• 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

Considering how close the switch 2 gpu was samsung node wise and Ampere cores wise to a 3060, not sure if it is that intriging of a move, lot of capacity becoming available again in between Switch 2 run with not that big of a change to make to shift toward making 3060 again (specially once the early 2026 surge get over).

Lot of lovelace use regular GDDR6 and could have lowered the memory clock to use slower module I imagine and the geforce 5050 also use GDDR6, if they do this, not sure if it is related to memory price.
 
LukeTbk said:
Considering how close the switch 2 gpu was samsung node wise and Ampere cores wise to a 3060, not sure if it is that intriging of a move, lot of capacity becoming available again in between Switch 2 run with not that big of a change to make to shift toward making 3060 again (specially once the early 2026 surge get over).

Lot of lovelace use regular GDDR6 and could have lowered the memory clock to use slower module I imagine and the geforce 5050 also use GDDR6, if they do this, not sure if it is related to memory price.
Click to expand...

"ASRock Develops HUDIMM Memory Standard: DDR5 with Just One Sub-channel​

by btarunr Today, 14:26 Discuss (9 Comments)
You know we've hit rock-bottom with memory affordability when ASRock innovates a new memory standard to lower prices. The new "HUDIMM," or half unbuffered DIMM, is an ASRock-innovated standard. It calls for UDIMMs with half a rank of memory, populating just one of the two 40-bit sub-channels. Such a DIMM would only offer half its bandwidth even at its rated memory clock, and of course half the density. The HUDIMM standard is targeted at entry-level builds and business desktops that just want a modern platform for everyday tasks, and something to tide over the DDR5 memory crunch. ASRock partnered with Team Group to manufacture the first HUDIMM memory modules, which it tested to work on its Intel 600-series, 700-series, and 800-series chipset motherboards. HUDIMM support probably requires some UEFI firmware-level awareness of the standard, and ASRock is expected to release firmware updates for the same.

Here's the fun part—ASRock made it possible for end-users to pair HUDIMMs with regular UDIMMs that have two 40-bit sub-channels; for example, pairing an 8 GB HUDIMM with a 16 GB UDIMM, to achieve asymmetric capacities such as 24 GB, and the bandwidth of at least 3 DDR5 sub-channels. ASRock also developed HSODIMM, which—you guessed it—is a DDR5 SO-DIMM with just one sub-channel. Team Group will manufacture these HSODIMMs, and ASRock's Deskmini series mini-PCs that are based on MoD (mobile on desktop) platforms, will implement support for them. ASRock is perfectly faced to address the client computing market during the DDR5 famine; besides the HUDIMM/HSODIMM innovation, the company is developing a new crop of Socket LGA1700 motherboards with both DDR4 and DDR5 memory slots."

1776454410932.png
 

“Recent Linux VRAM Management Improvements Resurrect 4 GB AMD Radeon RX 6500 XT for Some Games​

by Cpt.Jank Today, 01:54 Discuss (6 Comments)
Natalie Vock, a well-known Valve contractor and Linux graphics driver developer, recently debuted a new patch that enabled better VRAM management on Linux for GPUs with low VRAM. When we originally covered the set of kernel patches, we noted that it could cut VRAM usage in half in some applications, potentially making certain aging graphics hardware viable for gaming where it may not have been before the patches. Aside from some early tests by Vock herself, not much other data was available at the time to draw any conclusions about the potential performance improvements. Now, thanks to NJ Tech on YouTube, we have some idea of how the patch could improve performance on a GPU like the AMD Radeon RX 6500 XT, which has a mere 4 GB of VRAM. The YouTuber tested the GPU across nine games, including some heavy hitters, like Crimson Desert, Hogwarts Legacy, and Cyberpunk 2077 in CachyOS, which was the first Linux distribution to package and release Vock's kernel patches in what it calls GPU Booster. In the current GPU market, it would be nice to have a silver bullet to make low VRAM GPUs viable, but the results are far more varied than that, with some games seeing no improvement and others seeing up to 100% increases in FPS.

