Thats why we need current x86 ecosystem where youre free to match hardware with any other compatible parts you want.
Except for memory of Lunar Lakes, but they confirmed that they will not continue this soldered memory fashion. It was just to prove that x86 can fight against ARM when it comes to energy efficiency
> Thats why we need current x86 ecosystem where youre free to match hardware with any other compatible parts you want.
Let's remember that this ecosystem is a mistake that no party would willing recreate again. Given the chance to improve the x86 with 64-bit capability, Intel chose to form a cabal and create a patent-locked monstrosity that no one else could approach.
IBM didn't want to create an open-platform, they just did want someone else to set the standard and knew that they were too slow to create something from scratch so they had to grab off-the-shelf parts. On their second attempt they tried to create their own OS (/2) and bus standard (MCA) with egregious terms. Sony memory sticks have been cut open to be found to be micro-SD cards with extraneous packaging.
Sun tried to open up SPARC... and then close it up again with UltraSPARC... and only reopening it as a 11th hour hail-Mary.
That is why both RISC-V and GNU/Linux are so special. They are really the only systems from the get-go designed and intended to be open.
We should really appreciate and support those because these things don't usually happen on purpose and are not easily replaced.
AFAIK now the ram gets to be way closer this way (which is why the M series gets to destroy everyone on benchmarks where it can show off that).
I think that it'd be cool to have that memory become an L4 cache and still have cheaper ram as a backing store with more capacity to fence off the CPU from the abysmal latencies that SSDs and, satan forbids, HDDs have.
The RAM speed comes from the 512bit wide bus, not the direct connection. If motherboard makers were to support quad or octa channel interfaces, PCs would be just about as fast.
There is no technical reason that forced Apple to make the RAM soldered. They could have easily made a proprietary but replaceable socket/connector, but chose not to because of the sweet, sweet markups.
There still is today; any custom built desktop can be upgraded easily. Most laptops that are more enterprise focused still have RAM and SSD slots, and replaceable batteries too.
AI bubble is finally forcing companies like Apple to upgrade memory. 8GB RAM is now not enough, but 32GB RAM is still a luxury!
Finally I will not have provide support for "development workstation" with 8GB RAM! Some people do data analytics on such machine, because M1 Macs are suppose to be fast! Then soldered SSD melts from swapping and write wear.
We had similar situation with Windows Vista. Hardware makers with 512MB or even 256MB RAM were finally forced to upgrade!
I feel 16G Mac is singlehandedly stopping the arms race of bloated apps and beefier machines, especially since things become progressively slow (RAM compression and swapping) instead of crashing.
I don't really care how they price different spec'd models, as long as desktop units are user-upgradable for both memory and storage.
I thought I'd be excited for the Mac Mini announcement but when I heard it, I wanted to be excited but couldn't really see a use for me. If I want a beefy AI GPU machine I wouldn't buy from Apple, similarly for a gaming PC. Otherwise I'd use a MacBook or Linux desktop/server.
If you were buying a beefy AI gpu machine today would you get? I’m in the market and because I’m quite Mac based was considering a MacBook Pro M4 Max but they are painfully expensive. So then I was considering the Mac mini as “chuck stuff at this box” but keep my existing Mac laptop. But if I’m just chucking stuff at the box it doesn’t really need to be a Mac at all…
It costs $400 just to get 32GB RAM on the M4 mini. Retail 32GB DDR5 kits are about $100, 64GB kits about $200. Yes, yes, technically it's LPDDR5 and whatever.
Realistically the BOMs are nowhere near the retail prices in either case. It's likely that if upgrades were sold at their BOM difference the lower end machines would be more expensive.
In other words the low end machines are subsidized by the upgrades.
As this makes them more accessible I think this is ok, but only if the base model has serviceable specs. I'm glad they finally do.
Vertical scaling is usually more expensive than horizontal scaling. This is not a surprise. What surprises me is that it’s only $1 more expensive here. I suppose that’s because you don’t get double the number of CPUs in the vertical case.
Well, maybe it's a wrong analogy, but when you see a 1l glass bottle of stuff selling for almost exactly twice the price of a 0.5l glass bottle of the same stuff, then there is something off because glass bottles themselves ain't free either.
Thats why we need current x86 ecosystem where youre free to match hardware with any other compatible parts you want.
Except for memory of Lunar Lakes, but they confirmed that they will not continue this soldered memory fashion. It was just to prove that x86 can fight against ARM when it comes to energy efficiency
> Thats why we need current x86 ecosystem where youre free to match hardware with any other compatible parts you want.
Let's remember that this ecosystem is a mistake that no party would willing recreate again. Given the chance to improve the x86 with 64-bit capability, Intel chose to form a cabal and create a patent-locked monstrosity that no one else could approach.
IBM didn't want to create an open-platform, they just did want someone else to set the standard and knew that they were too slow to create something from scratch so they had to grab off-the-shelf parts. On their second attempt they tried to create their own OS (/2) and bus standard (MCA) with egregious terms. Sony memory sticks have been cut open to be found to be micro-SD cards with extraneous packaging.
Sun tried to open up SPARC... and then close it up again with UltraSPARC... and only reopening it as a 11th hour hail-Mary.
That is why both RISC-V and GNU/Linux are so special. They are really the only systems from the get-go designed and intended to be open.
