Posted by runako 16 hours ago
That trend of the last several decades has been completely destroyed in the last year or so. Sure you can buy a new computer that might be slightly better than one a few years old, but it will be significantly more expensive to the point where an upgrade is almost certainly not worth it unless money is no object. Even replacing a computer from the pre-AI era would be virtually impossible at anywhere near what it cost you to buy it new unless it was verging on truly ancient by computer standards. Computers have always been a counterpoint to inflation everywhere else, but now that seems to be not only no longer true, but computer costs are inflating significantly worse than other goods.
I find myself increasingly in favor of the movement against data centers. Not because I think data centers are bad on their own, but because of the hording of computing resources they represent and the impact that has on people trying to buy any sort of computing device for their personal use. The PC I built 2 years ago is basically irreplaceable at anywhere near the price I paid for it. That's a truly unacceptable state of affairs in personal computing.
The industry has little incentive to fix this problem and as the market shrinks so does the incentive.
We lived thru the PC revolution. It looks over now to me.
Whether OEMs like Apple pass these savings along to consumers is tbd though.
And then the opposite can happen like an unknown drone with 100 kg warhead striking memory production facility.
Even when the cost of the cheapest computers bottomed out, it likely had more to do with lower prices not making sense. At a certain point, the cost of things that encapsulate the computer and I/O (like the enclosure and screen) end up overwhelming the cost of the actual compute.
This is capitalism, as demand increases, so does price, regardless of the cogs.
TSMC, plus flash and memory manufacturers with their own facilities, have dramatically cut production of almost anything that isn't in demand for AI, plus Macs were just about the only way to get more than 16GB of 'VRAM' without spending multiple thousands of dollars, so Mac Minis and Studios with more than about 24GB of ram have been under months-long backorders, which is unprecedented. Tim Apple made is name in supply chain logistics. Even he can't logistic his way out of this. Making things worse, the RAM that Macs use is really fast stuff, nearly the speed of high end desktop GPUs.
Supply for anything that isn't exactly what NVIDIA, Meta, OpenAI, Anthropic, and MS want has pretty much evaporated. And even that stuff has skyrocketed in cost because, essentially, the race has turned into who can hoard the most of TSMC's and NVIDIA's production capacity in a particular year.
The pathetic thing is that it isn't even actual purchases in some cases. In some cases, it's OpenAI or Anthropic merely "promising" they will buy X% of a memory manufacturer's capacity.
I can't wait for all of this to come crashing down.For all this hardware that is useless for anything else because of its density and power consumption - to absolutely crater in price and become e-waste.
What will happen once people are priced out of buying computers? I find this worrying.
We've been through.. 4? 5? memory boom-bust-boom cycles already. Nothing is pointing to this one being any different, aside it probably becoming the mother of all busts since AI isn't adding much productivity and as businesses realize this, they'll stop paying top dollar for access meaning AI vendors cancel all their reserved fab capacity.
The fabs already exist at that point so it's less of a loss for the memory fabs to crank out memory chips and have the RAM OEMs sell them at bargain bin prices. No pseudo-cartel pricing possible either with CMXT becoming a major player. $50 for 32GB-64GB of DDR5 in 2029 here we come babyy
no one remembers the last time things got better rather than worse. shitcoin-driven scarcity seamlessly transitioned into COVID-driven scarcity and now into hyperscaler-driven scarcity. the chipmakers' cabal decided to meet increased demand with four digit margins rather than expanding production.
It's Deja vu for NVMe and SSD drives this time as prices are way up there.
Then they sold out and most recently literally sold out to this software company from my neck of the woods. Every single article now feels like it's written by an LLM optimized to write like a crypto/AI bro.
I used to use it daily and now not at all.
Would love to know where this comes from. I am not sure about that. Could you point to data/anecdata?
There's been a few studies. Most find productivity gain from 'AI' to be negligible.
