpriced) listing on ebay. I've had an ebay/paypal account now for 25+ years.
> The price increases bring the 16GB Pi 5 up to $299.99.
Meanwhile, a refurbished corporate laptop with 16GB RAM and a 512GB SSD can be yours for $199 [1]
I'm sure there will still be people who want the Pi 5 but at these prices, I ain't one of them.
[1] https://www.ebay.com/itm/327079631563
Those will dry up soon enough. Corporate laptop refreshes will be drawn out as they try and cost save on the increased price.
You also better hope the aliexpress dont figure out a way to get the RAM out of those things because they will start harvesting it for sure if there is money to made.
> Those will dry up soon enough.
We're talking about a pi replacement. The Pi 5 is slower than a 10yo laptop. That's gives us a very vast pool of used laptops.
> You also better hope the aliexpress dont figure out a way to get the RAM
That is a real worry and I can see used machines being gutted because selling DDR3/4/5 sticks is way easier and profitable than the whole machine. Adapters for SODIMM to regular DIMM are readily available and cheap, too.
For most older laptops it's easy enough, you just open them up and take the RAM sticks out. There are SO-DIMM to DIMM adapters to fit a laptop memory stick in a DIMM socket.
The EDU Neo is $500, too bad it’s not as versatile.
it's probably time to call those old retired programmers to ask them how to reduce software memory footprint
or to teach that again
I can save everyone a few Mb of memory now :
1. Check that you really need a SaaS SPA to solve the communication issues between your team members.
2. HTML and css should be enough for 99% of corporate websites.
3. Resize the images on your websites, they're too big.
4. Use teams in the browser, not as stand-alone app.
Seems unnecessary. We can simply ask the LLMs to do it after, of course, imploring them to not make any mistakes.
Taking a big, complex, already well optimised program like Chrome or the linux kernel and optimising the memory footprint is hard. But 90% of programs are just crappy web apps that nobody has even bothered to optimise at all. (Sometimes wrapped in electron or something.)
If you go look, you often discover that 90% of the requests are useless, or at least could be combined. That 60% of bandwidth is used up by 3 high res images which get displayed at 30x30 pixels. That CPU performance is dominated by some rubbish code that populates an array of a million items every call, then looks up 1 element then throws the whole thing away, only to regenerate the exact same list again a few microseconds later.
We have plenty of RAM. In absolute terms, 8gb of ram in the macbook neo is 8 billion bytes. 64 billion ones and zeros. You don't need rocket science to make a CRUD app that runs well with that much ram.
Computers don't get slower over time. If we were merely as lazy with computing resources as programmers 10 years ago, most programs would scream on modern hardware.
This is happening, sort of. All the big tech companies have major initiatives going to reduce RAM usage.
The old graybeards who know how to optimize efficiency may not work for them anymore, though.
The art is not lost, just not funded. Feel free to fund the programmers for your own software projects.
It isnt lost but it also isnt a common skill set in programmers any more.
Most programmers are JS web devs writing client side code or server side CRUD.
I would guess < 10% of programmers writing code today get perf / valgrind out on the regular. I know I dont.
You can still write JS or TypeScript code that tries its best to keep memory use under check. JavaScript was around in the late 90s when the memory footprint of software was at least an order of magnitude lower, so it's absolutely doable.
You don't have to go that deep. 99% of the time our analytics or risk management teams have some really memory inefficient Python and they want me to write them one of our "magic C things" it turns out to be fixable by replacing their in-memory iterations with a generator.
Most people don't have the chance to do that, but hopefully we can see some other languages get first class access on the web. At least there is the whole WASM project.
Why not ask your LLM?
Just rewrite your biggest memory hogs in Rust, it routinely slashes RAM footprint and demand for RAM throughput. The effect is even bigger than the typical reduction in CPU use. You can even ask AI to help you with the task, it will use a lot less RAM for it than the rewrite will save down the road.
Why would we need rust, if the AI can just write really good code in C that doesn't exhibit any of the issues that rust protects you from?
Rust's compile-time checks are actually a nice set of guardrails for LLMs.
Nobody who works with LLM generated code believes that LLMs produce fault-free code.
'rewrite in C, make sure there are no memory leaks'. You first.
Why is that less realistic than saying 'rewrite in rust, make sure there are no memory leaks'?
My point, which I should have been clearer with, is that we aren't at a state where you can just one shot a rewrite of a complex application into another language and expect some sort of free savings. Once we are at that state, and it's good enough to pull it off, why wouldn't the AI be able to pull it off in C as well?
