Main-frame (thin) -> PC (fat) -> Internet/Cloud (thin) -> Mobile (fat) -> AI (thin)
I expect this to continue until the next technology transition.
In each of these shifts, and there have been others, things are not completely fat or thin, more of an in-between state but leaning to local vs cloud.
People who are willing to drop $20k on a computer might not be affected much tho.
They probably won't, but those willing to drop $3-10k will be if the consumer and data-center computing diverge at the architectural level. It's the classical hollowing out the middle - most of the offerings end up in a race-to-the-bottom chasing volume of price-sensitive customers, the quality options lose economies of scale and disappear, and the high-end becomes increasingly bespoke/pricey, or splits off into a distinct market with an entirely different type of customers (here: DC vs. individuals).
https://us.ugreen.com/collections/usb-c-hubs - these docks only require a single USB port to connect to. That could be a SBC working as a handheld. These docks could end up being the largest cost component in the new era of all-in-ones. UGreen could be the next Apple as screens and processors snap-on to these hubs, in addition to their own range of power banks and SSD enclosures. Their quality is high too.
In fact, I would go so far as to say we are entering a tinkering culture, and free-energy technologies are upon us as a response oppressive economic times. Sort of like how the largest leaps in religious and esoteric thought have occurred in the most oppressive of circumstances.
People will reject their crappy thin clients, start tinkering and build their own networks. Knowledge and currency will stay private and concentrated - at least at first.
But indeed, once you have USB-C support on your device, you can connect all kinds of periphery through it, from keyboards to 4K screens. Standardized device classes obviate the need for most drivers.
This will likely extend further and further, more into the "normie" territory. MS Windows is, of course, the thing that keeps many people pinned to the x64 realm, but, as Chromebooks and the Steam Deck show us, Windows is not always a hard requirement to reach a large enough market segment.
The large PC builders (Dell, HP, Lenovo) will continue down the road of cost reduction and proprietary parts. For the vast majority of people pre-packaged machines from the "big 3" are good enough. (Obviously, Apple will continue to Apple, too.)
I think bespoke commodity PCs will go the route, pricing wise, of machines like the Raptor Talos machines.
Edit: For a lot of people the fully customized bespoke PC experience is preferred. I used to be that person.
I also get why that doesn't seem like a big deal. I've been a "Dell laptop as a daily driver" user for >20 years now. My two home servers are just Dell server machines, too. I got tired of screwing around with hardware and the specs Dell provided were close enough to what I wanted.
I'm very excited about the Steam Machine for the reasons you mention - I want to buy a system, not a loose collection of parts that kind-of-sort-of implement some standard to the point that they probably work together.
We're out here with amazing performance in $600 laptops that last all day on battery and half of this comment section is acting like personal computing is over.
Raspberry Pi is way cheaper than those things, and I'm sure you could hook one up with an all-day battery for $100-200.. Doesn't mean it's "better".
They’re not ideal for all use cases, of course. I’m happy to still have my big Linux workstation under my desk. But they seem to me like personal computers in all the ways that matter.
I also want housing as cheap as it was a couple of years ago.
But I can imagine that it would become less prevalent on personal machines, maybe even rare eventually.
Prior to the crunch, you could have anything from 48-64 cores and a good chunk of RAM (128GB+). If you were inordinately lucky, 56 cores and 64GB of onboard HBM2e was doable for 900-1500 USD.
They’re not Threadrippers or EPYCs,but sort of a in between - server chip that can also make a stout workstation too.
Many a people need only a basic device for Netflix, YouTube, google docs or email or search/but flights tickets. That will be amazing.
Many have job supplied laptop/desktop for great performance (made rubbish by AV scanners but that's different issue)
I was looking up an old video game homepage the other day for some visual design guidance. It was archived on the Wayback Machine, but with Flash gone, so was the site. Ruffle can't account for every edge case.
Flash was good. It was the bedrock of a massive chunk of the Old Net. The only thing awful are the people who pushed and cheered for its demise just so that Apple could justify their walled garden for the few years before webdev caught up. Burning the British Museum to run a steam engine.