This goes back before AI, with computers not being found to be improving productivity numbers: Solow's paradox.
* https://en.wikipedia.org/wiki/Productivity_paradox
* https://www.brookings.edu/articles/the-solow-productivity-pa...
One of the problems here is that things like this only count as "increased productivity" in the economic sense if it results in more income at the end of the funnel. Are you actually making more product faster in all your new-found free time?
Very few orgs I've worked in have actually been limited by the speed of programmers writing code. Usually it's communication overhead and management processes eating up most of the bandwidth - and those tend to expand to absorb any additional bandwidth.
Look at all the shit github README before llm. Most repos didn't bother because it was effort. those few that did, were because of autistic focus. Of those, very few were "great" user documentation.
now llms are here, the effort barrier is gone, and now we're inundated with shit README that provide no value to users nor respect their time.
Once you start applying SKILL.md that revolve around fixing HOW to communicate value to users in a way that respects their time, you'll see this story change in noticeable places.
I doubt it will change everywhere because it also requires that you give a shit about doing this in the first place, so those people Ed Zitron talks about... business idiots won't care, because they're ultimately nihlist.
This strikes me as a problem with how we define productivity in this context. If I can do my job in 1 hour, when it took me 8 before, am I not more productive in a rate-based sense? Sure, I might only work 5 hours in a week instead of 40, so my net production is the same, but my rate of productivity was certainly much higher.
It's like "work" meaning one thing to a normal person, but something very specific and counterintuitive in Newtonian physics (if you carry a fifty-pound weight in a circle and stop at your starting point, you've done no work).
https://www.economist.com/business/2026/07/27/america-has-be...
If it persists for years people will have time to spin up more fabs and meet demand
What happens is that people start buying Chinese mini PCs, some of those even have spyware preinstalled along with another malware that is Windows 11
I’m sure some kind capitalist will be happy to rent them a thin client linked to a computer in the cloud.
Leasing not only allows Apple to keep "selling" people new devices, it will also mostly like accelerate the second-hand sales and resource recycling arms of the business. Consumers will be more willing to upgrade sooner and stay in perpetual payments vs having to buy out the device at lease end - and Apple gets back newer devices where the recycled tech is arguably more valuable.
Companies will prioritize components for cloud computing and lease us thin clients...
Maybe we'll see a return to the late 1990's, when companies would give you a free computer as long as you allowed it to constantly display an ad banner on the top of the screen.
But yes, they all went out of business during when the dot-com bubble burst.
- Lots of options competing on quality, price, user experience, user needs
- Every marketplace innovation that we talk about on this platform
So somebody might offer exactly that and if it's a good option, people will use it. If it's not, they'll use an alternative.
Government regulation will lead to:
- Giant entrenched companies that are difficult to disrupt thanks to insider connections
- Lack of competition, price increases, terrible customer service and pretty much everything that "anti-capitalists" today complain about in regards to capitalism.
Everything government involvement touches gets more expensive. See healthcare, college tuition, student loan debt as some easy examples.
Medicare is cheaper that US private insurance:
* https://www.kff.org/medicare/how-much-more-than-medicare-do-...
Also:
> A single-payer universal health care system could cover every American, save more than 100,000 lives a year, and still cost $1 trillion less than the system it would replace, according to a new preprint study led by researchers at the Yale School of Public Health.
* https://ysph.yale.edu/news-article/universal-health-coverage...
Oh, you mean like Apple, Google, and Microsoft?
> - Lack of competition, price increases, terrible customer service and pretty much everything that "anti-capitalists" today complain about in regards to capitalism.
Have you looked around? This is already happening.
And it leads to comments like the parent blaming "capitalism"
Literally everything listed for 'government regulation will lead to' can and has happened in unchecked capitalism. Insider connections rule everything right now in this current regime of chest-beating 'capitalists.'
Capitalism and socialism aren't real things outside of freshman year econ textbooks.
