128GB used to be 400$ in June, and now it's over $1,000 for the same 2x64GB set..
I have no idea if/when prices will come back down but it sucks.
(That's just me eyeballing it, feel free to do the math)
Nobody is going to do anything if they can't be sure that they'll be able to run the fab they built for a long time and sell most of what they make. Conversely fabs don't tend to idle a lot. Sometimes they're only built if their capacity is essentially sold already. Given how massive the AI bubble is looking right now, I personally wouldn't expect anyone to make a gamble building a new fab.
* Someone explained this at length on here a while ago, but I can't seem to find their comment. Should've favorited it.
"Hyperscalers" already have multi-year contracts going. If the demand really was there, they could make it happen. Now it seems more like they're taking capacity from what would've been sold on the spot or quarterly markets. They already made their money.
Following your reasoning, which is common in manufacturing, the capex needed is already allocated. So, where does the 2x price hike come from if not supply/demand?
The cost to produce did not go up 100%, or even 20%
Actually, DRAM fabs do get scaled down, very similar to the Middle East scaling down oil production.
It absolutely is supply/demand. Well, mostly demand, since supply is essentially fixed over shorter time spans. My point is that "cost per square mm [of wafer]" is too much of a simplification, given that it depends mostly on the specific production line and also ignores a lot of the stuff going on down the line. You can use to look at one fab making one specific product in isolation, but it's completely useless to compare between them or when looking at the entire industry.
It's a bit like saying the cost of cars is per gram of metal used. Sure, you can come up with some number, but what is it really useful for?
https://www.tomshardware.com/pc-components/storage/perfect-s...
...except current peak in demand is mostly driven by build-out of AI capacity.
Both inference and training workloads are often bottlenecked on RAM speed, and trying to shoehorn older/slower memory tech there would require non-trivial amount of R&D to go into widening memory bus on CPU/GPU/NPUs, which is unlikely to happen - those are in very high demand already.
Even China has no reason to reduce prices much for memory sold to the US when they know we have no choice but to buy at the prices already set by the cartel. I expect that if China does start making memory they'll sell it cheap within China and export it at much higher prices. Maybe we'll get a black market for cheap DRAM smuggled out of China though.
So, it would happen even with best intentions and no conspiracies. AI boom already hiked GPU prices, memory was next in line.
If there is high demand for the former due to AI, they can increase production to generate higher profits. This cuts the production capacity of consumer DRAM, and lead to higher prices in that segment too. Simple supply & demand at work.
Each memory DIMM/stick is made up of multiple DRAM chip. ECC DIMMs have an extra chip for storing the error correcting parity data.
The bottleneck is with the chips and not the DIMMs. Chip fabs are expensive and time consuming, while making PCBs and placing components down onto them is much easier to get into.
A consumer DDR5 module has two 32-bit-wide buses, which are both for example implemented using 4 chips which each handle 8 bits operating in parallel - just like RAID 0.
An enterprise DDR5 module has a 40-bit-wide bus implemented using 5 chips. The memory controller uses those 8 additional bits to store the parity calculated over the 32 regular bits - so just like RAID 4 (or RAID 5, I haven't dug into the details too deeply). The whole magic happens inside the controller, the DRAM chip itself isn't even aware of it.
Given the way the industry works (some companies do DRAM chip production, it is sold as a commodity, and others buy a bunch of chips to turn them into RAM modules) the factory producing the chips does not even know if the chips they have just produced will be turned into ECC or non-ECC. The prices rise and fall as one because it is functionally a single market.
I don't really understand why every little thing needs to be spelled out. It doesn't matter. We're not getting the RAM at an affordable price anymore.
Do we really think the current level of AI-driven data center demand will continue indefinitely? The world only needs so many pictures of bears wearing suits.
I feel like the question of how many computers are needed to steer a bunch of self-driving taxis probably has an answer, and I bet it's not anything even remotely close to what would justify a decade's worth of maximum investment in silicon for AI data centers, which is what we were talking about.
In a traditional pork cycle there's a relatively large number of players and a relatively low investment cost. The DRAM market in the 1970s and 1980s operated quite similarly: you could build a fab for a few million dollars, and it could be done by a fab which also churned out regular logic - it's how Intel got started! There were dozens of DRAM-producing companies in the US alone.
