Bro, that was intense. I appreciate you ❤️

I remember playing ASCII games as a kid, I gotta get back into it. There was a program or website where one could download community created ASCII games and I remember building one or two.

Ha, that's the funny part. They could have kept pumping out AM4 and taken market share slowly over time while making normal money, making normal jobs, but noooooo line must go up every quarter. So we instead got new products that are twice as much money (when ddr5 came out) that are less stable and not much better, and then to make line go up more fuck selling to people lets all do AI bullshit instead. And now we are in a world where now there is no current viable product, less jobs, and less money people are willing to spend on things that cost 5X what they should.

I sat down to log into a couple MUDs yesterday. Didn't stick with it --- the combat still hasn't evolved enough for me --- but you could play that on anything capable of displaying text on a screen and running telnet.

It looks like some crazy person has been doing a TCP/IP stack for the 128K Mac, the first Macintosh ever released, from 1984, as well as a telnet client. So you can technically lug a 42-year-old computer out of an antique store and play currently-being-developed Internet games...though you won't be getting color, since that came a while down the road.

If you get some device that can expose a serial console on some system to TCP/IP --- not sure how far back you need to go for that --- you could technically play it on a teletype from the 1930s.

The "some device" will have to be later, though, so that's maybe kinda cheaty.

Technically, Debian Linux has been run on an Intel CPU from 1971, but it isn't fast enough to be a practical host for such a teletype in that environment. Even stripped down forms of Linux are going to be "too big" to be such a host.

It does sound like the Commodore 64 has a package, Novaterm 10, that runs TCP/IP and telnet, but I don't know whether it can output to the C64's serial port rather than video display; you could play locally on one of those, but probably not run on a teletype. That being said, it probably shows that it's technically-possible, since I'm sure that if it can run a virtual terminal program, it has the resources to just dump the text to an actual terminal via a serial port. I'd guess that there's probably some system out there circa 1980 that someone has probably built that can both run a TCP/IP stack and expose a serial console to a 1930s teletype to play current Internet games.

Doom Rogue Like.

Not sure if Jupiter Hell(successor by the creators) will run on dos though.

Youre complaining that they attempt to advance technology to acquire marketshare?

I'd agree that its an AI bubble, I can understand that at least. But those peoples money will be set on fire far more than you with a higher DDR price, as it was with the dotcom bubble; then you'll get cheaper DDR.

I just did the same thing 100 bucks 32gb ram I had lying around, still very cheap. Managed 4.7 GHz overclock with my existing 8350 cooler. It's the best CPU with DDR3 though I hear there are comparable xeons.

CO on Zen 3 X3D chips is always an undervolt, not an overclock.

If it was server hardware they probably needed more than just ECC. They were probably RDIMM or LRDIMM which have built in registers to handle addressing larger amounts of memory.

I've been playing a lot of Doom recently and I used to run that on an old 486. People just kept on making free wads for that for like 35 years.

checks

The 5700X3D looks like it goes for about $350 on eBay. The 5800X3D looks like about $450.

I don't think there will be a DDR6. I think the AI bubble is going to pop & all these data centers will become "mainframe centers" that your minimum spec'd home terminal connects to to do all the computing for you on "Our lightning fast multi-core super computer with terabytes of memory!"

😐😭🤮

Sticking to DDR3 until society finally collapses , I am also building a nice ddr2 build for super cheap that still runs all my physical games collection!

My DDR3 mini itx board already has the max capacity installed years ago, which was 16gb. It's reached the ceiling :(

Just change the mobo man but stick to ddr3. I am gonna die on this hill man, computing died in 2013

Youre complaining that they attempt to advance technology to acquire marketshare?

Yes, it is a colossal industry wide issue of "needing" a new product in a rapidly decreasing window regardless of the utility of the product.

noo, good cats are not for sale

I doubt it. Those AI computers are built in a really weird way and have a lot of hardware that isn't really useful outside an AI/HPC context. Some stuff like the weird card to card network topology can be reconfigured but the rest of it can't easily be. The servers are rather agressively designed around keeping as many GPUs fed as possible making them kinda weird for other jobs. Those datacenter cards are missing enough video hardware (for example texture units) to make gaming hard and I'm not sure there's that much consumer demand for linear algebra accelerators. If they can't find more HPC jobs they may go under. Movie studios could have interesting opportunities here but they are still primarily using CPUs in all their software IIRC.