We should really appreciate and support those because these things don't usually happen on purpose and are not easily replaced.
To note that UNIX was also only open due that AT&T wasn't allowed to profit from their research at the time.
> Let's remember that this ecosystem is a mistake that no party would willing recreate again.
The human species is not known from learning from its mistakes. /s
In old times there were an option to upgrade ram and storage using stock parts form hardware store, but now everything is soldered.
AFAIK now the ram gets to be way closer this way (which is why the M series gets to destroy everyone on benchmarks where it can show off that).
I think that it'd be cool to have that memory become an L4 cache and still have cheaper ram as a backing store with more capacity to fence off the CPU from the abysmal latencies that SSDs and, satan forbids, HDDs have.
The RAM speed comes from the 512bit wide bus, not the direct connection. If motherboard makers were to support quad or octa channel interfaces, PCs would be just about as fast.
1 reply →
There is no technical reason that forced Apple to make the RAM soldered. They could have easily made a proprietary but replaceable socket/connector, but chose not to because of the sweet, sweet markups.
6 replies →
A CAMM2 memory based design would likely work just as well.
Intel Lunar Lake is proving this is not necessary.
It is still current times in PC desktop land, and why I keep on Windows/Linux land.
There still is today; any custom built desktop can be upgraded easily. Most laptops that are more enterprise focused still have RAM and SSD slots, and replaceable batteries too.
Well you don't need to let that stop you.
AI bubble is finally forcing companies like Apple to upgrade memory. 8GB RAM is now not enough, but 32GB RAM is still a luxury!
Finally I will not have provide support for "development workstation" with 8GB RAM! Some people do data analytics on such machine, because M1 Macs are suppose to be fast! Then soldered SSD melts from swapping and write wear.
We had similar situation with Windows Vista. Hardware makers with 512MB or even 256MB RAM were finally forced to upgrade!
I feel 16G Mac is singlehandedly stopping the arms race of bloated apps and beefier machines, especially since things become progressively slow (RAM compression and swapping) instead of crashing.
Is that actually helpful? More requirements & more ram. Doesn’t this leave you where you started?
Or maybe you aren’t very interested in the AI stuff and you just get a better machine.
I don't really care how they price different spec'd models, as long as desktop units are user-upgradable for both memory and storage.
I thought I'd be excited for the Mac Mini announcement but when I heard it, I wanted to be excited but couldn't really see a use for me. If I want a beefy AI GPU machine I wouldn't buy from Apple, similarly for a gaming PC. Otherwise I'd use a MacBook or Linux desktop/server.
They aren't though. Everything is soldered down or uses custom stuff that's hard to replace.
32GB RAM is not enough for AI either - the problem is that they lock down any user upgradeability and over price every possible change.
If you were buying a beefy AI gpu machine today would you get? I’m in the market and because I’m quite Mac based was considering a MacBook Pro M4 Max but they are painfully expensive. So then I was considering the Mac mini as “chuck stuff at this box” but keep my existing Mac laptop. But if I’m just chucking stuff at the box it doesn’t really need to be a Mac at all…
Those days are long gone in Apple land.
Previously (10 points) https://news.ycombinator.com/item?id=42040972
RAM and storage upgrades have always been absurdly expensive at Apple. It never stopped millions of people from buying them anyway.
It used to be pretty easy to upgrade storage on Mac’s up until apple silicon
The last few intel generations were soldiered too. I'm still using one.
But how many more millions did it stop?
That's great! You're getting an SoC, chassis, power supply, cooler, ports etc. for free! :D
Sounds like someone has yield problems...
Market segmentation.
The people who want the extra RAM and storage will pay. Others will make do, or use external storage.
One of the many apple taxes people seem to love overpaying for
So the 16GBx256GB Mac Minis are discounted?
Or the 32GiBx256GB ones are overpriced.
It costs $400 just to get 32GB RAM on the M4 mini. Retail 32GB DDR5 kits are about $100, 64GB kits about $200. Yes, yes, technically it's LPDDR5 and whatever.
1 reply →
Realistically the BOMs are nowhere near the retail prices in either case. It's likely that if upgrades were sold at their BOM difference the lower end machines would be more expensive.
In other words the low end machines are subsidized by the upgrades.
As this makes them more accessible I think this is ok, but only if the base model has serviceable specs. I'm glad they finally do.
I think the base configuration recently jumped from 8GB to 16GB while keeping the same price.
1 reply →
basic answer is yes; essentially you are getting more memory bandwidth compensated by people buying higher RAM.
Vertical scaling is usually more expensive than horizontal scaling. This is not a surprise. What surprises me is that it’s only $1 more expensive here. I suppose that’s because you don’t get double the number of CPUs in the vertical case.
Well, maybe it's a wrong analogy, but when you see a 1l glass bottle of stuff selling for almost exactly twice the price of a 0.5l glass bottle of the same stuff, then there is something off because glass bottles themselves ain't free either.
Of course, when it comes to bottled water my experience is the bigger bottle is cheaper than the small one!
> Well, maybe it's a wrong analogy
Yep. Try comparing the price of a 1 L bottle with 1 kg of water to a 1 L bottle with 2 kg of water. Vertical scaling can be very, very expensive.