But these days the market looks completely different. The market is roughly equally divided up between SK Hynix, Micron, and Samsung. Building a fab costs billions and can easily a year of 5 - if not a decade - from start to finish. Responding to current market conditions is basically impossible, you have to plan for the market you expect years from now.
Ignoring the current AI bubble, DRAM demand has become relatively stable - and so has the price. Unless there's a good reason to believe the current buying craze will last over a decade, why would the DRAM manufacturers risk significantly changing their plans and potentially creating an oversupply in the future? It's not like the high prices are hurting them...
Will the company be evicted from the country in 6 months? A year? Will there be 100% tariffs on competitions imports? Or 0%? Will there be an anti-labor gov’t in effect when the investment might mature, or a pro-labor?
The bigger the investment, the longer the investment timeframe, and the more sane the returns - the harder it is to make the investment happen.
High risk requires a correspondingly high potential return.
That everyone has to pay more for current production is a side effect of the uncertainty, because no one knows what the odds are of even future production actually happening, let along the next fancy wiz-bang technology.
But people do need the current production.
Also worth mentioning DRAM and NAND's profit from Samsung is what keep the Samsung Foundry fighting TSMC. Especially for those who thinks TSMC is somehow a monopoly.
Another things to point out which is not mentioned yet, China is working on both DRAM and NAND. Both LPDDR5 and Stacked NAND are already in production and waiting for yield and scale. Higher Price will finally be perfect timing for them to join the commodity DRAM and NAND race. Good for consumer I suppose, not so good for a lot of other things which I wont go into.
Is this the ‘but he loves me, he wouldn’t hit me again’ of the tech world?
The firms can coordinate by agreeing on a strategy they deem necessary for the future of the industry, and that strategy requires significant capital expenditures, and the industry does not get (or does not want) outside investment to fund it, and if any of the firms defects and keeps prices low the others cannot execute on the strategy, so they all agree to raise prices.
Then, after the strategy succeeds, they have gotten addicted to the higher revenues, they do not allow prices to fall as fast as they should, their coordination becomes blatantly illegal, and they have to get smacked down by regulators.
As long as it doesn't fall into the "collusion" prohibitions of the relevant competition law.
> “People of the same trade seldom meet … but the conversation ends in a conspiracy against the public, or in some contrivance to raise prices.”
Adam Smith, The Wealth of Nations (1776)
So said and did the infamous Phoebus cartel, to unnaturally "fix" the prices and quality of light bulbs.
https://spectrum.ieee.org/the-great-lightbulb-conspiracy
https://en.wikipedia.org/wiki/Phoebus_cartel
For more than a century, one strange mystery has puzzled the world: why do old light bulbs last for decades while modern bulbs barely survive a couple of years?
The answer lies in a secret meeting held in Geneva, Switzerland in 1924, where the world’s biggest light bulb companies formed the notorious Phoebus Cartel.
Their mission was simple but shocking: control the global market, set fixed prices, and most importantly… reduce bulb lifespan.
Before this cartel, bulbs could easily run for 2500+ hours. But after the Phoebus Cartel pact and actions, all companies were forced to limit lifespan to just 1000 hours. More failure meant more purchases. More purchases meant more profit. Any company who refused faced heavy financial penalties.
The most unbelievable proof is the world-famous Livermore Fire Station bulb in California, glowing since 1901. More than 120 years old. Still alive. While our new incandescent bulbs die in 1–2 years.
Though the Phoebus cartel was dissolved in the 1930s due to government pressure, its impact still shadows modern manufacturing. Planned obsolescence didn’t just begin here… but Phoebus made it industrial.
I still try to fight that habit of not unnecessarily cycling even tho all my lights are LED.
Unfortunately commodity business is not sexy, it doesn't get the press, nor does it get told even in business schools. But a lot of the times these call called price fixing is a natural phenomenon.
I wont even go into what get decided in court doesn't always mean it is right.
I will also add we absolutely want the DRAM and NAND or in fact any industries to make profits, or as much profits as it could. What is far more important is where do they spend not those profits. I didn't look into SK Hynix but both Samsung and Micron spends significant amount of R&D at least try to lower the total production cost of DRAM per GB. We want them to make healthy margin selling DRAM at $1/GB, not losing money and then go bankrupt.