This is what I'm afraid of. As more stuff moves to the cloud helped in part by the current prices of HW, the demand for consumer hardware will drop. This will keep turning the vicious cycle of rising consumer HW prices and more moves to the cloud.
I can already see Nvidia rubbing their hands together in expectation of the massive influx of customers to their cloud gaming platform. If a GPU is so expensive, you move to a rental model and the subsequent drop in demand will make GPUs even more expensive. They're far from the only ones with dollar signs in their eyes, between the money and total control over customers this future could bring.
Being entirely reliant on someone else's software and hardware is a bleak thought for a person used to some degree of independence and self sufficiency in the tech world.
It's also a nightmare from any sort of privacy perspective, in a world that's already becoming too much like a panopticon.
Roblox is not popular because of its graphics. Younger gamers care more about having fun than having an immersive experience.
The problem I describe is companies pushing towards the "rent" model vs. "buy to own". Nvidia was just an example by virtue of their size. Microsoft could be another, they're also eying the game streaming market. Once enough buyers become renters, the buying market shrinks and becomes untenable for the rest, pushing more people to rent.
GPUs are so expensive now that many gamers were eying GeForce Now as a viable long term solution for gaming. Just recently there was a discussion on HN about GeForce Now where a lot of comments were "I can pay for 10 years of GeForce Now with the price of a 5090, and that's before counting electricity". All upsides, right?
In parallel Nvidia is probably seeing more money in the datacenter market so would rather focus the available production capacity there. Once enough gamers move away from local compute, the demand is unlikely to come back so future generations of GPUs would get more and more expensive to cater for an ever shrinking market. This is the vicious cycle. Expensive GPU + cheap cloud gaming > shrinking GPU market and higher GPU prices > more of step 1.
Roblox is one example of a game, there are many popular games that aren't graphics intensive or don't rely on eye candy. But what about all the other games that require beefy GPU to run? Gamers will want to play them, and Nvidia like most other companies sees more value in recurring revenue than in one time sales. A GPU you own won't bring Nvidia money later, a subscription keeps doing that.
The price hikes come only after there's no real alternative to renting. Look at the video streaming industry.
Also, if gamers demand infinitely improving graphics so much that they would rather pay for cloud gaming than relax their expectations and be happy with, say, current gen graphics, then that is more a claim about modern self-pwned gamer behavior than megacorp conspiracy.
But I don't buy that either. The biggest games on Steam Charts and Twitch aren't AAA RTX 5090 games.
Riddle me this: does anyone pursue a self-pwn intentionally?
"Conspiracy theory" is just dehumanizer talk for falling prey to business as usual.
This what always happens in capitalism. Scarcity is almost always followed by glut
Memory makers, for example, have sold out their inventory for several years, but instead of investing to manufacture more, they’re shutting down their consumer divisions. They’re just transferring their consumer supply to their B2B (read AI) supply instead.
Thats likely because they don’t expect this demand to last past a few years.
Many users will not want to risk their privacy, data, and workflow on someone else's rapidly-enshittifying AI cloud model. Right now we don't have much choice, but there are signs of progress.
Many new games cannot run max settings, 4k, 120hz on any modern gpus. We probably need to hit 8k before we max out on the returns higher resolution can provide. Not to mention most game devs are targeting an install base of $500 6 year old consumer hardware, in a world where the 5090 exists.
My phone has 16gigs of ram and a terabyte of storage, laptops today are ridiculous compared to anything I studied with.
I'm not arguing mind you, just trying to understand the usecases people are thinking of here.
Running Electron apps and browsing React-based websites, of course.
Perhaps the math will change if the hardware market stagnates and people are keeping computers and phones for 10 years. Perhaps it will even become a product differentiator again. Perhaps I'm delusional :).
Well, some of the "old school" has left the market of natural causes since the 2000s.
That only leaves the rest of 'em. Wer dey go, and what are your top 3 reasons for how the values of the 2000s era failed to transmit to the next generation of developers?
I have an example.