The one notable exception is if you're a group of workers and you negotiate a contract with your employer: the state can and will interfere to void it if it prevents the workers from being properly exploited.
Macs, as nice as the hardware is, are hostile to repair.
The m6 with same specs are $2000.
Also m4 Mac mini seem to now be in stock on their refurbished store.
I may need to replace some parts at some point, but this far it's not been necessary.
I'd love to have the 16", which didn't exist when I bought the 13", but thanks to Altman, hardware prices are insane these days. :/
Same with my M1 Air for being on the move, even notebooks would be more expensive than they’re worth.
It will last longer and hopefully the GPU prices will recover in a few years.
I hope
If you are running Steam in the background, one of the buttons triggers an HUD keyboard, where the left trackpad types keys on the left side of the keyboard and the right trackpad for the right side of the keyboard. My guess is the new controller works the same way.
Also, some media center software, such as Kodi, have gamepad support. So a gamepad will feel completely natural to control it without having to resort to using the mouse and HUD keyboard.
Personal assistant (e.g. openclaw) in the mac ecosystem. You need external models to power it ofc, but cool to have all the Apple integration and local network in your assistant. The claw can be addressed in multiple ways and can command your local hardware if you so wish.
My son is using my old ~2011 Mac mini with Linux Mint on it. Slow but works.
Not that they're a bad deal now, but at European prices of over €1000 for M6/16GB/256GB it's a psychological barrier that's been broken.
I can't justify a more than 2x price increase for what I'm guessing are marginal improvements. I want to replace my home lab but as it stands my underpowered NAS will have to continue as it has the past few years
https://everymac.com/global-mac-prices/all-msrp-mac-mini-pri...
Though $599 for the M4 in 2024 seems like a local optimum.
We should be buying European solutions anyway. Different psychological barrier that was broken long before.
(250,000 pre-orders, vs now ~5000 sales/quarter)
It is big. No dashboard. square steering wheel without turn signals or gear stalk. The rear-view mirror doesn't see out unless you lower the cover in the back. The side mirrors are strangely shaped and small. The front corners are a mysterious mostly unpredictable distance out the front window.
But then... It has cameras in all directions including the front bumper. There is a top-down sort of heat-map display when you park. All 4 wheels can be turned for a tighter turning circle. and it can self-park.
So... I would feel embarrassed to be seen in it, but it was definitely an interesting experience.
Don't give them ideas...
I don't care how the M6 compares to the M1, even when the M6 2.3x compared to M1 in Excel spreadsheet. That is no measurement even in accounting. The dead time is not the Excel.
I am to only one?
170GB/s memory bandwidth Configurable up to 32GB unified memory
307GB/s memory bandwidth Configurable up to 64GB unified memory
That is also the factor you want to take into account for AI inference, of course.
- Apple M5 Ultra chip
- 30-core CPU with 10 super cores and 20 performance cores
- 64-core GPU
- Neural Accelerators
- Hardware-accelerated ray tracing
- 32-core Neural Engine
- 1.2TB/s memory bandwidth
one point 2 terrabyte a second memory bandwidth....
512gb coming later this year
but at the cost of a small car
I just wait for practical real live numbers with real stacks i use.
And btw. they reduced the memory bandwdith on the M4 in comparision to its previous version by 40gb/sec or so, you need to wait to see.
Says 1.2x for multithreaded and doesn't specify (I presume smaller) for single-threaded. That's compared to M5 not M5 Pro, so M5 Pro is likely significantly faster.
Don’t know, but this as main ad headline feels almost like a threat to my personal home computing needs
I have an old M1 Max with 32GB that runs inference models fine, especially those using MLX.
Dedicated graphics cards are faster but not to the point that it matters.
That’s a six year old machine.
My newer machine M5 max 128GB will far outperform your typical 32GB gaming card once the model exceeds memory.
To even get close to that on a dedicated graphics card you are paying upwards of $20K.