It's kinda sad when you grow up in a period of rapid hardware development and now see 10 years going by with RAM $/GB prices staying roughly the same.
A house is $500,000
A GPU is $500
You could put GPUs into the inflation bucket and it wouldn’t change anything. Inflation trackers count cost of living and things you pay monthly, not one time luxury expenses every 4 years that geeks buy for entertainment.
I put my G4 PowerBook into it once, and then vowed never to look at it again.
That period of time had some benefits. Programmers learned to squeeze absolutely everything out of that hardware.
Perhaps writing software for today's hardware is again becoming the norm rather than being horribly inefficient and simply waiting for CPU/GPU power to double in 18 months.
I was lucky. I built my am5 7950x Ryzen pc with 2x48gb ddr5 2 years ago. I just bought 4x48gb kit a month ago with an idea to build another home server with the old 2*48gb kit.
Today my old g.skill 2x48gb kit costs Double what I paid for the 4x48gb.
Furthermore I bought two used rtx3090 (for AI) back then. A week ago I bought a third one for the same price... ,(for vram in my server).
But you’re cherry picking prices from a notable period of high prices (right now).
If you had run this comparison a few months ago or if you looked at averages, the same RAM would be much cheaper now.
We’re just consuming a lot of DRAM in general.
Same order, same bill of materials, 17.5K USD per unit today.
That is roughly a 5.5k increase for 768GB of DDR5 ECC memory and the 4 2tb nvme ssds.
Really wish that I could replace my old skylake-x system but even ddr4 rdimms for an older xeon are crazy now let alone ddr5. Unfortunately I need slots for 3xTitan V's for the 7.450 TFLOPS each of FP64. Even the 5090 only does 1.637 TFLOPS for FP64, so just hopping that old system keeps running.
Upgraded by adding 64GB.. last Friday I sold the 32 GB I took out for what I paid for the 64 GB in July... insane
(Including the submitter. In their comment history is "Tip: You can sell used server RAM or desktop modules through BuySellRam to recover value from old hardware." at https://news.ycombinator.com/item?id=45800881 and all of the submissions of this domain are from this user: https://news.ycombinator.com/from?site=buysellram.com )
Old RAM that comes out of the PCs being sold at fire sale prices isn’t really in demand though. Even slower DDR4 grades aren’t seeing much demand.
Actually, the textile market is pretty volatile in the US these days with Joan's out of business. Pick a poison, I guess? There's little room for stability in a privately-owned-world.
Close a few Chrome tabs, and save some DDR5 for the rest of us. :-)
You're welcome.
If you are working on an application that has several services (database, local stack, etc.) as docker containers, those can take up more memory. Especially if you have large databases or many JVM services, and are running other things like an IDE with debugging, profiling, and other things.
Likewise, if you are using many local AI models at the same time, or some larger models, then that can eat into the memory.
I've not done any 3D work or video editing, but those are likely to use a lot of memory.
RAM uses power.
However this does not make sense, as for more than a decade the processors have only grown increasing the number of threads, therefore two channels sounds like a negligent and deliberately imposed bottleneck to access the memory if one use all those threads (Lets say 3D render, Video postproduction, Games, and so on).
And if one want four channels to surpass such imposed bottleneck, the mainboards that nowadays have four channels don't contemplate consumer use, therefore they have one or two USB connectors with three or four LAN connectors at prohibitive prices.
We are talking about consumer quad-channel DDR4 machines ten years old, wildly spread, keeps being competent compared with current consumers ones, if not better. It is like if all were frozen along this years (and what remains to be seen with such pattern).
Now it is rumoured that AMD may opt for four channels for its consumer lines due to the increased number of pin connectors (good news if true).
It is a bad joke what the industry is doing to customers.
You need to re-check your sources. When AMD started doing integrated memory controllers in 2003, they had Socket 754 (single channel / 64-bit wide) for low-end consumer CPUs and Socket 940 (dual channel / 128-bit wide) for server and enthusiast destkop CPUs, but less than a year later they introduced Socket 939 (128-bit) and since then their mainstream desktop CPU sockets have all had a 128-bit wide memory interface. When Intel later also moved their memory controller from the motherboard to the CPU, they also used a 128-bit wide memory bus (starting with LGA 1156 in 2008).