I use Logos (a Bible study app, library ecosystem, and tools) partially for my own faith and interests, and partially because I now teach an adult Sunday school class. The desktop version has gotten considerably worse over the last 2-3 years in terms of general performance, and I won't even try to run it under Wine. The mobile versions lack many of the features available for desktop, but even there, they've been plagued by weird UI bugs for both Android and iOS that seem to have been exacerbated since Faithlife switched to a subscription model. Perhaps part of it is their push to include AI-driven features, no longer prioritizing long-standing bugs, but I think it's a growing combination of company priorities and framework choices.
Oh, for simpler days, and I'm not sure I'm saying that to be curmudgeonly!
--
[0] - Having https://sectograph.com/ as a watch face is 80%+ of value of having a modern smartwatch to me. Otherwise, I wouldn't bother. I really miss Pebble.
("Why don't you just close firefox?" No thanks, I've lost tab state too many times on restart to ever trust its sessionstore. In-memory is much safer.)
You have to close Firefox every now and then for updates though. The issue you describe seems better dealt with on filesystem level with a CoW filesystem such as ZFS. That way, versioning and snapshots are a breeze, and your whole homedir could benefit.
For the rest: I agree with you.
And when, for whatever reason, having a "desktop application" becomes a priority to developers, what do they do? Write it in Electron and ship a browser engine with their app. Yuuuuuuck!
I haven't noticed any kind of difference when using Teams. That piece of crap is just as slow and borken as it always was.
> I haven't noticed any kind of difference when using Teams.
If the device is a laptop, also the thermal design (or for laptops that are in use: whether there is dust in the ventilation channels (in other words: clean the fans)) is very important for the computer to actually achieve the performance that the hardware can principally deliver.
The issue is with applications that have no business being entitled to large amount of resources. A chat app is a program that runs in the background most of the time and is used to sporadic communication. Same for music players etc. We had these sorts of things since the 90's, where high end consumer PCs hat 16mb RAM.
these days individual _tabs_ are using multiple gb of ram.
tl;dr, no one is looking for their RAM to stay idle. They're looking for their RAM to be available.
In not trying to excuse crappy developers making crappy slow ad wasteful apps, I just don't think electron itself is the problem. Nor do I think it's a particularly big deal if an app uses some memory.
The issue with Electron is that it encourages building desktop apps as self-contained websites. Sure, that makes it easier to distribute apps across systems and OSes, but it also means you've got front end web devs building system applications. Naturally, they'll use what they're used to: usually React, which exacerbates the problem. Plus it means that each app is running a new instance of a web browser, which adds overhead.
In real life, yeah, it's rare that I actually encounter a system slowdown because yet another app is running on Electron. I just think that it's bad practice to assume that all users can spare the memory.
I'll admit that my concern is more of a moral one than a practical one. I build software for a living and I think that optimizing resource usage is one way to show respect to my users (be they consumers, ops people running the infra, or whatever). Not to mention that lean, snappy apps make for a better user experience.
This is why the top model of the previous generation of the iPhone (the iPhone 16 Pro Max) has only 8 GB of RAM, bumped to 12 GB for the current top model (the iPhone 17 Pro Max at the higher tiers of additional storage). If Apple had decided to put more RAM than that into any iPhone, even the models where the price is irrelevant to most buyers, they would not have been serving their customers well.
So, now you have to pay a penalty in either battery life or device weight for the duration of your ownership of any device designed for maximum mobility if you ever want to having a good experience when running Electron apps on the device.
Are they slow because they're Electron? No idea. But you can't deny that most Electron apps are sluggish for no clear reason. At least if they were pegging a CPU, you'd figure your box is slow. But that's not even what happens. Maybe they would've been sluggish even using native frameworks. Teams seems to do 1M network round-trips on each action, so even if it was perfectly optimized assembly for my specific CPU it would probably make no difference.
I wonder if there’s a computer science law about this. This could be my chance!
https://en.wikipedia.org/wiki/Wirth%27s_law
Not exactly the same (it's about power rather than price). But close enough that when you said it, I thought, "oh! there is something like that." There's also more fundamental economics laws at play for supply and demand of a resource / efficiencies at scale / etc. Given our ever increasing demand of compute compared increasing supply (cheaper more powerful compute), I expect the supply will bottleneck before the demand does.