The clusters in the UAE and Saudi Arabia might be repurposable for nuclear weapons research which isn't great.

I'm running an AM4 Ryzen 9, with 64Gb of DDR4, and the thing cooks. AM4 is just great.

This is universally true, so, it’s implied they are a bad cat unfortunately

I’m just getting into my first MUD! It’s neat!

I've been happily sitting on my R5 2600x for so many years now lol. My MB is an MSI B450 Pro Carbon and it can easily handle a 5950x or 5800x3D or whatever.

Add another 2x8 gb ddr4 for 32GB total and I'll be seeing some insane performance relative to what I'm already happy with, which will last me another 5 years easily unless the mobo dies.

It's really insane how well designed AM4 is.

Nice, Beyond Sunset kinda reminds me of Neon Overdrive, which came out last year. I wonder if NO took aesthetic inspiration from the BS setting. I haven't played through a ton of the total conversion wads but it's always impressive what people manage to do with the (tweaked) Build engine. Are those two standouts for you? Looks like Ashes got a cacoward runner up in 2018. I had a lot of fun recently with one called Impping, which is a slaughter wad where the enemies are mostly imps and your ammo is mostly rockets.

https://doomwiki.org/wiki/Neon_Overdrive
https://doomwiki.org/wiki/Impping!

Oh, and apparently there was a bit of drama a few months back over the GZDoom guy using some AI in his coding so some people switched to the UZDoom fork

https://lemmy.zip/post/54431428

I have love for this machine. It has a 6W TDP CPU and 16GB LPDDR3. Unfortunately the CPU is soldered to the board. Recently added a PCI expansion for 1 extra SATA slot for an extra disk. Works great, but it's pretty much maxed out now. Gonna hold onto this for as long as it goes.

Chiplet designs and integrated GPU do benefit a lot from ram speed at least. So its speeding up CPU and GPU as well, providing a higher CPU core count and cheaper CPU.

GPGPU usage is probally going to see some real usage. There was an interesting talk at the xorg conf even about turn the video hardware into virtual services running on GPGPU focused hardware.

Ive talked with some of the HPC programers too who are trying to find creative repurposes already lol

"Rescue"
Lol. What is the rescue if even DDR4 explodes? What does this help?

This is cool. I just hope game developers also get on this bandwagon. We could use a 4-year moratorium on increasing minimum system requirements.

Unless he's some kind of purebred, you probably won't even get enough for one stick.

DDR4 is already ridiculously expensive, just not quite as bad as DDR5

I still have some DDR3, but no motherboard for it. We need DDR4, but I don't suppose anyone will want to trade in that direction.

I think that it's fair to say that AI is not the only application for that hardware, but I also think that carpelbridgesyndrome's point was that they aren't really well-suited to replace conventional servers, where all local computing just moves to a server, which is the sort of thing that ouRKaoS was worried about. Maybe for some very specialized use cases, like cloud gaming in some genres. I'd also add that the physical buildings have way more cooling capacity than is necessary for conventional servers, so they probably wouldn't be the most-cost-effective approach even if you replaced the computing hardware in the buildings.

I got my 5700X3D for 280 about this time last year, so that's actually not an awful price increase.

not gonna happen with GTA6 releasing soon

Rockstar are good at optimising their games. GTA V ran better than GTA IV on the same hardware. I wouldn't be surprised if GTA VI runs well on the Steam Deck.

I'm not sure that memory --- and I'm speaking more-broadly than HBM, even --- optimized for running neural nets on parallel compute hardware and memory optimized for conventional CPUs overlap all that much. I think that, setting aside the HBM question, that if we long-term wind up with dedicated parallel compute hardware running neural nets, that we may very well wind up with different sorts of memory optimized for different things.