There's never been a desktop CPU socket with a memory bus wider than 128 bits that wasn't a high-end/workstation/server counterpart to a mainstream consumer platform that used only a 128-bit wide memory bus. As far as I can tell, the CPU sockets supporting integrated graphics have all used a 128-bit wide memory bus. Pretty much all of the growth of desktop CPU core counts from dual core up to today's 16+ core parts has been working with the same bus width, and increased DRAM bandwidth to feed those extra cores has been entirely from running at higher speeds over the same number of wires.
What has regressed is that the enthusiast-oriented high-end desktop CPUs derived from server/workstation parts are much more expensive and less frequently updated than they used to be. Intel hasn't done a consumer-branded variant of their workstation CPUs in several generations; they've only been selling those parts under the Xeon branding. AMD's Threadripper line got split into Threadripper and Threadripper PRO, but the non-PRO parts have a higher starting price than early Threadripper generations, and the Zen 3 generation didn't get non-PRO Threadrippers.
> Even gaming is now more reliant on GPU performance (which in principle ought to benefit from the high PCIe bandwidth of server parts)
A gaming GPU doesn't need all of the bandwidth available from a single PCIe x16 slot. Mid-range GPUs and lower don't even have x16 connectivity, because it's not worth the die space to put down more than 8 lanes of PHYs for that level of performance. The extra PCIe connectivity on server platforms could only matter for workloads that can effectively use several GPUs. Gaming isn't that kind of workload; attempts to use two GPUs for gaming proved futile and unsustainable.
That number of threads will hit a bottleneck accessing only through to channels of memory.
I don't understand why you brought up the topic of single-threading in your response to the user, given that processors reached a frequency limit of 4 GHz, and 5 GHz with overclocking, a decade ago. This is why they increased the number of threads, but if they reduce the number of memory channels for consumer/desktop...
The OP is talking about a specific niche of boosting single thread performance. It’s common with gaming pcs since most games are single thread bottlenecked. 5% difference may seem small, but people are spending hundreds or thousands for less gains… so buying the fastest ram can make sense there.
It depends. It takes more energy, which can be undesirable in battery powered devices like laptops and phones. Higher end memory can also generate more heat, which can be an issue.
But otherwise more RAM is usually better. Many OS's will dynamically use otherwise unused RAM space to cache filesystem reads, making subsequent reads faster and many databases will prefetch into memory if it is available, too.
$ ~/dev/mozlz4-tool/target/release/mozlz4-tool \
"$(find ~/Library/Application\ Support/Firefox/Profiles/ -name recovery.jsonlz4 | head -1)" | \
jq -r '[.windows[].tabs | length] | add'
5524
Real memory size: 2.43 GB
Virtual memory size: 408.30 GB
Shared memory size: 746.5 MB
Private memory size: 377.3 MB
Usually after the companies are fined for price-fixing
When the AI bubble bursts we can get back to the old price
If those retailers didn’t increase their prices when the price hike was announced, anyone building servers would have instantly purchased all of the inventory anyway at the lower prices, so there wouldn’t actually have been weeks of low retail RAM prices for everyone.
Every once in a while you can catch a retailer whose pricing person missed the memo and forgot to update the retail price when the announcement came out. They go out of stock very rapidly.
But that retailer would have made a lot of money in a very short time.
No matter what, you are not getting those modules at the old price. There are few things that trip up people harder than this exact scenario, and it happens everywhere. Concert tickets, limited releases, water during crises, hot Christmas gift, pandemic GPUs, etc.
Once understood you can stop getting mad over it like it's some conspiracy. It's fundamental and natural market behavior.
Years, or when the AI bubble pops, whatever comes first.
Similar situation with QLC flash and HDDs btw.
Those price increases are not normal at all. I understand that most of it still comes from market demands but this is also skewing the market now in unfair manners. Such increases smell of criminal activity too.
You want to penalize companies for buying things and penalize companies for selling things are market rate?
There are a lot of good examples through history about how central planning economics and strict price controls do not lead to good outcomes. The end result wouldn’t be plentiful cheap RAM for you. The end result would be no RAM for you at all because the manufacturers choose to sell to other countries who understand basic economics.
Building a PC in a cost efficient manner generally requires someone to track parts prices over years, buy parts at different times, and buy at least a generation behind.
The same applies to many other markets/commodities/etc...