I guess this might be happening with LLMs already
The constant increases in website and electron app weight don't feel great either.
A lot of commercial CAD software exists for a very long time, and it is important for industrial customers that the backward compatibility is very well kept. So, the vendors don't want to do deep changes in the CAD kernels.
Additionally, such developments are expensive (because novel algorithms have to be invented). I guess CAD applications are not that incredibly profitable that as a vendor you want to invest a huge amount of money into the development of such a feature.
For word processing, basic image manipulation, electron app (well...) even the "cheap" Macbook Neo is good enough, and it's a last year phone CPU. But that's not enough for a lot of use case.
Most affordable laptops have exactly that, 16gigs of ram and a terabyte of storage. Think about THAT!
That's "non powerful" to you?
This absolutely boggles my mind. Do you mind if I ask what type of computing you do in order to justify this purchase to yourself?
I'm also into motorcycles. Before I owned a house with a garage, I had to continuously pack my tools up and unpack them the next day. A bigger project meant schlepping parts in and out of the house. I had to keep track of the weather to work on my bikes.
Then, when I got a house, I made sure to get one with a garage and power. It transformed my experience. I was able to leave projects in situ until I had time. I had a place to put all my tools.
The workstation is a lot like that. The alternative would be renting. But then I'd spend a lot of my time schlepping data back and forth, investing in setting things up and tearing them down.
YMMV. I wouldn't dream of trying to universalize my experience.
I would bet it continues to be more affordable to buy reasonable specs with current consumer hardware, rather than buying a top system once.
768GB of RAM is insane…
Meanwhile, I’ve been going back and forth for over a year about spending $10k on a MacBook Pro with 128GB. I can’t shake the feeling I’d never actually use that much, and that, long term, cloud compute is going to matter more than sinking money into a single, non-upgradable machine anyway.
I don't know your workloads, but for me personally 64 GB is the ceiling buffer on RAM - I can run entire k8s cluster locally with that and the M5 Pro with top cores is same CPU as M5 Max. I don't need the GPU - the local AI story and OSS models are just a toy for my use-cases and I'm always going to shell out for the API/frontier capabilities. I'm even thinking of 48 config because they already have those on 8% discounts/shipped by Amazon and I never hit that even on my workstation with 64 GB.
No, it won't. The power drain of merely refreshing DRAM is negligible, it's no higher than the drain you'd see in S3 standby over the same time period.
Similarly, having more cache may mean less SSD activity, which may mean less energy draw overall.
If I had a chip to put on the roulette table of this "what if" I'd put it on the "it won't make a difference in the real world in any meaningful way" square.
It wasn't my primary motivator but it hasn't made me regret my decision.
I hummed and hawed on it for a good few months myself.
How is this going to work? You need uncontrolled compute for developing software. Any country locking up that ability too much will lose to those who don't.
I've read about companies where all software developers have to RDP to the company's servers to develop software, either to save on costs (sharing a few powerful servers with plenty of RAM and CPU between several developers) or to protect against leaks (since the code and assets never leave the company's Citrix servers).
Oh you sweet summer child :(
You think our best and brightest aren't already working on that problem?
In fact they've fucking aced it, as has been widely celebrated on this website for years at this point.
All that remains is getting the rest of the world to buy in, hahahaha.
But I laugh unfairly and bitterly; getting people to buy in is in fact easiest.
Just put 'em in the pincer of attention/surveillance economy (make desire mandatory again!).
And then offer their ravaged intellectual and emotional lives the barest semblance of meaning, of progress, of the self-evident truth of reason.
And magic happens.
---
To digress. What you said is not unlike "you need uncontrolled thought for (writing books/recording music/shooting movies/etc)".
That's a sweet sentiment, innit?
Except it's being disproved daily by several global slop-publishing industries that exist since before personal computing.
Making a blockbuster movie, recording a pop hit, or publishing the kind of book you can buy at an airport, all employ millions of people; including many who seem to do nothing particularly comprehensible besides knowing people who know people... It reminds me of the Chinese Brain experiment a great deal.
