• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

DRAM Pricing Jumps 50%, Only 70% of Orders Getting Filled

erek said:

“Micron reportedly offers pennies on the dollar for Crucial RAM return, only offers to reimburse original MSRP despite it being only 37% of market value — chipmaker later reverses course with a better solution​

​

Crucial RMA scandal following the brand's exit from the consumer market

Micron used to produce some of the best RAM available to mainstream users through its Crucial brand. Despite the company's exit from the consumer market through its now-defunct Crucialbrand, Micron had pledged to continue honoring warranties on existing Crucial products. A recent case on Reddit, however, highlights ongoing challenges with Micron's warranty process, where it reportedly offered a customer a ludicrouspayout for an RMA. Only after further communication did Micron provide an improved offer. Another user also chimed in with a similar story about an SSD return with Micron. We have reached out to Micron for comment.“

https://www.tomshardware.com/pc-com...-later-reverses-course-with-a-better-solution
Click to expand...
AceGoober said:
That is some serious garbage deals. Micron, and any other RAM manufacturer, at the minimum should offer current market price refund so the user can purchase replacement RAM.
Click to expand...

View: https://youtu.be/OTJvDOJgUA8?is=pW-E1yvLITfl_KlM
 

“Samsung reaches 80% HBM4 yield, HBM4E crosses 70%; raising pressure on SK Hynix​

Samsung Electronics’ HBM4 yield has climbed into the 80 percent range, a milestone reached roughly six months after mass production kicked off, and it’s shaking up a memory market SK Hynix has controlled for years.
What’s driving this jump is Samsung’s integrated approach.
Instead of memory, Foundry, and packaging teams working separately, they collaborated from the design phase itself, and that synergy appears to be paying dividends faster than expected.
SEDaily suggests Samsung also cracked one of its toughest technical hurdles, the Thermal Compression Non-Conductive Film process, managing to balance speed, power efficiency, and heat control all at once.
To be fair, SK Hynix hasn’t been sitting idle. Industry estimates suggest its own HBM4 yield has also touched the 80 percent mark, and the company continues to lean on its mass-production track record.“

https://www.sammyfans.com/2026/08/09/samsung-reaches-80-percent-hbm4-yield/
 

“NVIDIA’s Vera Rubin Memory Bill Doubles Grace Blackwell’s, As DRAM Cost Swells 2.5x In A Single Generation, With One Rack Holding 4,500 Smartphones’ Worth Of Memory​


NVIDIA's Vera Rubin racks feature an unprecedented amount of memory, and that comes with a hefty price tag, as disclosed in the "Bill of Materials".​



Memory Accounted For 53% in Grace Blackwell Servers, but Vera Rubin Pushes That To 62% With Double The Price Bump​

Vera Rubin racks are crazy expensive, and while one would expect to pay a premium for these next-generation enterprise powerhouses, the main cost driver is, unsurprisingly, the memory & that accounts for 62% of the cost of the Vera Rubin platform alone.”

https://wccftech.com/nvidia-vera-ru...es-grace-blackwells-as-dram-cost-swells-2-5x/
 
zehoo said:
Nanya/Winbond/PSMC

This wouldn’t be a short term solution, probably taking up to 10 years. But Apple has played the long game multiple times before in replacing suppliers that screw them.
Click to expand...

Are those fabs capable of producing cutting edge RAM that Apple requires, in the quantities that Apple requires? Will they sell RAM to Apple for the prices Apple wants when other customers are willing to pay orders of magnitude more? Apple's not getting "screwed" by their suppliers, their suppliers simply know that whatever isn't bought by Apple will be bought be someone else, so Apple's at their mercy. Apple screwed their suppliers (and customers) for years by pounding them down to the lowest prices and then turning around and selling basic RAM upgrades for exorbitant prices to their customers.

