New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Is any seller planning to sell the ram they own lol?
You can't. The price of all hosting is going to increase substantially this year.
Case in point, I paid $210 for a 64gb ddr5 kit a few years ago for my build. The same kit is $810. That's bonkers, and providers have to eat that cost.
At least it is not from Nice Parfume LLC Romania, in that case you would get..... a brick
Love this guy!
Me too!
It's a real one, you don't see them so often anymore.
😂
would be great if China becomes the number one source for ddr5 in 2027, we can get lower prices, memory is not that hard to copy/make (not like a gpu)... I guess usa memory makers can fall to the wayside... too expensive... bye bye Gskill
Let me be sincere, I could not care less who makes parts, CN, US, EU, Africa..... That is completely irrelevant to me.
What does matter is quality control. You don’t need quantum level genius anymore to manufacture chips. We’re over three decades into mass semiconductor production; these parts are no longer exotic. What is exotic today is greed.
As a product gets to mass production usually within 3-5 years it reaches maturity ( by that tame the assembly line gets perfected, bugs get fixed and so on), this is valid for any type of product. Also, by the 3-5'th year manufacturer knows where he can cut costs and where not to. ( by try and error on the expanse of the consumer as usual )
What still amazes me is that we don’t have slow, large-capacity storage drives based on flash memory today, something in the ~200 MB/s range, for example. I honestly doubt anyone in the industry truly cares whether that comes in a 3.5″ or 2.5″ form factor. Capacity and price should matter more than peak benchmark numbers.
I am one of those that played with 40MB drives in the early 90's, we are in 2026, that is ~35+ years of technological advancements, yet still, for the past 10 years, the 512MB-1TB Flash Storage and 8-16GB RAM is the standard, and now it is at a 2-5x price. WTF? If that is not greed, I don't know what is.
Yes, advanced lithography is complex. But are we seriously expected to accept that on a planet of 8 billion people, only 3–5 companies are capable of manufacturing chips at scale? the math does not add up here. ( greed does make sense
)
Sincerely, I hope the CN will teach the "west" a lesson. not all CN products are bad or low quality, only the dirt cheap ones are trash, yet, those are as bad as the ones made by the "west". The logo on the box doesn’t magically change that.
PS: I am pissed.
Cheers!
EDIT:
And if we’re already in the stupidity zone, let’s not forget this little gem from history.
Back in the early ’90s, IANA was practically shitting it's pants over IPv4 exhaustion. NAT didn’t exist yet. Every printer, workstation, server thingy was expected to have a public IPv4 address. There was no real concept of LANs/vlans/security zones/ACL as we understand them today; it was, ironically, a true “Internet of Things” long before the term existed.
Network Address Translation was introduced in 1994, formalized in RFC 1631, largely driven by Kjeld Borch Egevang, and it immediately changed the game. Overnight, IPv4 went from “shit we’re about to run out” to “this will last a lot longer than anyone expected.”
And that’s the inconvenient truth: NAT effectively solved the IPv4 exhaustion problem in practice, which is exactly why IPv6 adoption stalled for so long. That, and the fact that putting letters in IP addresses still makes my head hurt
At the time, NAT did come with real performance concerns. Dedicated ASICs for address translation were expensive and largely nonexistent, and the first practical NAT implementations ran on general-purpose PCs of that era. Early commercial routers lacked the compute headroom to perform large-scale, stateful translation efficiently, and hardware acceleration mechanisms such as TCAM were either unavailable or not designed for this purpose.
That problem quietly disappeared by the late ’90s and early 2000s as general-purpose CPU performance increased dramatically and networking silicon evolved to support higher packet rates and more complex stateful processing.
So once again, advancements in semiconductor compute power and efficiency solved a problem that had caused widespread panic. That panic led to the design of an “interesting” architectural fix which, in practice, still has limited real-world necessity even today.
As a bonus, if I opened a can of worms here, and wasted 5 minutes of your time reading this:
But more importantly, NAT solved the numbers game.
Once address translation became viable, large organizations no longer needed entire /8 networks. Vast blocks of IPv4 space that had been allocated early often massively over-provisioned by today’s standards and suddenly became reclaimable, transferable, and usable by the public Internet.
In other words, NAT didn’t just delay IPv4 exhaustion it corrected a fundamentally wasteful allocation model.
In engineering, “simple” almost always beats “complicated”
As a result, the industry doubled down on the workaround, and the original fix lost urgency.
Just to be clear, I’m not saying IPv6 has no advantages. It absolutely does. I’m saying those advantages are often overstated relative to real-world needs.
NAT may be inelegant old school from a purist networking perspective, but it enforces a clear boundary. It forces intent. Things are reachable because someone explicitly made them reachable not because they exist.
Traffic should always be routed intentionally. Services should be reachable because you explicitly allow them to be, defined by clear rules and policy — not simply because a cable was plugged in. Default reachability is sloppy at best, and dangerous at worst.