So, if you're running neural nets, you have extremely predictable access patterns. Software could tell you what its next 10 GB of accesses to the neural net are going to be. That means that latency is basically a total non-factor for neural net memory, because the software can request it in huge batches and do other things in the meantime.

That's not the case for a lot of the memory used for, say, playing video games. Part of the reason, aside from hardware vendors using price discrimination that PCs (as opposed to servers) don't use registered memory (which makes it easier to handle more memory) is because it increases latency a little bit, which is bad when you're running software where you don't know what memory you're going to need next and have a critical path that relies on that memory.

On the other hand, parallel compute hardware doing neural nets are extremely sensitive to bandwidth. They want as much as they can possibly get, and that's where the bottleneck is today for them. Back on your home computer, a lot of software is oriented around doing operations in serial, and that's more prone to not saturate the memory bus.

I'd bet that neural net parallel compute hardware does way more reading than writing of memory, because edge weights don't change at runtime (on current models! That could change!).

searches

Yeah.

https://arxiv.org/html/2501.09605v1

AI clusters today are one of the major uses of High Bandwidth Memory (HBM). However, HBM is suboptimal for AI workloads for several reasons. Analysis shows HBM is overprovisioned on write performance, but underprovisioned on density and read bandwidth, and also has significant energy per bit overheads. It is also expensive, with lower yield than DRAM due to manufacturing complexity.

But there are probably a lot of workloads where your CPU wants to do a ton of writes.

I'd bet that cache coherency isn't a huge issue for neural net parallel compute hardware, because it's going to be a while until any value computed by one part of the hardware is needed again, until we reach the point where we can parallel-compute an entire layer at one go (which...I suppose we could theoretically do. Someone just posted something that I commented on about someone making an ASIC with Llama edge weights hard-coded into the silicon, which probably is a step in that direction). But with CPUs, a big problem is making sure that a value written by one CPU core reaches another CPU core, that the second doesn't use a stale value. That's gonna impact the kind of memory controller design that's optimal.

GTA IV was released for the same generation of consoles, and also there was only a 5 year difference between releases.

GTA V came out in 2013... almost 13 years ago and we are two generations ahead console-wise.

I'm just skeptical at the idea that GTA VI will run well on Steam Deck, maybe on the future Steam Machine

Well, it did take a while for GTA IV to run well on anything, really...

Well shit. I've been going off the now long-outdated memory, pun partially intended, of HBM2e being positioned as a potential replacement for DDR. Maybe we'll see more manufacturers doing something like Apple and jamming the RAM into the CPU die directly. That seems to be working out well for them so far, at least performance-wise.

i think its a bit much to still ask the steam deck to run new releases well, especially with valves refusal to release performance updated version unless they can make major, foundational improvements such that it would basically be a steam deck 2

the steam deck is 4 years old, afterall, and it wasnt exactly a top end machine when it released.. being little more than a slightly modified APU based system.

it still doesn't in some modern PCs lol

My understanding is that the AI companies push their servers so hard that the components are basically consumables. Consumers don’t really press their machines to the point of physical exhaustion.

Is it?

I feel like I could have raised a family in the time between its supposed release date and now, and it's still not out yet.

Doesn't GTA IV famously soft lock if the frame rates too high because soon aspects of the animation cycle are blocked to frame rate?

You can't run normal programs on their weird AI architecture. This is the problem everyone has with all of the ram as well, when the AI bubble pops we won't get loads of cheap RAM because it's all configured for AI and doesn't really work on anything else. They can't just pivot, that's why they're so eager to make AI a thing.

And no where near as much capacity. It's a pretty terrible alternative.

I'm sure you can't do it easily, but I'm sure there will be ridiculous AI vibe-coded attempts at making it work that end in a catastrophic failure/data breach/scandal.

My understanding is that the RAM architecture is built around insanely quick read write access but doesn't really store data for more than three or four seconds at a time. Most modern programs expect the RAM to hold on to the data for basically however long they need until they access it. So most programs just won't fit into memory configured like that, and I think it's a hardware thing, not something you can change with software.