In Alan Wake II, VRAM use is actually increased, but there's a more than 2× increase in FPS, going from 14 FPS to a very playable 42 FPS average. In Resident Evil: Requiem, VRAM use is identical, but there's a 16% FPS increase, and in Silent Hill f, the story is similar, with identical VRAM utilization but marginal performance increases. Crimson Desert saw a decrease in VRAM usage, but there was no measurable performance increase as a result, as was the case with Hogwarts Legacy and Cyberpunk 2077, both of which saw reduced VRAM usage but an increase of 1 FPS average in the former and identical performance in the latter. The Last of Us Part 2 actually saw a 1 FPS drop in both average and 1% low FPS, but it seems as though there was just too much graphics memory pressure for Vock's VRAM patches to mitigate the issue. Death Stranding 2 and Marvel's Spider-Man 2 also saw little to no improvement despite reduced VRAM utilization in the former. The full video by NJ Tech follows.“
 
erek said:
“Enthusiast fabricates his own RAM in a garden shed cleanroom

Last time we saw Dr. Semiconductor, he had completed his backyard shed with all the equipment and clean room capability to fabricate silicon chips.

Now he posts about checking whether someone can DIY themselves out of the DRAM crisis.”

https://blog.adafruit.com/2026/04/23/enthusiast-fabricates-his-own-ram-in-a-garden-shed-cleanroom/
Click to expand...
Not exactly. He made part of DRAM. A very tiny part of it.

View: https://youtu.be/h6GWikWlAQA?si=Uu4XkW5r1i5wIAJm
 
Red Falcon said:
From the article, and very accurate:
View attachment 799606

As the old saying goes, let it burn. 🔥 🔥 🔥
Click to expand...

🔥 “Valve Confirms Staggered Hardware Launch Due to DRAM Crisis: "This Doesn't Have RAM in It"​

by Cpt.Jank Today, 01:40 Discuss (2 Comments)
Valve confirmed not too long ago that it had delayedannouncing the pricing of the Steam Machine, Steam Frame, and Steam Controller because of the ongoing DRAM shortage and price volatility crisis that has plagued the PC hardware industry in recent months. Eventually, the gaming giant caved and announced the Steam Controller ahead of the rest of the hardware that was meant to ship alongside it, and in a recent interview with Polygon, Valve confirmed what we all suspected: The Steam Machine and Steam Frame are still being held back by the DRAM crisis. As Valve hardware engineer, Steve Cardinali, explained about the early launch of the controller: "This doesn't have RAM in it, and it's not as complicated to start getting out the door for us. We're ready for it. We wanted to build up quantity so that we could try to address everybody who wants one at launch, but it's possible that the demand for it far exceeds our expectations."

He goes on to explain that there was never a hard-and-fast rule that the Controller, Frame, and Machine had to launch together, although he confirmed that Valve would not launch the Machine ahead of the Controller. It's safe to assume, then, that the Steam Machine and Steam Frame were originally meant to launch around May as well—Cardinali says that the Controller and Machine are "a pair made in heaven," so it only makes sense to try to launch them around the same time, even if "there's no point in holding it back while we work through the other stuff." According to VR industry insider, Brad Lynch on X, the Steam Machine is more severely affected by the memory supply issues than the Steam Frame, since it relies on on-package RAM and on mobile RAM, which doesn't appear to be as severely impacted by the shortages. Reading between the lines, it seems entirely possible that the Steam Frame may launch ahead of the Steam Machine as long as RAM prices remain elevated and unstable.“
 
A surcharge eh? Such doom and gloom today. And here I thought some memory (desktop) had stabilized the past two weeks or so.
 
  • Like
Reactions: erek
like this

"Framework Laptop 16 Gets NVIDIA RTX 5070 12 GB Upgrade Module for Eyewatering Price​

by Cpt.Jank Today, 16:30 Discuss (1 Comment)
Framework, known for its repairable and upgradeable gaming and productivity laptops, has officially released a Laptop 16 graphics upgrade module with the new NVIDIA GeForce RTX 5070 12 GB. Framework's original Laptop 16 graphics expansion module, which upgrades the original $350 AMD Radeon RX 7700S, is based on the NVIDIA GeForce RTX 5070 8 GB and costs $699. Now, the RTX 5070 12 GB upgrade comes in at a whopping $1,199—a 72% price increase for a 50% increase in VRAM. It's unclear where the drastic price increase comes from, but some of it can likely be explained by the low volume Framework has to contend with, and some of it can be attributed to the ongoing DRAM crisis wracking the PC hardware market."
 
erek said:

"Framework Laptop 16 Gets NVIDIA RTX 5070 12 GB Upgrade Module for Eyewatering Price​

by Cpt.Jank Today, 16:30 Discuss (1 Comment)
Framework, known for its repairable and upgradeable gaming and productivity laptops, has officially released a Laptop 16 graphics upgrade module with the new NVIDIA GeForce RTX 5070 12 GB. Framework's original Laptop 16 graphics expansion module, which upgrades the original $350 AMD Radeon RX 7700S, is based on the NVIDIA GeForce RTX 5070 8 GB and costs $699. Now, the RTX 5070 12 GB upgrade comes in at a whopping $1,199—a 72% price increase for a 50% increase in VRAM. It's unclear where the drastic price increase comes from, but some of it can likely be explained by the low volume Framework has to contend with, and some of it can be attributed to the ongoing DRAM crisis wracking the PC hardware market."
Click to expand...
"It's unclear where the drastic price increase comes from"

Have they been living in a cave?
 
sharknice said:
I had a friend tell me ram prices were high because of tarrifs. 🤦
When I told him it's because of AI he tried to push back.
This is what Reddit does to people's brains.
Click to expand...
Well it's a confusing thing to have to explain, that's why. It's easier to blame ORANGE MAN BAD and call it a day than having to actually research and challenge one's own biases.
 
kram182 said:
Worth it if it gets us to adopt nuclear (as it seems it is/might)
Click to expand...
Modern nuclear designs are extremely safe. The exposure levels of radiation are higher in basements from naturally occurring radon, even after mitigation is installed. The problem is that many have opinions on nuclear reactors from technology data from 50 to 60 years ago. Miniature nuclear reactors for modern submarines have been low exposure since the 1990's for example. Chemical exposure from consumer products and manufacturing with carcinogenic forever chemicals are a much higher threat to humanity yet these same people that fear nuclear power do not fear what currently gives them cancer.

If only AI could figure out a way to create lower cost GPUs and memory chips? We need AI to eliminate the scarcity it is creating.
 
Last edited:
sharknice said:
I had a friend tell me ram prices were high because of tarrifs. 🤦
When I told him it's because of AI he tried to push back.
This is what Reddit does to people's brains.
Click to expand...
To be fair, it is both ... its just that AI is doing 90% of the lifting.
 
Ididar said:
To be fair, it is both ... its just that AI is doing 90% of the lifting.
Click to expand...
Actually, it's more than 90%. Nobody is going to balk at an increase of a few dollars (tariffs, talking full stick of 16GB DDR5 increasing by less than $5).
 
Taiwan semiconductors are exempt from US tariffs currently. A 25% tariff is currently in place for Japan and S.Korea. AI demand for memory (and other components) is by far the main driving cause for the price increases.
 
Mega6 said:
Taiwan semiconductors are exempt from US tariffs currently. A 25% tariff is currently in place for Japan and S.Korea. AI demand for memory (and other components) is by far the main driving cause for the price increases.
Click to expand...
True, but when the price rises by a factor of 4, the tariff also rises by that same factor of 4 since its based on the sale price of the good and not how much it actually cost to make or anything.
 
sfsuphysics said:
True, but when the price rises by a factor of 4, the tariff also rises by that same factor of 4 since its based on the sale price of the good and not how much it actually cost to make or anything.
Click to expand...
So why exactly did the price increase by a factor of four? AI. Tada.
 
I clipped AI's wordy response down to it's conclusion:

In short, the price spike is less about the inherent value of DDR5 and more about the **market's inability to meet the massive, specialized
demand** that the AI boom and the technological transition required.
Click to expand...
 
Ididar said:
Are we really arguing over a few percentage points? AI is mostly the cause, yes. Tariffs add insult to injury.
Click to expand...
Idk you and your pal are the ones who keep bringing up (in your own words - a few percentage points) tariffs and making a big deal out of it. Little wonder why MU is up 400% this year.
 
Ididar said:
Are we really arguing over a few percentage points? AI is mostly the cause, yes. Tariffs add insult to injury.
Click to expand...
I wouldn't be surprised if it turns out it wasn't AI but it's price fixing like every other time computer parts got high and it turns out the fires ECT they claim it was because never happened ECT
 
tunatime said:
I wouldn't be surprised if it turns out it wasn't AI but it's price fixing like every other time computer parts got high and it turns out the fires ECT they claim it was because never happened ECT
Click to expand...
I know everything is a conspiracy nowadays and it couldn’t possibly be simply supply vs demand economics.
 
The Cobra said:
Here is a good article over the memory crap that is happening. I gifted the article.

https://www.wsj.com/tech/ai-has-mad...a?st=Gs7k43&reflink=desktopwebshare_permalink
Click to expand...