Incidentally, those industries taught you most of what you know about "how to human"; their products were also a staple in the lives of your parents; and your grandparents... if you're the average bougieprole, anyway.
---
Anyway, what do you think the purpose of LLMs even is?
What's the supposed endgame of this entire coordinated push to stop instructing the computer (with all the "superhuman" exactitude this requires); and instead begin to "build" software by asking nicely?
Btw, no matter how hard we ignore some things, what's happening does not pertain only to software; also affected are prose, sound, video, basically all electronic media... permit yourself your one unfounded generalization for the day, and tell me - do you begin to get where this is going?
Not "compute" (the industrial resource) but computing (the individual activity) is politically sensitive: programming is a hands-on course in epistemics; and epistemics, in turn, teaches fearless disobedience.
There's a lot of money riding on fearless disobedience remaining a niche hobby. And if there's more money riding on anything else in the world right now, I'd like an accredited source to tell me what the hell that would be.
Think for two fucking seconds and once you're done screaming come join the resistance.
Before this price spike, it used to be you could get a second-hand rack server with 1TB of DDR4 for about $1000-2000. People were massively underestimating the performance of reasonably priced server hardware.
You can still get that, of course, but it costs a lot more. The recycling company I know is now taking the RAM out of every server and selling it separately.
Apple hardware is incredibly overpriced.
See a $1100 GPU on eBay, but it’s in the US? Actually a $1900 GPU.
A colleague were just talking about how well he timed the purchase of his $700 24GB 3090.
As someone who just bought a completely maxed out 14" Macbook Pro with an M5 Max and 128GB of RAM and 8TB SSD, it was not $10k, it was only a bit over $7k. Where is this extra $3k going?
Turns out the heatsink in the 14" isn't nearly enough to handle the Max with all cores pegged. I'd get about 30 seconds of full power before frequency would drop like a rock.
I'm wondering if there was something wrong with your particular unit?
How can you say this when Apple is releasing extremely fast M5 MacBook Pros? Or the $600 MacBook Neo that has incredible performance for that price point?
Even x86 is getting some interesting options. The Strix Halo platform has become popular with LLM users that the parts are being sold in high numbers for little desktop systems.
If you haven't tried out a desktop CPU in a while, I highly recommend you giving it a try if you're used to only using laptops, even when in the same class the difference is obvious.
For CPU-bound tasks like compiling they’re not that different. For GPU tasks my desktop wins by far but it also consumes many times more power to run the giant GPU.
If you think laptops are behind consumer desktops for normal tasks like compiling code you probably haven’t used a recent MacBook Pro.
What are the exact CPU models used here though? Since my point was about CPUs in the "same class", and it's really hard to see if this is actually the case here.
And yes, I've played around with the recent Apple CPUs, all the way up to M4 Pro (I think, thinking about it I'm not 100% sure) and still I'd say the same class of CPUs will do better in a desktop rather than a laptop.
If you want to compare it in the Apple ecosystem, compare the CPUs of a laptop to one of the Mac Mini/Mac Studio, and I'm sure you'll still see a difference, albeit maybe smaller than other brands.
The same chip perform basically the same in the different form factors.
For all of the definitive statements you're making in this thread, you don't seem to know much about Apple M-series silicon.
A 300W GPU released in 2025 is about 10x M5 perf. The difference is going to be smaller for CPU perf, but also not close.
This is not true. The recent MacBook Pros are every bit as fast as my Zen 5 desktop for most tasks like compiling.
For GPU there is a difference because both are constrained by thermal and power requirements where the desktop has a big advantage.
For CPU compute, the laptop can actually be faster for single threaded work and comparable for multi threaded work.
Anyone claiming laptop CPUs can’t keep up with desktop CPUs hasn’t been paying attention. The latest laptops are amazing.
Bad example. That's highly parallel, so a higher core-count die is going to destroy the base M5 here.
I don't typically compile Linux on my M5, so I don't really care, but at least online available clang benchmarks put it at roughly half the LOC/s of a 9950X, which released in 2024.
Anything single threaded it should match or even edge ahead though.