The current situation Apple's in is the same situation gamers are in when they say Nvidia's "going to get what's coming to them" once some other competitor comes out with big time supply and undercuts them on price. Nvidia knows the gamers aren't serious and they'll always come back to buy an Nvidia card once they're available. The gamers know they're not serious and will always come back to buy an Nvidia card once they're available. Nvidia knows the gamers know they're not serious because they'll always come back to buy an Nvidia card once they're available. Same deal here. Micron, Samsung, and SK Hynix know Apple's coming back, and if they don't, no worries, Sam Altman's going to buy it anyway. Or they can go to Nanya and buy DDR4 and lower the specs on their premium products, their choice.

It's a crappy situation all around, really.
 
kram182 said:
Everyone's (including Microsoft) talking about 8GB meanwhile Microsoft's own system requirements, that they made and wrote and haven't changed, list 4GB. They should be optimizing to that, as that would then optimize for 8GB as well, secondarily (and my poor, cheap, but was only $100 new 5 years ago, 4GB calibration laptop 😭)
Click to expand...

Why optimize for lower RAM usage when they can instead bloat your OS with apps that track you and sell you advertisements in the OS you're already a paying customer for?
 
Optimizing for 4 and optimizing for 8 aren't going to produce the same results. 4 divides evenly into 8, so there will be some benefit, but not as much as for devices with 4GB of ram.

That said, I think optimizing for 4 is a decent middle-ground.
 
Nobu said:
Optimizing for 4 and optimizing for 8 aren't going to produce the same results. 4 divides evenly into 8, so there will be some benefit, but not as much as for devices with 4GB of ram.

That said, I think optimizing for 4 is a decent middle-ground.
Click to expand...

They're unfortunately optimizing for information extraction so they can sell ads and AI "solutions" to already paying customers. The only thing that will change this is a rise in Linux as an alternative, otherwise I'm not convinced they particularly care.
 

“CXMT’s Capacity Woes Likely Mean Its Partnership With Apple Won’t Blossom To The Same Heights As Korean Rivals; DRAM May Be Limited To Product Testing Only​


Companies such as Samsung, SK hynix, and Micron have chosen to prioritize AI customers because they can place mammoth-sized DRAM orders, pushing Apple into a corner as it strategizes how to overcome this supply chain headache. While CXMT was previously reported to be a beacon of hope, the Chinese memory maker is facing a string of problems, suggesting that even as Apple tests the manufacturer’s DRAM in its products, the deal may not progress further.

Even if Apple receives the Trump Administration’s blessing to procure CXMT’s DRAM shipments, the Chinese firm needs to solve its production problems​

Testing CXMT DRAM in a variety of Apple products has been discussed previously, but The Wall Street Journal is adamant that the Chinese manufacturer’s memory chips will be used in Apple products sold exclusively in China. While Apple’s aim appears to be broadening its supply chain horizon so it can protect itself from future uncertainty regarding shipments, it matters little if CXMT cannot keep up with demand.”

https://wccftech.com/cxmt-apple-dram-capacity-testing-limits/
 

“TSMC Holding $1 Billion Of Apple’s A20 Pro Supply Refuted By Analyst, Says That DRAM Shortage Is Very Real, But Report Underestimates The Two’s Robust Operational Planning​


The DRAM shortage has certainly forced a myriad of companies to contemplate their uncertain future, but Apple doesn’t appear to be on this list. Earlier, we reported that the iPhone maker was facing a supply chain migraine of epic proportions, where TSMC was sitting on a whopping $1 billion worth of A20 Pro supply because of inadequate DRAM supply. Fortunately for Apple, an analyst believes these claims are nothing but hot air and that Apple’s and TSMC’s tightly integrated and efficient operations are being severely underestimated.

Analyst Ming-Chi Kuo believes Apple has planned its 2nm chipset production schedule at least three months in advance, aligning volume with the exact amount of DRAM supply available”​

https://wccftech.com/tsmc-apple-a20-pro-dram-shortage-1-billion-refuted/
 

“CXMT Hits 90% Yield on 17nm DDR5 Chips, Closing the Gap With Micron & Others – Report​


Chinese memory upstart CXMT is making important progress with its DDR5 memory chips, according to a report in the Chinese media. The firm has become a regular part of industry news as it grows in importance due to high demand from the local industry and Apple's interest in buying its chips to ease the pressure of the global memory chip shortage.
Now, today's report suggests that CXMT's DDR5 memory chips, marketed as being manufactured on the 17-nanometer manufacturing process, have reached a 90% yield to allow the firm to make major progress in its ability to ship large volumes.”