"Samsung Q1 2026 Results: Memory Profit Up Nearly 50x, Warns of 2027 Shortage​

by Nomad76 Today, 09:11 Discuss (13 Comments)
Samsung has reported a record quarterly profit for Q1 2026, with chip division operating income jumping to 53.7 trillion won ($36.15 billion), a 49% increase from the 1.1 trillion won ($745 million) posted in the same quarter last year. That figure accounted for 94% of the company's total operating profit of 57.2 trillion won ($38.7 billion), which itself was up from 6.69 trillion won a year prior. Overall revenue rose 69% year over year to 133.9 trillion won ($90.6 billion). Its record profit came mostly from the AI data center sector that pushed demand for advanced memory well beyond what Samsung and its peers can supply. The company has signed multi-year binding supply contracts with customers looking to lock in capacity, though it hasn't disclosed names or terms. Samsung's memory chief Kim Jaejune told analysts that supply is already falling well short of demand, and that based on orders already received, the gap in 2027 is expected to be even wider than in 2026.

On HBM specifically, Samsung said it began mass-production sales of HBM4 for NVIDIA's Vera Rubin platform in February and is on track to more than triple HBM revenue this year versus last. However, rising memory prices are affecting the profit of other Samsung businesses. Mobile and network division profit fell 35% to 2.8 trillion won, due to higher component costs, and the display division saw operating profit drop 20% to 400 billion won ($270 million). Reuters also reports the risk of a strike, with unions representing a large portion of Samsung's South Korean chip workforce considering work stoppages over pay disputes. For the rest of 2026, Samsung expects that increasing demand will keep pushing memory, foundry, and display businesses."
 
robots and cars of the futur local compute capcity could be something with more computer than a full desktop tower now.

As for consumer dram revenu being more the double than last year for Samsung, does that show a trend for it to disapear how just show how popular it is and people ready to pay for it (i.e. 2 elements that assure a nice market for it coming up), if no one does it, HardOCP member just have to start doing it and own 100% of that market.
 
gtrguy said:
I figure we're probably the last generation to own their own personal computing devices with decent capabilities. Looking like the cloud will be the future for most.
Click to expand...
AI: yes.... Yes.... YES..... oh YES....
 
Mega6 said:
I know everything is a conspiracy nowadays and it couldn’t possibly be simply supply vs demand economics.
Click to expand...
If only there was a demand for AI which is causing ram supply to be so high. Not only that, but the demand is for ram not yet made, for graphic cards not yet made, for data centers that may not get made. I'm not saying AI doesn't have uses, but we're obviously going nuts with who's paying for it.
ceos AI.jpg
 
sfsuphysics said:
True, but when the price rises by a factor of 4, the tariff also rises by that same factor of 4 since its based on the sale price of the good and not how much it actually cost to make or anything.
Click to expand...
Tariff costs are dutiable value times tariff rate, not the market price.

This is the price to the foreign seller plus shipping and insurance costs.

If there multiple sellers, it gets more complicated and from what I have read is usually cheaper.

So fictional example might be a 40k car with factory whole price of 28k, plus maybe 1k shipping and tariffs only apply to that number (made up numbers). The retail prices are usually set to some variable marginal price above costs

So Not sure how impactful tariffs are to retail specifically, but it's less than a straight percentage of retail price.

Unless someone knows more details, that's my understanding.
 
1.1.2.3.5... said:
So Not sure how impactful tariffs are to retail specifically, but it's less than a straight percentage of retail price.

Unless someone knows more details, that's my understanding.
Click to expand...
yes that about it, it is how much a third party would pay for the raw shipment receive, so if samsung memory bring memory for samsung phone in the country they cannot use say the cost of making them, or the cost samsung phone division paid necessarily, how much Apple and others would have paid for it, the market for the big distributor that buy big volume is what they try to use as a cost to apply the tarrif, not the in store resales values (that you cannot necessarily know anyway, they can be sold months from now, with rebate and what not).

So in your example the car importer cannot use factory whole price, but how much toyota/Honda USA and relesser would be ready to pay for that type of cars, but that number would be higher than gross fabrication/transport cost but lower than end user cost because of some margins of profit on them to be made.

It tend of course to follow (the more memory stick are sold on newegg, the more memory stick importer/distributor are ready to pay for them) but it is not 1:1
 
Last edited:
You must log in or register to reply here.
Back
Top