You don't have to trust the AI to do it with Rust, you just have to ensure certain conventions are followed and you can formally prove you're 'safe' from certain classes of issue, no AI magic dice-roll.
A lot of people are very excited by the idea that now language capabilities (and almost every other technical nuance) somehow don't matter but much like gravity they will continue to assert themselves whether you believe in them or not.
So far humans have proven unable to write large apps in C without those issues, given their work is the training basis for LLMs this creates two problems, one being that they don't 'know' what a safe app looks like either and any humans reviewing the outputted code will be unable to validate that either.
There are classes of bug that are easy to write in C that are impossible to express in Rust.
The Rust ecosystem and build tools are much easier to use than C. The value of a language isn't just syntax.
LLMs are great at C, probably because C is historically the most popular language in the world, by far. It only declined slightly very recently. But there's insane amount of code written in it.
It can't, because there is no really good code to train off of.
Because it can't?
Based on a FIDO2 spec I used it to write a reasonably compliant security token implementation that runs on top of Linux USB gadget subsystem (xcept for attestation, because that's completely useless anyway). It also extracted tests from an messy proprietary electron based compliance testsuite that FIDO alliance uses and rewrote them in clean and much more understandable C without a shitton of dependencies that electron mess uses. Without any dependencies but openssl libcrypto, for that matter.
In like 4 hours. (and most of that was me copy pasting things around to feed it reasonable chunks of information, feature by feature)
It also wrote a real-time passive DTLS-SRTP decryptor in C in like 1 hour total based on just the DTLS-SRTP RFC and a sample code of how I write suckless things in C.
I mean people can believe whatever they want. But I believe LLMs can write a reasonably fine C.
I believe that coding LLMs are particularly nice for people who are into C and suckless.
Have you tried asking Claude 4.6 Opus?
Am I allowed to complain about this or do I have to get my VC's approval first
am I crazy for thinking that the 16GB Pi 5 is just there to absorb money from people who purchase the most expensive version of things? Like really nobody needs that much RAM on a Pi?
I am running a bunch of stuff on my 8GB Pi and I've run out of memory to put more stuff on. I use it as a low power server running a bunch of Docker containers. Some of these require at least 200mb and some use 2G of memory.
I was going to buy a small nuc and load it up on memory but I've acquired an old Mac Mini with 16GB of ram, which will do.
Yes, you are crazy for thinking that. The extra ram is useful for small LLMs and also running lots of dock containers. The very low power consumption makes it ideal for a low end home server.
I use the 16GB SKU to host a bunch of containers and some light debugging tools, and the power usage that sips at idle will probably pay for the whole board over my previous home server, within about 5 years.
You can just as well not run docker. 1GiB machine can run a lot of server software, if RAM is not wasted on having duplicate OSes on one machine.
Docker is about containerization/sandboxing, you don't need to duplicate the OS. You can run your app as the init process for the sandbox with nothing else running in the background.
I think that on linux docker is not nearly as resource intensive as on Mac. Not sure of the actual (for example) memory pressures due to things like not sharing shared libs between processes, granted
Containers are not Virtual Machines. 1GB cannot run a lot of server software.
If stuff is written in .NET, Java or JavaScript. Hosting a non-trivial web app can use several hundred megabytes of memory.
Most Pis are sold for embedded customers, some of which no doubt can use 16gb.
No you are not crazy. It's silly to try to use a raspberry pi 5 16GB (or equivalent priced product) as a desktop workstation with a GUI on it when much better actual x86-64 based workstations exist. Ones with real amounts of PCI-E lanes of I/O, NVME SSD interfaces on motherboard, multiple SATA3 interfaces on motherboard, etc. In very small form factors same as you'd see in any $bigcorp office cubicle.
It’s an incredibly lopsided machine. The Pi 5 is decently powerful, but you really really should not be attempting to use one as a desktop replacement. While theoretically possible you are so much better off with a $50 used SFF PC.
The SBC markets been on life support for a long time. Youtubers making videos about them don't seem to grasp that and keep pumping out reviews and projects like its still 2019. The pi specifically has plummeted in popularity and for most use cases they just aren't a cost effective option when second hand micro pcs are dirt cheap and vastly more capable.
PCs don't have GPIO. They're different markets and the desktop replacement never materialized.
For most projects using GPIO, a <$10 ESP32 board or Arduino clone will suffice
MCUs are great but a lot of projects require linux and an application processor. Pi is the industrial standard.
