There’s LPCAMM2 but they are already expensive even before the recent shenanigans. Maybe there other ways such as cpus for quad-channel RAM
I’m not saying we should pay more to get less RAM. I think we should stop buying new RAM sticks and stick to whatever we currently have, including but not limited to RAM sticks
That only works if you already have some.
I would personally prefer ai companies not hoarding all of it using some imaginary money and the definitely not price fixing ram triopoly doing absolutely nothing to meet demand or compete with each other.
If there is anything good coming from this then it’s a slight pressure on software manufacturers to do some optimization but I’ll believe it when I see it.
I already seen it XD a while ago I bought a budget laptop (N150, 4000MT/s RAM, 512GB SSD). It’s super laggy on stock configuration(pre installed Windows 11), however it runs faster than most gaming laptops(many players run Windows for some popular games that have kernel level anticheat, which is essentially malware and fundamentally incompatible with Linux) after I installed Linux on it.
Expect the prices of DDR (SO)DIMMs, SSDs and video cards (demand + DDR VRAM) to go up and up until (at least) Q3 2026 according to what I’ve heard and read until now.
So if you’re on the market for a new Framework laptop, upgrades, or EGPU setup amongst others the time to invest is now.
I’m currently waiting until my Framework 16 is shipped but the acquisition of the two SSDs and memory I wanted was not that much fun to say the least.
When I was first putting my laptop together the 2x64Gb memory modules I wanted for my new F16 were ~~$500 (which already was ridiculous given earlier prices). When I finally finalized my order and purchased the modules they were ~$750. Now they are nearing ~$900 on the market (at least at Amazon, the other good store in my country has them in stock for -i shit you not- $1500).
Consider this a friendly warning.
The AI bubble is similar to the nuclear arms race. Many countries currently see getting their fair share of now-limited compute as an existential issue, so barring some legendary paper proving without a shadow of a doubt that transformer-derived LLMs can’t solve a general class of problem even with exquisite engineering, you can expect every country with deep pockets to finance this bubble until an entire economy burns up and collapses from not being able to support it
This implies that the bubble will go on long enough for semiconductor fabs to renegotiate their locked-in prices and in some years you’ll see even the prices of consumer grade chips ballooning as well
Sadly you might/will be right. But still happy that they have stayed strong so far. But it would be understandable.
I was paranoid enough to buy my ram and SSDs back when I first ordered my FW16, but if I had known how bad it would get I would’ve ordered a bigger SSD.
It may be a different problem but we as a society have found ways around said technical problems many times in the past. I remember in the early 2000’s hearing that we’re running into the limit of physics for how small we can make transistors, about how we’re running into thermal limits, and the end of Moore’s Law. At the time we were making processors in the realm of 180-90nm, nearly 20x larger than we are today.
We were also hitting limits of how large hard drives could be made, 80GB or so in a 3.5 inch form factor, when IBM came trotting out with their “pixie dust”. For flash member, which was still relatively new, there were concerns over reliability, scalability, etc., that came down to the same physics limits that were a concern with processors.
However, we’ve surpassed all that, and I’m certain we’ll find a way past this.
There were several reasons for switching from DIP to something else. Signal integrity would have been one, sure, however I think before we really got to that point DIPs became untenable due to the amount of I/O needed. That, and simply the need to shrink components down to make things smaller would have been a driver in and of itself.
Yes which is where the various camm implementations come into play.
Which isn’t entirely wrong, most of the improvements have been from making transistors differently and not just smaller since then. They have gotten smaller but they also got folded and rearanged.
It has certainly slowed down signifficantly.
I am not arguing that socketable ram is impossible, I am just doubting the survival of the sodimm form factor. On desktop/server we can justify just using violence to make the longer traces work at the cost of power but on a mobile platform that is a lot less justifiable. To find a way past it there would have to be a will and a reward for doing so. As a manufacturer why would you make the effort to make a product that has worse battery life and does not allow you to charge whatever the hell you want for bigger memory capacity?
I mean its already reached its end. DDR5-5600 is the end of the road.
Not a bad opportunity to jump ship altogether and go CAMM at some point … once new platform drops
Omg I didn’t see the pricing till now. Ram pricing right now is insane.
Yes… I have several Crucial 128GB ram kits that I bought for $364 when they were released in February, and according to camelcamelcamel has been as low as $250. Now the same kit is $750-$1,100 depending on where you find them.
did framework mention pricing/availability for framework desktops?
was looking forward to get the 128gb model for a local llm but seeing the rampocalypse its not something i need THAT bad…
Yikes I hope this is a short spike.
They said they are not changing the price of their desktop ram!
I wouldn’t hold your breath… The only thing we can really hope for is for TSMC and other chip fabs increasing capacity to get supply in check. IIRC that took quite a few years for GPUs and then AI just blew that out of the water too, at least for NVIDIA.
I would be interested to find out how micron made their decision to exit the consumer market. Was it based on revenue or profit?
With the laptop ram being over 100% profit now, due to the current peek in prices, that is quite something to walk away from.
Maybe the server ram has an even higher profit/margin?
That being said, people are talking of an AI crash, so investing in more fabs is a difficult one to take.
I guess the next big thing would be the cost of fab equipment dropping to the cost of a 3d printer, thus everyone could own their own fab machine if they wished.
Maybe? 
“A single rack of NVIDIA’s GB300 solution uses 20TB of HBM3E and 17TB of LPDDR5X. That’s enough LPDDR5x for a thousand laptops, and an AI-focused datacenter is loaded with thousands of these racks“
They make absolute epic shit tons of revenue and profit on the AI crazy train (the golden toilets kind), until the whole thing ofcourse collapses and crashes even more spectacularly than the dot com crazy train in the late 90s/early 00s.
By the way, SK Hynix has recently announced they are “mostly” leaving the consumer market behind aswell. Same reason.
Samsung are having problems with selling their RAM chips. They actually have over-supply.
The problem with there RAM chips is quoted as “(Higher power consumption)”.
This is suppressing the demand for Samsung chips, in favor of Hynix and Crucial/Micron.
That actually make the desktop right now a very good deal. I hope they are still not losing money on it …