That said, models are getting smaller and faster which allows me to run multiple different models with ease.
[edit] Just checked the 5 models I use take up 57GB when all loaded at the same time.
To be specific, 5 models at 57 gb means you are using crap quantized models, which suck for any real agentic work. I mean, sure they give you some inference, but compared to the full parameter models like Qwen3.8 and Gemma4 that can run full agentic loops, you may as well just use cloud inference for the price.
You of course could "run" those larger models, but we both know that the tok/sec is dogshit on Macs for those.
And 57 gb is split across 3 cards quite easily, which will all be cheaper than your comparable Mac and way faster.
You really need to educated yourself on how running local models works and what the models like Gemma 4 are capable of, so you don't continue to waste money on Macs.
It definitely matters, if you're intending to run a Claude Code/OpenCode style agent workflow. Most of those harnesses start with 8-12k token contexts, which is a lot of prefill for a Mac but cheap for a CUDA GPU. The GPGPU compute on the fastest Macs is still trailing behind Nvidia's laptop GPUs; the highest-bandwidth Apple Silicon chip (now the M5 Ultra) has ~7x lower memory bandwidth than a single B100 card.
There's a good reason why Apple Silicon isn't to be found anywhere in the datacenter buildout. It's nonviable for training, and wastes electricity running real-world inference workloads.
As a disclaimer though, I haven't even tried to really run local models on the M1 as I also have a 'gaming PC' that can run them fine.
However the main reason I would upgrade is to run LLMs locally and RAM prices are making this a no-go. One day if RAM comes down to earth I'll upgrade. If not I'll stick to this old machine for a long time.
I mean, local LLMs already are really good if you own Nvidia Cards, not sure what you are waiting for. 2x3090 will blow any current Mac out of the water in tok/sec
I'm not sure I'd consider a gaming PC 'regular consumer' hardware especially with today's prices. But yeah, it's definitely doable.
Because otherwise a lot of this hardware (including their blazing fast SSDs) is wasted. Cuz my M1 mac mini is awesome otherwise.
For example: it is copy-on-write, and my corruption happened in the superblock checkpoint. There is no built-in command in Disk Utility to "mount" an older checkpoint (and accepting a small amount of dataloss). Also the "First Aid" command allegedly runs volume repair top-down when it is recommended to do it inside-out.
It's the Linux kernel that deliberately does not provide user land a way to change file creation times.
Meanwhile XNU/macOS does.
Interesting that Apple quotes performances when playing Cyberpunk 2077 ! Maybe they will want to care more about gaming ?
And while i'm at it, I was wondering a few weeks ago if I should try this game on my M3 pro. Anyone has played it on Apple and has feedback ?
Not like Mac gaming is good anyway...they have "Game Porting Toolkit" but it's not free and open like Proton so nobody will use it unless Apple pushes them. Removing Rosetta doesn't help either.
The game gets a bad rap for the last-gen console performance, but even 2020's iGPUs could handle it: https://youtu.be/oUm5GmJcBPA
I don't know what this means - game porting toolkit is free
They include game metrics in their marketing often, because they know their customers care about gaming. But no—for whatever reason, it never translates into Apple caring about gaming. I don't get it, but that's the way it has been for a long time.
They’ve also said when they pull Rosetta for intel compatibility, it will still be available for old games to use, which is a huge improvement from their usual stance of “that game studio that hasn’t existed since 2008 should dust off their code and update it to our new architecture.”
So is this performance gain probably due to the new Neural Accelerators?
With larger (also really fast) unified memory on the chip, it could easily load larger weights, and the thing is that the quantized results on the M5s are really good, I believe that for most local inferencing, users would be using INT4 (maybe more bits per weight sometimes) quantized models, that might be where the Neural Accelerators kick in. In raw power, a desktop 5090 easily outperforms the M5 Max, but for this specific use case, I believe that the M5 Max is good enough.