Done, and now back to business fixing stuff.
And if you made it this far thank you for scrolling to the bottom 😄
Yep Host-C been through that all, since the mid 80's here with my apple iic assembly language, and 300 baud modem, working at digital equipment chips/fab which became Compaq, then HP and then finally Intel (felt like a bank being bought out over and over),.
I remember running around the office seeing if someone left off the terminator for the "high speed" coax cable network we had and programming in good old turbo pascal. I was also one of the first to get high speed cable internet from Comcast at 1 mbps!! (whoo hoo)
I'm still a big fan of my G Skill Trident Neo sticks in my machines but will switch if needed..
If you don't mind my curiosity, which distro you run on a 1 GB machine?
Have a couple of old notebooks which I like, but both collecting dust because of 1 GB RAM limitation.
As usual, it's geopolitics. Advanced lithography machines are made by the dutch with parts from german and US, but US has control over it since it's made using US technology.
On other hand, the advanced materials and high purity chemicals are also required. For example, flouro gas, silicon wafers, quartz (the real bottleneck).
It's just a game
True, do not hate the game
Still, PISSED I AM !
The latest version of Falkon (Blink browser) runs all the modern web in only 300 MB memory, I run it on a 1 GB Intel Atom occasionally.
The desktop in the picture has 2 GB memory, Debian. It runs everything that I care to run under it. It cannot run the Garbage Web—those web apps that take a minute to load and seize CPU and memory—but that is garbage by definition and I avoid it anyway.
Sigh, gotta replace 7.5 years old DDR4 16GB stick, got quoted for $135 (incl. tax).
I don't even know whether that price is good or not.
I just paid $250 for 10 used 8gb sticks. That price seems about normal for new sticks from what I'm seeing.
Yes, RAM is a problem right now 😅. Many are optimizing server loads, looking at used or ECC memory, renting from data centers with existing memory, and often sending servers to colocation—it's easier to control costs when prices fluctuate.
just make a swapfile of 20 terabyte
It's just crazy. For $135 I could've purchase low-end Samsung phone with 8GB of ram.
my laptop was built before the ram shortage and i did get 2x 32GB DDR5 (64GB DDR5 RAM) and for my desktop i got it at a weird time back in 25 with only 32GB of DDR5 ram.. i don't feel like buying more until ram prices crashes again...
Makes me wonder...I have about 1-2TB of ddr4 ecc registered in my filing cabinet right now. I guess I'm rich?
If you don't use them now or soon 100% sell them! Prices like these may not return, you can make a great profit!
My estimation: late 2027 to mid 2028. Industry analysts says it's the new normal. But, I disagree since the fundamental problem will be solved by then.
I hope LET hosts doesn't went all-in with the RAM craze. @tentor probably will get shouted& slapped by his accountant because of the asset write off
Seems like my whining about wanting more CPU and less RAM when it's sale time might come true now; glass so fucking half full.
Wait, your kinda right… if they don’t offer much ram they can offer more cpu! But it’s also true that cpu may get even more oversold!
We can't ever have nice things -.-
Yeah, it's not about all goods from CN being poor quality, but rather that they will design to the quality level, and more importantly price point, demanded by the customer.
There's a separate issue though that sometimes manufacturers will sometimes try their luck substituting in cheaper components / processes as time goes by to try to increase their profit margins and hoping the customer won't notice, especially on products that are directly shipped to the end customer. So, if you have specified a certain quality level, it's important to have your own QA test things rigorously.
Within China, companies are often fairly upfront about the real quality of given things, and for instance you can go into a mall in Shenzhen and tell someone you want 1000 USB sticks that present as a 256GB drive that's actually only 16GB, or whatever you want. They'll happily take a good 16GB drive and reprogram it as the 256GB one and stick it in the packaging for the 256GB one. If the quantities are reasonably low, they'll even do that for you on the spot!
I have no doubt that China absolutely can make RAM as good as e.g. SK Hynix or Micron, but equally buying it as an end customer is far riskier, especially in small quantities where returning it is more hassle than eating the loss.
But I can also see that this is a competitive advantage for China at the moment - with the US actively trying to stifle China's tech progress in AI, the US-led AI bubble having self-inflicted supply issues while China continues to have a steady supply for its own market seems like a win for them, and maybe they'd consider limiting exports outside China for as long as the US is limiting tech exports to China.
was ~35 usd at most ~6 moths ago, I might not recall that precisely, we only buy 64 GB 2400T ECC R and upwards modules, those are x4 the price today then 6 moths back.
Yeah. On September 2025, the RAM I bought was around that price too. It's actually 10-20% more expensive than the early 2025 pricing.
By the time RAM pricing went down, the cost of running LLM AI will be cheaper by factor of 2 or 3. Say, will you start hosting public model and selling the API tokens given the chance?