I don't think comparing new Pis to used micro pcs is fair. Compare a _used_ Pi with a used micro pc. If you have any geek friends, it's probably not hard to find a used Pi for free.
Time to break out the Small Web protocols and start living within our means!
Sorry, best I can do is coal powered datacenter vibe coding.
It's not easy to do hobbies.. I keep needing more money..
The extreme DRAM market has had an unexpected side effect of triggering a lot of panic buying. I know several people who delayed PC upgrades for years but then panic bought new systems in this market. The trigger was seeing all of the "It's only going to get worse" and "This is the end of personal computing" headlines.
They're already regretting spending so much now that prices have started to tick downward.
I keep telling everyone: If you don't have a pressing need to buy right now, please wait 6 months and check again.
wasn't "panic" buy but I built a new comp early 2025, cuz at worst case would be complete supply crash and at best case it was going to be more expensive.
Def don't regret doing that, though I regret not springing for the extra RAM.
Same. I got 64gb for my new build the day this whole thing started but I kind of wish I had gotten 128 just for bragging rights.
That's actually a reasonable response to market volatility and illiquidity. It's not just high prices, but prices that still fail to be representative of the actual market stance despite the rises.
It's not a reasonable response. If you don't need a PC right now, buying in the middle of a demand spike is the worst time to do it.
What’s interesting is mini pcs are dirt cheap. The RAM for them costs as much or more than a barebones Ryzen 7 mini pc.
High speed NVME is soaring too. Some popular Samsung kits are up 3X compared to 12 months ago.
Does this mean the Atom 8GB boxes I have laying about are now more valuable?
Yep. I just bought a Pi CM5 for my son, for his ClockworkPi uConsole. CAD $200 for the 8GB module. I bought a whole Pi5 16GB not long ago for under CAD $200.
I will not be buying any more SBC's at this price point. I wonder if Raspberry PI will survive.
Got my RPi 5 16GB quite a while ago for around $160 and already thought that was expensive... It’s still powerful enough for almost everything I throw at it, honestly a bit overkill in most scenarios.
With prices steadily going up, for me it's starting to feel more sensible to repurpose the RAM sticks I've collected from old PC builds / laptops and just throw together small amd64 boxes instead of buying more RPis.
I wonder if there are low power Intel or AMD boards that accept DDR3. So many sticks of 2 / 4 / 8GB DDR3 inside laptops going into recycling or landfills which would do perfectly fine for low power purposes. Hell, performance for standard workloads scales with access times, not bandwidth, and DDR3 sits nicely at CAS8 1600MHz and CAS10 2133MHz..
Does anyone mind explaining why the 2GB model only increased by 20% in price while the 16GB model nearly tripled in price?
The main cost input is presumably ram. They are passing it through.
If everything on the board but the ram costs $30, and ram is going from $10/gb to $20/gb, then they have to change the price $50 -> $70 to break even on the 2gb board, and $190 -> $350 for the 16gb board.
In other words, the raspi is now priced like a stick of ram with a bonus computer attached because ram is massively more expensive than the rest of the computer.
The 16GB model has eight times more ram?
Bought a couple of 32gb SBCs before this all hit the fan. And also built a SSD NAS before the wave hit.
So timed that all pretty great. What worries me is my desktop is up for a full new buy somewhere around early '28. That could be a train wreck depending on how taiwan situation goes
> So timed that all pretty great. What worries me is my desktop is up for a full new buy somewhere around early '28
That's a very specific date / timeline. How do you decide to do a full new buy? I ask because I own a desktop that I built 15 years ago which I was flirting with replacing completely last year, but unfortunately I didn't pull the trigger ... oops :(
My old rig is still going strong. The motherboard can only take up to 32GB DDR3 though. CPU is an Intel i7-4790k which is still very fair today if you are not running a resource hog OS (looking at you Windows). Overall it is completely serviceable for my needs. Being honest with myself the only reason I wanted to upgrade was for nerd cred but I don't game much anymore and don't do any ML tasks that require lots of local compute.
Apple and Sama didn't do the consumers any favors this year.
What did Apple do?
what are the barriers to new DRAM supply coming online?
Huge capital outlays and no guarantee the prices stay high.
Startup costs measured in the billions, with no guarantee of success, and a long payback time horizon in a market that almost everyone thinks is - in one way or another - a bubble.
Oh yeah, the market is also getting intense scrutiny from powerful geopolitical entities that are quite explicit that they don't believe in fair play or consistent, stable rules.