https://wccftech.com/cxmt-hits-90-y...ps-closing-the-gap-with-micron-others-report/
 
erek said:

“CXMT Hits 90% Yield on 17nm DDR5 Chips, Closing the Gap With Micron & Others – Report​


Chinese memory upstart CXMT is making important progress with its DDR5 memory chips, according to a report in the Chinese media. The firm has become a regular part of industry news as it grows in importance due to high demand from the local industry and Apple's interest in buying its chips to ease the pressure of the global memory chip shortage.
Now, today's report suggests that CXMT's DDR5 memory chips, marketed as being manufactured on the 17-nanometer manufacturing process, have reached a 90% yield to allow the firm to make major progress in its ability to ship large volumes.”

https://wccftech.com/cxmt-hits-90-y...ps-closing-the-gap-with-micron-others-report/
Click to expand...
We now live in a world where we're rooting for China... what's next, North Korea RAM?
 
Honestly there will never be enough memory to do what they want. I wouldn't be surprised if we ran out of the required minerals before we ran out of oil. At least oil is somewhat easy to get from the ground.

They'll be demolishing data centers to construct mines before it's all over.
 
Nobu said:
Honestly there will never be enough memory to do what they want. I wouldn't be surprised if we ran out of the required minerals before we ran out of oil. At least oil is somewhat easy to get from the ground.

They'll be demolishing data centers to construct mines before it's all over.
Click to expand...
what about all the buried minerals within landfills
 
erek said:
what about all the buried minerals within landfills
Click to expand...
For mining or recovery of minerals is all about how concentrated it is which is related to cost effectiveness. Landfills are not are not concentrated enough therefore not cost effective.
 
erek said:
what about all the buried minerals within landfills
Click to expand...
You thought mining was expensive, wait until they have to pay someone to sort and separate all that mess. Besides, many landfills had things built on top of them. The whole vehicle maintenance dept here was built on a landfill -- they had to drive piles to keep the building(s) from sinking, and you can see a large space open under the body shop concrete.
 
Nobu said:
You thought mining was expensive, wait until they have to pay someone to sort and separate all that mess. Besides, many landfills had things built on top of them. The whole vehicle maintenance dept here was built on a landfill -- they had to drive piles to keep the building(s) from sinking, and you can see a large space open under the body shop concrete.
Click to expand...
remember the Love Canal debacle?

History and Contamination
    • The Digging: Started in the 1890s by William T. Love as a planned hydroelectric power canal, the project was abandoned and left as a partial ditch. [1, 2]
    • Chemical Dumping: Between 1942 and 1952, Hooker Chemical used the site to dump chemical byproducts, sealed it with a clay cap, and sold the property for $1 in 1953. [1, 2, 3]
 
El_Capitan said:
We now live in a world where we're rooting for China... what's next, North Korea RAM?
Click to expand...

I'd buy a 64 GB kit of DDR5-6000 labeled "Dear leader's 100% hole in one special version memory" if it was sold for $200.


erek said:
remember the Love Canal debacle?

History and Contamination
    • The Digging: Started in the 1890s by William T. Love as a planned hydroelectric power canal, the project was abandoned and left as a partial ditch. [1, 2]
    • Chemical Dumping: Between 1942 and 1952, Hooker Chemical used the site to dump chemical byproducts, sealed it with a clay cap, and sold the property for $1 in 1953. [1, 2, 3]
Click to expand...

Fully familiar with it, it's about 35 miles from where I live. Much of Niagara County (where love canal is located) is a polluted mess. Dumping was prevalent there, not just love canal.

