I’d also note that OP only created their account a day ago.

I wonder how much exact duplication each process has?

https://www.kernel.org/doc/html/latest/admin-guide/mm/ksm.html

Kernel Samepage Merging

KSM is a memory-saving de-duplication feature, enabled by CONFIG_KSM=y, added to the Linux kernel in 2.6.32. See mm/ksm.c for its implementation, and http://lwn.net/Articles/306704/ and https://lwn.net/Articles/330589/

KSM was originally developed for use with KVM (where it was known as Kernel Shared Memory), to fit more virtual machines into physical memory, by sharing the data common between them. But it can be useful to any application which generates many instances of the same data.

The KSM daemon ksmd periodically scans those areas of user memory which have been registered with it, looking for pages of identical content which can be replaced by a single write-protected page (which is automatically copied if a process later wants to update its content). The amount of pages that KSM daemon scans in a single pass and the time between the passes are configured using sysfs interface

KSM only operates on those areas of address space which an application has advised to be likely candidates for merging, by using the madvise(2) system call:

int madvise(addr, length, MADV_MERGEABLE)

One imagines that one could maybe make a library interposer to induce use of that.

I mostly use terminal-based software on Linux.

I think that the only programs I use much that embed a web browser are:

• Firefox

• Steam

• Some games that are Web-based and which I only run one of at once (Neo Scavenger, some RPGMaker-based games, probably some others).

I’ve also noticed that is you want a chest smaller than DDD, it’s almost impossible with some models — unless you specify that they are a gymnast.

That’s also another point of present generative AI image weakness — humans have an intuitive understanding of relative terms and can iterate on them.

So, it’s pretty easy for me to point at an image and ask a human artist to “make the character’s breasts larger” or “make the character’s breasts smaller”. A human artist can look at an image, form a mental model of the image, and produce a new image in their head relative to the existing one by using my relative terms “larger” and “smaller”. They can then go create that new image. Humans, with their sophisticated mental model of the world, are good at that.

But we haven’t trained an understanding of relative relationships into diffusion models today, and doing so would probably require a more sophisticated — maybe vastly more sophisticated — type of AI. “Larger” and “smaller” aren’t really usable as things stand today. Because breast size is something that people often want to muck with, people have trained models on a static list of danbooru tags for breast sizes, and models trained on those can use them as inputs, but even then, it’s a relatively-limited capability. And for most other properties of a character or thing, even that’s not available.

For models which support it, prompt term weighting can sometimes provide a very limited analog to this. Instead of saying “make the image less scary”, maybe I “decrease the weight of the token ‘scary’ by 0.1”. But that doesn’t work with all relationships, and the outcome isn’t always fantastic even then.

There are also things that present-day generative AI is not very good at in existing fields, and I’m not sure how easy it will be to address some of those. So, take the furry artist. It looks like she made a single digitally-painted portrait of a tiger in a suit, a character that she invented. That’s something that probably isn’t all that hard to do with present-day generative AI. But try using existing generative AI to create several different views of the same invented character, presented consistently, and that’s a weak point. That may require very deep and difficult changes on the technology front to try to address.

I don’t feel that a lot of this has been hashed out, partly because a lot of people, even in the fields, don’t have a great handle on what the weaknesses are and what might be viably remedied and how on the AI front. Would be interesting to try to do some competitions in various areas, see what a competent person in the field and someone competent in using generative AI could do. It’ll probably change over time, and techniques will evolve.

There are areas where generative AI for images has both surpassed what I expected and underperformed. I was pretty impressed with its ability to capture the elements of what creates a “mood”, say, and make an image sad or cheerful. I was very surprised at how effective current image generation models were, given their limited understanding of the world, at creating things “made out of ice”. But I was surprised at how hard it was to get any generative AI model I’ve tried to generate drawings containing crosshatching, which is something that plenty of human artists do just fine. Is it easy to address that? Maybe. I think I could give some pretty reasonable explanations as to why consistent characters are hard, but I don’t really feel like I could offer a convincing argument about why crosshatching is, don’t really understand why models do poorly with it, and thus, I’ve no idea how hard it might be to remedy that.