Would you place that bet?
Totally agree. Just like graphics card prices. Is it worth building a pc now?
After discovering Dell Alienware clearance and graphics card availability in those Alienware computers, I haven't felt the need to build a computer for the last five years.
I looked on their site. I don’t see any section for clearance.
We're somehow in a race between LLMs curing cancer, destroying the planet by "You're right to be mad, I shouldn't have issued those launch codes, it's even in my Claude.md file, I'm sorry," and rendering modern technological civilization uneconomical. I know this is statistically the best time in history to live, but lord, I could use a vacation.
It's terrible. Fake money is fueling the exhaustion of real resources in search of questionable outcomes ("AGI"). Imagine if all of these money were invested in curing cancer.
Imagine if AI cures cancer.
Let's also imagine an alternative reality where some reasonable percentage of the $2.5T in current year AI spending was instead invested in the "general intelligence" researchers we already have for the same purpose. I think it's a pretty reasonable expectation that 1) they'd probably make more progress and 2) that money would help a lot more people in the process (through jobs and economic activity).
You can imagine all you want, but my understanding is there is no credible evidence that scaling LLMs will result in true AGI.
Obviously there's no "evidence". Why would you even think we need AGI? But I'm happy to hear your reasoning if you were one of the few/only? people who imagined that software that could predict the next word could do what it now is doing.
An already-ageing population living even longer while nobody wants kids anymore?
You don't have to imagine, it will hallucinate you a slop with full confidence every time.
I've already seen at least one person who was pretty sure that the preprint paper they co-authored with AI (read: AI wrote for them) was going to cure cancer and make them billions of dollars.
There was only one problem. The paper jumped straight from "this paper will show how our new treatment could cures cancer forever" to "as you can see, these results clearly show that our treatment cures cancer" - with neither any actual results nor any specifics on the treatment. And I don't just mean that the paper didn't go into details; writing the paper was the full extent of their "research".
So QED then, I guess.
AI was used fundamentally for COVID vaccine development. AI is used for research in all modern drugs. It’s a certainty if cancer gets cured AI will have played a fundamental role since it’s already fundamental to precursors.
My PDP-11 runs fine on 512K
I bet the power draw is at least 50x though
Unless you're really using the GPIO pins or other weird I/O, I really fail to see the purpose in having an 8GB or 16GB RAM Raspberry Pi (at a much higher price than it used to be) as a desktop workstation with a GUI on it.
The idea of putting sixteen gigs of RAM in a raspberry pi is nuts. The legit thing you want to use a raspberry pi (or a competitor) for as an embedded headless thing with no KB/mouse/display attached should run fine in 2GB of RAM or less, assuming an ordinary debian-based OS environment.
I would much rather have a used, ex-corporate/ex-lease, small form factor or ultra small form factor x86-64 desktop PC (Dell, HP, Lenovo, whatever) with 16GB of RAM in it and an SSD on a SATA3 or NVME interface. Whatever is the "best" SFF that you can buy via huge eBay used equipment dealers on any given month.
Despite being many years old, whatever you can buy on ebay for 200 bucks (at least before the recent RAM fiasco) with some recent-ish quad core core i5/i7 or Ryzen in it will run circles around a raspberry pi 5.
SBCs are not just RPis. Other brands can still be bought cheaper.
Is there anything (technically) preventing SBC manufacturers adding SODIMM slots?
I was expecting the Milk V Titan to avoid this memory nonsense since it has two unpopulated DDR4 slots, but it has fallen off the radar like several other SBCs.
SODIMMs are huge compared to a BGA memory package which is a problem if your goal is to minimize your board size (e.g. I don't think there's a reasonable way to fit it into a Raspberry Pi form factor without something weird and expensive like a mezzanine connector). Routing the signals is also somewhat more annoying because they all come out of one edge of the connector compared to a BGA package which has them fan out in every direction, giving more space for length matching traces, etc. You'll likely need additional PCB layers compared to a BGA chip.
DDR4 is also crazy expensive right now so this just depends on you having some around from a previous build
It actually seems to be slightly less expensive than DDR5, perhaps due to the lower throughput that makes it uncompetitive for AI-adjacent workloads.
The title should say: "Collusion of large corporations promoting LLMs with RAM manufacturers is killing the hobbyist SBC market (and bankrupting anybody trying to get a PC or laptop)".
Because we all know that DRAM prices have spiked since production is going to those infernal chatbot training data centers. Same as a lot of the electricity in some parts of the world, BTW.