It gets for worse for multi threaded perf if you leave behind consumer-grade hardware and compare professional/workhorse level CPUs like EPYC/Threadripper/Xeon to Apple's "pro" lines. That's just a slaughter. They're roughly 3x a 9950X die for these kinds of workloads.
The base M5 starts at 10 cores and scales to 18 cores. The performance is similar to high end dekstop consumer CPUs.
> I don't typically compile Linux on my M5, so I don't really care,
If you don't compile large codebases, why do you care then?
I do compile large codebases and I'm speaking from experience with the same codebase on both platforms. Not "LOC/s" benchmarks
There's a large C++ codebase I need to compile, but it can't compile/run on OSX in the first place, hence the desktop that I use remotely for that. Since it's also kind of a shitshow, that one has really terrible compile times: up to 15 minutes on a high powered Intel ThinkPad I no longer use, ~2 minutes on desktop.
I could do it in a VM as well, but let's be real: running it on the M5 in front of me is going to be nowhere near as nice as running it on the water cooled desktop under my desk.
And then, if your laptop is busy, your machine is occupied - I hate that feeling. I never run heavy software on my laptop. My machine is in the cellar, I connect over ssh. My desktop and my laptop are different machines. I don't want to have to keep my laptop open and running. And I don't want to drag an expensive piece of hardware everywhere.
And then you need to use macOS. I'm not a macOS person.
I would hope so, given that you can buy multiple M5 laptops for the price of that CPU alone.
I made a comment about how impressive the M5 laptops were above, so these comments trying to debunk it by comparing to $12,000 CPUs (before building the rest of the system) are kind of an admission that the M5 is rather powerful. If you have to spend 3-4X as much to build something that competes, what are we even talking about any more?
(Of course, I don't disagree with the notion that consumerism produces an extraordinary amount of worthless trash, but that's a different matter. The main problem with consumerism is consumerism itself as a spiritual disease; the material devastation is a consequence of that.)
The planet has a certain resource-bound carrying capacity. It's a fact of physics. Just because we aren't there yet as of (checks time) 2026-03-27, doesn't mean Malthusians are wrong.
Although to be fair to the other side, I think with abundant renewable energy we'll be able to delay resource depletion for a very long time thanks to recycling (and lower standards of living of course).
Also C: nations that are both A and B, needlessly causing oil volatility with unplanned military dickheadedness.
And even if we figured out how to electrify everything (which we didn't as I just said), we would still run into resources shortages for batteries, wires (copper etc.), nuclear fuel (uranium)...
There is not a risk of resource shortage of copper. The doomer and prepper talking points you're parroting are not based in reality.
And I don't even understand your other points to be honest. What do you mean "consumer vehicles" ? Are you taking about individual's cars ? I'm not taking about that, these don't matter that much. I'm taking about trucks, boats, planes, the stuff actually shipping you your lifestyle.
Look up what it means to have a conversation in "good faith" vs in "bad faith" and you might learn something useful about conversation tools. For example, lying about what someone says and calling it "peculiar" is "bad faith".
This is where I think current hackers should be headed. I grew up with lots of family who were backyard mechanics, wrenching on cars and motorcycles. Their investment in tools made my occasional PC purchase look extremely affordable. Based on what I read, senior mechanics often have five-figure US dollar investments in tools. Of course, I guess high quality torque wrenches probably outlast current GPU chips? I'd hate to be stuck making a $10K investment every 24 months on a new GPU . . .
I have been renting GPU resources and running open weight models, but recently my preferred provider simply doesn't have hardware available. I'm now kicking myself a little for not simply making a big purchase last fall when prices were better.
I've replaced transmissions, head gaskets, and done all work for our family cars for two decades based on a Costco toolkit, and 20 trips to the autoparts store or Walmart when I needed something to help out.
Maybe I'm being a little forgetful that yes I bought a jack, and Jack stands, and have a random pipe as a breaker bar, and other odds and ends. But you can go very far for $1k as a DIYer.
It really feels like we're slowly marching back to the era of mainframe computers and dumb terminals. Maybe the democratization of hardware was a temporary aberration.