They're still finding contamination to this day, in fact 4 days ago to be precise: https://www.wivb.com/news/investiga...-after-uncovering-suspected-industrial-waste/
 

"Even if Apple receives the Trump Administration’s blessing " Here it is again that five letter word beginning with 'T'​

 

“Micron poised for "structural reset" in earnings power, UBS says, as HBM squeeze tightens further​


Micron poised for "structural reset" in earnings power, UBS says, as HBM squeeze tightens further
Micron Technology Inc (NASDAQ:MU)'s earnings power looks headed for a structural reset, according to UBS, which pointed to memory supply so tight that Nvidia was forced to de-spec its upcoming VR300 GPU.
UBS said Nvidia was forced to scale back the amount of high bandwidth memory, or HBM, in each VR300 GPU due to tight supply, cutting planned content to 512GB of HBM4E per chip from 768GB. VR300 is part of Nvidia's upcoming Rubin Ultra lineup of AI accelerators. The reduced memory content lets Nvidia build more GPUs with the same constrained memory supply, and the bank said the resulting increase in VR300 shipments is significant enough to lift its overall 2027 HBM consumption estimate to 61.5 billion gigabits, up from a prior 58.7 billion.“

https://www.proactiveinvestors.com/...-as-hbm-squeeze-tightens-further-1096831.html
 

“Nvidia reportedly testing lower memory configs of Rubin Ultra as memory shortage bites back — designs tested include as little as 192 GB and step back to HBM4​


The chickens are apparently coming home.

Nvidia is reportedly testing variations of its upcoming Rubin Ultra accelerator with less memory due to concerns it won't be able to source enough HBM. Some versions include just 192 GB of memory and use HBM4 instead of HBM4E, as originally announced, according to The Information. The report confirms an earlier comment from firm SemiAnalysis about a potential Rubin Ultra memory downgrade.“

https://www.tomshardware.com/pc-com...ude-as-little-as-192-gb-and-step-back-to-hbm4
 

“Intel CEO hints at return to the memory business — says market is ripe for innovation, hints at stacking memory and CPU​


Memory again?

Intel CEO Lip-Bu Tan has revealed that new memory architectures — once thought a commodity business — are now strategically interesting and one of his pet projects, while speaking about the comeback of the American chip industry. He further noted that the memory industry is ripe for innovation, and also hinted at exploring ways to stack memory on top of a CPU.“

https://www.tomshardware.com/pc-com...r-innovation-hints-at-stacking-memory-and-cpu
 
erek said:

“Intel CEO hints at return to the memory business — says market is ripe for innovation, hints at stacking memory and CPU​


Memory again?

Intel CEO Lip-Bu Tan has revealed that new memory architectures — once thought a commodity business — are now strategically interesting and one of his pet projects, while speaking about the comeback of the American chip industry. He further noted that the memory industry is ripe for innovation, and also hinted at exploring ways to stack memory on top of a CPU.“

https://www.tomshardware.com/pc-com...r-innovation-hints-at-stacking-memory-and-cpu
Click to expand...
That seems like a mistake to me.

Intel's not in a good position for a massive multi-year R&D effort, even ignoring the risk of the current bubble popping in a few years and them entering the market just in time for the ram supercycle to enter it's oversupply and crashing prices phase.
 
erek said:
“Intel CEO hints at return to the memory business — says market is ripe for innovation, hints at stacking memory and CPU
Click to expand...
To me the titlte and this mention ... Given the current profitability of 3D NAND and DRAM makers, producing memory is certainly a good business again and will remain profitable for some time. ...

Does not necessarily match what Tan is saying there, seem more about new type of cpu cache or compute-memory layer than intel returning into the dram for other business...

That sait the TeraFab memory ambition do not have a fab partner yet... a lot of the risk could be financed by the non intel entity and maybe Intel could get involved back in traditional dram business in some capacity that way... but that does not sound like a pet project, something fully major.
 

“Micron's top brass wins the memory crisis corpo double speak award: 'We do not have line of sight as to when that supply vector will intersect the demand vector'​


A lot of words for "don't expect things to get better soon."

A slight bastion of hope for those affected by the memory crisis (ie, anyone who wants to buy new electronics in the next few years) is the planned arrival of new fabrication facilities, pumping out vast amounts of wafers that will be turned into memory modules. According to Micron, though, we shouldn't hold our breath.“

https://www.pcgamer.com/hardware/me...pply-vector-will-intersect-the-demand-vector/
 
erek said:

“Micron's top brass wins the memory crisis corpo double speak award: 'We do not have line of sight as to when that supply vector will intersect the demand vector'​


A lot of words for "don't expect things to get better soon."