Can you elaborate on the collusion aspect? Is the implication that OpenAI and Anthropic are coordinating their purchases in such a way that they target the hobbyist market? What’s the collusion angle here?
OpenAI signed letters of intent for 40% of the DRAM supply because they have no moat and want to starve their competition.
Only works so long as you eventually pay up... well unless the manufacturers make too much this way. That said are there some Chinese manufacturers that aren't part of the cabal and could undercut them?
Except that it doesn't work like that. If you buy DRAM and don't do anything genuinely worthwhile with it, you'll ultimately dump it all right back onto the market, and everyone knows that. The biggest worry is that it's actually OpenAI and their direct competition starving the rest of the market because they predict AI research and the like to be a highly valued use for the stuff, compared to building gaming PC battlestations or whatever the highest-valued use was before. Many observers think that this will also happen with GPUs and cutting-edge digital logic more generally.
> Collusion of large corporations promoting LLMs
> We all know that DRAM prices have spiked since production is going to those infernal chatbot training data centers
I know it's very fashionable here to talk about capitalism as some hand-washes-hand big corp organized scam, but if you put that ideology aside for a moment, you contradicted yourself here, I think.
I personally don't like conspiracy-theory-thinking. If I was a DRAM manufacturer and had to choose between servicing a single customer, who orders hundreds of millions worth of my product, or service a very large number of customers who order tiny amounts of the product a piece, then of course I would focus on the large client, because they are easier to service for the expected profit margin. I wouldn't even need to think about advertisement, sales, all that jazz. Looking at it from that perspective, it seems pretty logical to me that a spike in demand from datacenter operators would rise prices dramatically. I struggle to see room for collusion / conspiracy here.
A couple of issues, first there is a history of price collusion (see DRAM price fixing scandal on Wikipedia) and while it may be "logical" from a seller point of view to prefer large orders, this upsets a lot of people and used to be illegal in the United States (it may still be illegal, but it's not enforced)
Oh, I did not know that. Thanks for the clarification
Holy crap, it is super expensive now. I should have brought an extra one in the past.
It’s great that everything I love is getting ruined so that the most mediocre people on earth can generate slop on a daily basis.
“Killing” is strong phrasing.
Yes, a $250 mini PC I bought last year is now $350.
Is this pricing bad? Yeah, compared to what it was.
Is this the end of the world? Not really, and we’ve seen price spikes for all kinds of PC components in the past. It’s rarely permanent.
That sounds pretty nice. The same mini PC I paid $195 for in 2023 is now $450. Seems to be life in Canada sometimes.
It had caused me to look around though. I have found the Pi Zero 2W to be surprisingly capable for Pi sized jobs.
Not everyone earns tech bro salaries and can sustain a thousand cuts. Many hobbiests are scraping and saving money to acquire hardware. For some it very well msy be the end of their world.
We are talking about brand new latest gen hardware here. People with low budgets are always scraping and saving for deals and don’t need to buy something brand new from a pricey brand name like raspberry pi.
You can still jump on eBay and buy all kinds of dirt cheap used pieces of hardware.
My buddy just bought a used ThinkPad T14 with 32GB of RAM and 1TB of storage for about $500. You can get by with a whole lot less.
In this context, I will also present the idea that Rasperry Pi has represented quite poor cost value for many years now.
Have you looked at how expensive international shipping is? eBay covers just a few countries, the rest of us can't buy there because we'll be paying 10 times the cost of hardware to get it over here.
I already moaned about this recently, but to briefly reiterate: the only hardware that's becoming available for most people in my region are Frankenstein desktops built from heavily used 10+ year old Xeons running on suspicious motherboards made by obscure Chinese manufacturers you've never heard of. This is pushing ever more people towards smartphones and away from actual computers.
But at least we got the bullshit machine in return, that's something, I guess.
> Have you looked at how expensive international shipping is?
It really shocks me how bad shipping has gotten. It's nearly unaffordable to buy things on eBay from the US as a Canadian due to shipping costs, so I can only imagine just how bad it is for people from other countries.
That cheap stuff from eBay that people talk about all the time seems to be available only in North America, or in the best case Western Europe.
ThinkPad T14 which generation?
Yes, 90%+ of sellers refuse to ship here (and we're not even under any sanctions and/or political pressure of any sort). I hear about these magical 100$ Thinkpads all the time; I'm yet to see anything cheaper than 300$ (add another 100$+ for shipping).