Some fantastic images are really easy to create with generative image AI. Some are surprisingly difficult. To name two things that I recall !imageai@sh.itjust.works regulars have run into over the past couple years, trying to create colored car treads (it looks like black treads are closely associated with the “tire” token) and trying to create centaurs (generative AI models want to do horses or people, not hybrids). The weaknesses may be easy to remedy or hard, but they won’t be the same weaknesses that humans have; these are things that are easy for a human. Ditto for strengths — it’s relatively-easy for generative AI to create extremely-detailed images (“maximalist” was a popular token that I recall seeing in many early prompts) or to replicate images of natural media that are very difficult or time-consuming to work in in the real world, and those are areas that aren’t easy for human artists.

scavenging sources have dried up completely

Last I looked — which was, admittedly, a few days ago — Google Shopping still has some retailers that have stock and haven’t yet jacked their price up. Mostly dinky little companies.

https://shopping.google.com/

searches

https://hssl.us/silicon-power-64gb-ddr5-5600-mhz-udimm-desktop-memory-kit-2-x-32gb-5600-mhz-clock-speed-sp064gblvu560fd2ad/

Never heard of these guys, can’t vouch for their reputability, but they have 2x32GB of DDR5 for $159.99 listed as being in stock.

The University of Virginia Bookstore — which I imagine is reputable — appears to be selling 2x32GB DDR5 for $245.95, and it looks like one can buy as a guest, so one presumably doesn’t need to be a University of Virginia student or staff.

https://www.uvabookstores.com/product/829320?quantity=1

EDIT: I would advise, though, that if you plan to get memory in the near future, that you’re probably better-off getting it ASAP, since I’m sure that there are going to be scalpers combing over all those small retailers before too long if they haven’t gotten to someone yet.

4chan’s position is that they aren’t doing business in the UK, which is why they’re disregarding the UK regulator’s fines. The UK regulator might be able to block them in the UK if the UK rolls out a Great Firewall of the UK, say, a la China, but probably not get the US to enforce rulings against them. And, I’d add, such a Great British Firewall is going to have limited impact unless the Brits also ban VPNs in the UK that don’t also do such blocking internal to the VPN and additionally block external VPNs, a la Russia.

In the same way, lemmy.today is doing business in the EU.

Very unlikely, in the eyes of the US court system. They have no EU physical presence, and aren’t advertising targeting EU people.

Facebook

Yeah, now they might be affected, but they’re in the EU.

EDIT: For context, last year, this happened:

https://www.nbcnews.com/news/world/russia-fines-google-20-decillion-world-gdp-youtube-kremlin-war-ukraine-rcna178172

Russia fines Google more than the world’s entire GDP

Russian courts can hand down whatever rulings they want, but they don’t really have an effect elsewhere unless other legal systems view them as having jurisdiction.

Iran has the death penalty for blasphemy. But the US isn’t going to enforce rulings on blasphemy unless it views Iran as having jurisdiction over the person posting said content.

Note that the issue only affects websites in the EU legal jurisdiction. From the US legal standpoint, that means doing business in the EU. The body text has the qualification, but the headline does not.

The Court of Justice of the EU—likely without realizing it—just completely shit the bed and made it effectively impossible to run any website in the entirety of the EU that hosts user-generated content.

“in the entirety of the EU”

That’s kinda clickbaity. This isn’t going to apply to, say, lemmy.today.

Micron is one of the “Big Three” DRAM manufacturers.

Crucial is their “sell directly to consumers” brand.

https://netvaluator.com/en/top-10-ram-manufacturers-by-market-share/

Micron Technology stands as the third giant, with a market share close to 20%, or about 23 billion USD in DRAM revenue. Unlike Samsung and SK Hynix, Micron is headquartered in the United States, making it a critical supplier for Western markets. Its product portfolio covers both DRAM and NAND, giving it broader exposure to the memory industry.

The company’s consumer-facing Crucial brand is well recognized among PC builders and gamers worldwide. Micron also plays a vital role in supplying DRAM for servers and AI, competing directly in the HBM space. Its strategy focuses on quality, diversification, and maintaining a stable supply chain for North America and Europe. As the only American giant, Micron is strategically important in the geopolitical landscape of semiconductors.

for example dell, hp and lenovo run a large business laptop leasing business if they do not get their ram, it will sour their relationships with memory manufacturers

Lenovo is stockpiling memory to try to make it through the RAM winter.

https://www.tomshardware.com/pc-components/ram/lenovo-stockpiles-ram-as-prices-skyrocket-reportedly-has-enough-inventory-to-last-through-2026-memory-stock-claimed-to-be-50-percent-higher-than-usual-to-fight-pricing-shock

Lenovo stockpiles RAM as prices skyrocket, reportedly has enough inventory to last through 2026 — memory stock claimed to be 50% higher than usual to fight pricing shock

Lenovo is playing it smart and buying up as much memory inventory as it can

I don’t think that Lenovo is getting special deals with memory makers either, or they wouldn’t need to stockpile.