A slight bastion of hope for those affected by the memory crisis (ie, anyone who wants to buy new electronics in the next few years) is the planned arrival of new fabrication facilities, pumping out vast amounts of wafers that will be turned into memory modules. According to Micron, though, we shouldn't hold our breath.“

https://www.pcgamer.com/hardware/me...pply-vector-will-intersect-the-demand-vector/
Click to expand...
Wow that's a fancy way of saying "dunno when things will turn around..."
 
erek said:
A lot of words for "don't expect things to get better soon."
Click to expand...
that a strange way to say this, what ?

get better soon for Micron is not what the author of this strange article think it would mean... what the double speak in there ?
 
Nobu said:
Honestly there will never be enough memory to do what they want. I wouldn't be surprised if we ran out of the required minerals before we ran out of oil. At least oil is somewhat easy to get from the ground.

They'll be demolishing data centers to construct mines before it's all over.
Click to expand...

The market will adapt. The AI guys don't want to surrender margin to the memory companies. The current state of things demands more memory. That doesn't mean a future state can't be created that demands less. If memory is the constraint, you can bet they have their top guys working on ways to remove that constraint.
 
HockeyJon said:
The market will adapt. The AI guys don't want to surrender margin to the memory companies. The current state of things demands more memory. That doesn't mean a future state can't be created that demands less. If memory is the constraint, you can bet they have their top guys working on ways to remove that constraint.
Click to expand...
There really is no substitute for more memory. It is always better than less. You can compress data, but you still have to be able to decompress it before using it (except for some very specific data types and certain operations thereon). You could move the computation into memory, but that only optimizes the operation, not the storage efficiency.
 
That is to say, the best they could do is change memory technology to use material that is more available or more space efficient, but you'll have to maintain the same access speed and similar density for it to be useful in this capacity. Against RAM, that's a tough ask.
 
Nobu said:
There really is no substitute for more memory. It is always better than less. You can compress data, but you still have to be able to decompress it before using it (except for some very specific data types and certain operations thereon). You could move the computation into memory, but that only optimizes the operation, not the storage efficiency.
Click to expand...

Right, and improvements to compression algorithms can help.

https://arstechnica.com/ai/2026/03/...-ai-memory-usage-without-sacrificing-quality/

We don't know what the future will hold, but the current incentive is to reduce memory usage wherever possible. That incentive didn't previously exist because memory was cheap so you didn't have to worry as much about compression and efficient use, and therefore there wasn't as much development there. Now you have to worry about it, so resources will be dedicated to finding solutions.

The industry is in an arms race right now. Once Sam Altman declared he was buying 40% of the annual RAM production capacity, everyone else decided they better rush to get what they can. This won't last forever. I have little don't there will be technological advancements and adjustment to the supply side over time.
 
HockeyJon said:
Right, and improvements to compression algorithms can help.

https://arstechnica.com/ai/2026/03/...-ai-memory-usage-without-sacrificing-quality/

We don't know what the future will hold, but the current incentive is to reduce memory usage wherever possible. That incentive didn't previously exist because memory was cheap so you didn't have to worry as much about compression and efficient use, and therefore there wasn't as much development there. Now you have to worry about it, so resources will be dedicated to finding solutions.

The industry is in an arms race right now. Once Sam Altman declared he was buying 40% of the annual RAM production capacity, everyone else decided they better rush to get what they can. This won't last forever. I have little don't there will be technological advancements and adjustment to the supply side over time.
Click to expand...

"However, the companies creating this technology could also use the newly freed-up memory to run more complex models." Is exactly what will happen.
 
Nobu said:
"However, the companies creating this technology could also use the newly freed-up memory to run more complex models." Is exactly what will happen.
Click to expand...