Also, I wonder how many of us, even here on HN, have the ability to spend that amount of money on computer for personal use. Frankly I wouldn't even know what to do with all the RAM - should I just ramdisk every program I use and every digital thing I made in the last five years?
Anyhow, I suppose for the folks who can't afford hardware (perhaps by design), one ought to own nothing and be happy.
The RAM choice was because I have never regretted buying more RAM - it's practically always a better trade than a slightly faster CPU - and 96GB DIMMs were at a sweet spot compared to 128GB DIMMs.
That, and the ability to have big LLMs in memory, for some local inference, even if it's slow mixed CPU/GPU inference, or paged on demand. And if not for big LLMs, then to keep models cached for quick swapping.
I don't mean to judge, it's your money but to me it seems like an enormous waste. Just like spending $100k on a car when you can get one for $15k that does pretty much exactly the same job.
It's not so easy to get nice second-hand hardware here in Switzerland, and my HEDT is nice and quiet, doesn't need to be rack-mounted, plugs straight into the wall. I keep it in the basement next to the internet router anyway.
The "sensible" choice is to rent. It's the same with cars; most people these days lease (about 50% of new cars in CH, which will be a majority if you compare it with auto loan and cash purchase).
I think leasing might be okayish if you find a really good deal, but it's really not much different than buying new which is just a shit deal no matter how you turn it. A 1-4 year old car is pretty much new anyway, I don't see any reason to buy brand new.
That’s for everyone
Never really used it all, usually only about 40%, but it's one of those better to have than not need, and better than selling and re-buying a larger memory machine (when it's something you can't upgrade, like a Mac or certain other laptops)
We live in world where we optimised for globalization. Industry in china, oil in middle east, etc...
This approach proved to be fragile on the hands of people with money and/or power enough to tilt the scale
And I fear they will be equally confused and annoyed by disposing of all of them.
I thought the trend is the opposite direction, with RTX 5x series converging with server atchitectures (Blackwell-based such as RTX 6000 Pro+). Just less VRAM and fewer tensor cores, artificially.
Where is the divergence happening? Or you don't view RTX 5x as consumer hardware?
Tech feels increasingly fragile with more and more consolidation. We have a huge chunk of advanced chip manufacturing situated on a tiny island off the coast of a rising superpower that hates that island being independent. Fabs in general are so expensive that you need a huge market to justify building one. That market is there, for now. But it doesn't seem like there's much redundancy. If there's an economic shock, like, I dunno, 20% of the world's oil supply suddenly being blockaded, I worry that could tip things into a death spiral instead.
Are you kidding? Apple's mobile chips are now delivering perf that AMD & intel desktop never could or did.
Most applications don't make aggressive use of the SIMD instructions that modern x86 chips offer, thus you get this impression. :-(
What are you talking about?
My laptops are, and always have been, primarily places where I do local computing. I write code there, I watch movies there, I listen to music there, I play games there...all with local storage, local compute, and local control (though I do also store a bunch of my movies on a personal media server, housed in my TV stand, because it can hold a lot more). My smartphone is similar.
If you think that the vast majority of the work most people do on their personal computers is moving to LLMs, or cloud gaming, then I think you are operating in a pretty serious bubble. 99.9% of all work that most people do is still best done locally: word processing, spreadsheets, email, writing code, etc. Even in the cases where the application is hosted online (like Google Docs/Sheets), the compute is still primarily local.
The closest to what you're describing that I think makes any sense is the proliferation of streaming media—but again, while they store the vast libraries of content for us, the decoding is done locally, after the content has reached our devices.
It doesn't matter if a cutting-edge AI-optimized server can perform 10, 100, or 1000 times better than my laptop at any particular task: if the speed at which my laptop performs it is faster than I, as a human, can keep up (whatever that means for the particular task), then there's no reason not to do the task locally.
I don't share the same 1:1 opinion with regards to the article, but it is absolutely clear that RAM prices have gone up enormously. Just compare them. That is fact.