Serial compute isn’t doing the double-every-18-months-in-speed since something like the early 2000s.

Unlike with serial compute, not all problems can be solved, run faster, with parallel compute. But at some point, unless we figure out some sort of new way to play with physics, we pretty much have to move to parallel compute where we can if we want much more performance.

Damn, that was the only brand of RAM without LEDs and racing stripes on it

hits Google Shopping

https://www.gamestop.com/pc-gaming/pc-components/memory/products/pny-performance-16gb-ddr5-4800mhz-desktop-memory-md16gsd54800-tb/333745.html

Aside from being a black circuit board rather than green, that PNY DIMM doesn’t look especially blinged up.

The Brits are still doing a penny!

Also, I can’t wait to buy up used RAM for pennys when the bubble pops.

The memory that manufacturers are producing is HBM; they’re transitioning facilities that had been producing memory for DIMMs to producing HBM. HBM won’t be in DIMM form factor — you can’t just stick it into the slots on a PC motherboard.

I read an article yesterday that Samsung’s memory division wasn’t even willing to let Samsung’s own cell phone division lock in any long-term memory buying agreement with them, which the cell phone division hsd been trying to do. Too much money in selling HBM memory for parallel compute to datacenters.

https://www.reuters.com/world/china/ai-frenzy-is-driving-new-global-supply-chain-crisis-2025-12-03/

Some 6,000 miles away in California, Paul Coronado said monthly sales at his company, Caramon, which sells recycled low-end memory chips pulled from decommissioned data-center servers, have surged since September. Almost all its products are now bought by Hong Kong-based intermediaries who resell them to Chinese clients, he said.

“We were doing about $500,000 a month,” he said. “Now it’s $800,000 to $900,000.”

I threw away a bunch of large-capacity DDR4 DIMMs last year, figured that they’d be useless in the future. Kind of wish I hadn’t, now. Reusing old DIMMs is probably the only source of supply that can be ramped up in the near term.

In October, SK Hynix said all its chips are sold out for 2026, while Samsung said it had secured customers for its HBM chips to be produced next year. Both firms are expanding capacity to meet AI demand, but new factories for conventional chips won’t come online until 2027 or 2028.

Two or three years until manufacturing capacity will be ramped up.

Why? I mean, they aren’t compelled to manufacture DIMMs.

Right now, there is a window in time where there are companies willing to pay tons of money for HBM, more than most people and companies are for DIMMs. It’d be crazy for memory manufacturers not to make HBM if they have the capacity to do so, if they’re doing way better by doing so.

Will these even be useful on the second hand market, or are these chips gonna be on specialized PCBs for these machines?

If I understand aright, it’s going to be HBM, so it won’t be in DIMM form. Like, can’t just go stick it in a PC.

From my /etc/resolv.conf on Debian trixie, which isn’t using openresolv:

# Third party programs should typically not access this file directly, but only
# through the symlink at /etc/resolv.conf. To manage man:resolv.conf(5) in a
# different way, replace this symlink by a static file or a different symlink.

I mean, if you want to just write a static resolv.conf, I don’t think that you normally need to have it flagged immutable. You just put the text file you want in place of the symlink.

Also, when you talk about fsck, what could be good options for this to check the drive?

I’ve never used proxmox, so I can’t advise how to do so via the UI it provides. As a general Linux approach, though, if you’re copying from a source Linux filesystem, it should be possible to unmount it — or boot from a live boot Linux CD, if that filesystem is required to run the system — and then just run fsck /dev/sda1 or whatever the filesystem device is.

I’d suspect that too. Try just reading from the source drive or just writing to the destination drive and see which causes the problems. Could also be a corrupt filesystem; probably not a bad idea to try to fsck it.

IME, on a failing disk, you can get I/O blocking as the system retries, but it usually won’t freeze the system unless your swap partition/file is on that drive. Then, as soon as the kernel goes to pull something from swap on the failing drive, everything blocks. If you have a way to view the kernel log (e.g. you’re looking at a Linux console or have serial access or something else that keeps working), you’ll probably see kernel log messages. Might try swapoff -a before doing the rsync to disable swap.

At first I was under suspicion was temperature.

I’ve never had it happen, but it is possible for heat to cause issues for hard drives; I’m assuming that OP is checking CPU temperature. If you’ve ever copied the contents of a full disk, the case will tend to get pretty toasty. I don’t know if the firmware will slow down operation to keep temperature sane — all the rotational drives I’ve used in the past have had temperature sensors, so I’d think that it would. Could try aiming a fan at the things. I doubt that that’s it, though.