Jevon's Paradox, but in that situation, memory companies will see the demand is sustainable and build more production capacity to meet that demand. I doubt the current productive capacity will remain the same in the face of increasing demand.
 
erek said:

“Intel CEO hints at return to the memory business — says market is ripe for innovation, hints at stacking memory and CPU​


Memory again?

Intel CEO Lip-Bu Tan has revealed that new memory architectures — once thought a commodity business — are now strategically interesting and one of his pet projects, while speaking about the comeback of the American chip industry. He further noted that the memory industry is ripe for innovation, and also hinted at exploring ways to stack memory on top of a CPU.“

https://www.tomshardware.com/pc-com...r-innovation-hints-at-stacking-memory-and-cpu
Click to expand...
Actually, Intel is in the perfect position for this. You can make all the CPU's and GPU's in the world, but without ram then you're selling nothing. Considering Intel has fabs and DRAM doesn't use anything fancy beyond 10nm, then I could see Intel leveraging this advantage. You could go ARM and buy $2,000 worth of LPDDR5X memory, or go Intel and get the equivalent amount of ram 1/4 of the price. AMD should do the same thing with Global Foundries, their old fling. He who controls the ram controls the future of compute it seems.
 
DukenukemX said:
Actually, Intel is in the perfect position for this. You can make all the CPU's and GPU's in the world, but without ram then you're selling nothing. Considering Intel has fabs and DRAM doesn't use anything fancy beyond 10nm, then I could see Intel leveraging this advantage. You could go ARM and buy $2,000 worth of LPDDR5X memory, or go Intel and get the equivalent amount of ram 1/4 of the price. AMD should do the same thing with Global Foundries, their old fling. He who controls the ram controls the future of compute it seems.
Click to expand...
intel could tie up with SK Hynix

SK hynix Reportedly Eyed as Potential Partner to Operate Intel’s Ohio Fab After Denying Acquisition Plans​

2026-07-23

https://www.trendforce.com/news/202...els-ohio-fab-after-denying-acquisition-plans/
 
DukenukemX said:
Actually, Intel is in the perfect position for this. You can make all the CPU's and GPU's in the world, but without ram then you're selling nothing. Considering Intel has fabs and DRAM doesn't use anything fancy beyond 10nm, then I could see Intel leveraging this advantage. You could go ARM and buy $2,000 worth of LPDDR5X memory, or go Intel and get the equivalent amount of ram 1/4 of the price. AMD should do the same thing with Global Foundries, their old fling. He who controls the ram controls the future of compute it seems.
Click to expand...
It is smart, allows you to sell product with RAM without paying whatever markup other manufacturers charge, and then add your own markup to that product. It's easy margins, and you know your customers are desparate for an alternative.

Edit: the risk, of course, is the market imploding, or materials cost outstripping whatever savings you have from doing the manufacture yourself.
 
DukenukemX said:
Actually, Intel is in the perfect position for this. You can make all the CPU's and GPU's in the world, but without ram then you're selling nothing. Considering Intel has fabs and DRAM doesn't use anything fancy beyond 10nm, then I could see Intel leveraging this advantage. You could go ARM and buy $2,000 worth of LPDDR5X memory, or go Intel and get the equivalent amount of ram 1/4 of the price. AMD should do the same thing with Global Foundries, their old fling. He who controls the ram controls the future of compute it seems.
Click to expand...
That a complete misunderstanding I think about memory wafer and compute wafer complexity. There is a reason CXMT has such an hard time making 17nm work.

Compute wafer expertise do not simply translate to memory wafer like that (and vice versa) and HBM 3 level memory is not necessarily less fancy than what people use for compute TSMC 3N process, memory wafer individual cell are less thin on a 2d plane but have many more layers (they grow vertically), 10nm dram required extreme advanced UV (EUV) lithography.

Nobu said:
It is smart, allows you to sell product with RAM without paying whatever markup other manufacturers charge, and then add your own markup to that product. It's easy margins, and you know your customers are desparate for an alternative.
Click to expand...
Even without that markup, achieving to make ram cheaper than what you can buy it for instead is one of the hardest thing a group of people can start to try to do.
 
Last edited:
You must log in or register to reply here.
Back
Top