It may be cheaper lateron, but ... when will that happen? Is there a guarantee? Supply crunch can also mean that fewer people can afford something because the prices are now much higher than before. Add to this the oil crisis Trump started and we are now suddenly having to pay more just because a few mafiosi benefit from this. (See Krugman's analysis of the recent stock market flow of money/stocks.)
Open source efforts need to give up on local AI and embrace cloud compute.
We need to stop building toy models to run on RTX and instead try to compete with the hyperscalers. We need open weights models that are big and run on H200s. Those are the class of models that will be able to compete.
When the hyperscalers reach take off, we're done for. If we can stay within ~6months, we might be able to slow them down or even break them.
If there was something 80-90% as good as Opus or Seedance or Nano Banana, more of the ecosystem would switch to open source because it offers control and sovereignty. But we don't have that right now.
If we had really competitive open weights models, universities, research teams, other labs, and other companies would be able to collaboratively contribute to the effort.
Everyone in the open source world is trying to shrink these models to fit on their 3090 instead, though, and that's such a wasted effort. It's short term thinking.
An "OpenRunPod/OpenOpenRouter" + one click deploy of models just as good as Gemini will win over LMStudio and ComfyUI trying to hack a solution on your own Nvidia gaming card.
That's such a tiny segment of the market, and the tools are all horrible to use anyway. It's like we learned nothing from "The Year of Linux on Desktop 1999". Only when we realized the data center was our friend did we frame our open source effort appropriately.
We have this class of models already, Kimi 2.5 and GLM-5 are proper SOTA models. Nemotron might also release a larger-sized model at some time in the future. With the new NVMe-based offload being worked on as of late you can even experiment with these models on your own hardware, but of course there's plenty of cheap third-party inference platforms for these too.
Oh god no, please not more slop, you're already consuming over 1 percent of human energy output, could you, like, chill a bit?
I.e., /if/ I am going to consume LLM tokens, I figure that a local LLM with 10s of billions of parameters running on commodity hardware at home will still consume far more energy per token than that of a frontier model running on commercial hardware which is very strongly incentivized to be as efficient as possible. Do the math; it isn't even close. (Maybe it'd be closer in your local winter, where your compute heat could offset your heating requirements. But that gets harder to quantify.)
Maybe it's different if you have insane and modern local hardware, but at least in my situation that is not the case.
^ Fair. Yep, I agree the calculus changes if you don't have _any_ local hardware and you're needing to factor in the cost of acquiring such hardware.
When I did this napkin math, I was mostly interested in the energy aspect, using cost as a proxy. I was calculating the $/token (taking into consideration the cost of a KWh from my utility, the measured power draw of my M1 work machine, and the measured tokens per second processed by a ~20BP open-weight model). I then compared this to the published $/token rate of a frontier provider, and it was something like two orders of magnitude in favor of the frontier model. I get it, they're subsidizing, but I've got to imagine there's some truth in the numbers.
I wonder, does (or will) the $/token ratio fall asymptotically toward the cost of electricity? In my mind I'm drawing a parallel to how the value of mined cryptocurrency approximately tracks the cost of electricity... but I might be misremembering that detail.
- Our career is reaching the end of the line
- 99.9999% of users will be using the cloud
- if we don't have strong open source models, we're going to be locked into hyperscaler APIs for life
- piddly little home GPUs don't do squat against this
Why are you building for hobby uses?
Build for freedom of the ability to make and scale businesses. To remain competitive. To have options in the future independent of hyperscalers.
We're going to be locked out of the game soon.
Everyone should be panicking about losing the ability to participate.
Play with your RTXes all you like. They might as well be raspberry pis. They're toys.
Our future depends on our ability to run and access large scale, competitive, open weights. Not stuff you run with LM Studio or ComfyUI as a hobby.
Also, the only thing crashing down will be the economic participation of everyday people if we don't have ownership over the means of creation. Hyperscalers will be just fine.
Here's my retort: https://news.ycombinator.com/item?id=47543367
People laugh at young men for looksmaxxing. And then there’s this. I dunno. As someone who has been playing computer games since the 70s, I clearly do not understand the culture anymore. But what forces would drive a young man to spend the price of a used car to play a derivative FPS? It seems heartbreaking. Just like the looksmaxxer.