Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Pretty sure a fully kitted out IIfx could break that. Before counting inflation. Possibly a Macintosh II or IIx. Since memory cost even more.
The Macintosh IIfx sold for $10,000 - $12,000, depending on configuration, before inflation. Now with inflation... 😉
 
Because there will always be people that need 1.5 TB of Ram! Trust me there will be that guy that will always say “you can never have enough RAM” 👌🙃
With AI, we have a tech that can and will actually use all this RAM.

Other than that, this feat would be totally pointless.
 
This will be a great machine for running AIs locally. Also, it will come out in 2027 at peak memory pricing - now seems to be just the beginning - and be unaffordable because of AI
 
So this can't be unified memory only I guess. Will this be a unified + dimm slot design?

Largest chips available today are 32 or was it 48 Gbit? How are going to cram all those chips around the processor die?
 
So this can't be unified memory only I guess. Will this be a unified + dimm slot design?

Largest chips available today are 32 or was it 48 Gbit? How are going to cram all those chips around the processor die?

TSMC has developed two processes for this: Chip-on-Wafer-on-Substrate-L and stacking. The M7 will likely use LPDDR6 or HBM4, with each chip possibly being 128GB or 256GB, so a 12-stack or 6-stack configuration could provide 1.5TB, both of which are feasible.
 
If Apple EULA allowed for virtualization of more than 2 instances of MacOS at a time, I could think of a few uses for this much RAM.
 
Change in max memory do not affect 99.9% of Apple's customers.

Currently, people running big (open) models without silly levels of quantization (compression) use a cluster of Studios (e.g. 512GB each) via Thunderbolt 5 RDMA. Brute force. Having a single machine run the entire model would be somewhat faster obviously, plus architecture improvements, but wouldn't necessarily be a game-changer.

A game changer would be a big reduction in the size of the models required for state-of-art performance.
 
  • Love
Reactions: jido
Change in max memory do not affect 99.9% of Apple's customers.

Currently, people running big (open) models without silly levels of quantization (compression) use a cluster of Studios (e.g. 512GB each) via Thunderbolt 5 RDMA. Brute force. Having a single machine run the entire model would be somewhat faster obviously, plus architecture improvements, but wouldn't necessarily be a game-changer.

A game changer would be a big reduction in the size of the models required for state-of-art performance.

But what if folks could use a cluster of 1.5TB Mac Studios...? ;^p
 
The Macintosh IIfx sold for $10,000 - $12,000, depending on configuration, before inflation. Now with inflation... 😉
and that was for the base model with 4 MB memory, you could upgrade it to 128.

Edit: The "Dash 30fx" was much more expensive, a souped up version of IIfx, rebuilt by the company 68000. Overclocked and with a huge case. Built to order, some say it should have cost about $40.000 (unknown ram config)
 
Last edited:
  • Like
Reactions: MacHeritage
According to this forum's staunch supporters of greedy Silicon Valley CEOs, it's much more important for Silicon Valley corporations to build data centers and hog up memory chips for AI slop than it is for ordinary people to have easy access to and buy reasonably priced memory chips.

I’ll take strawmen no one ever said for $100, Alex.
 
Apple doesn’t drop support for any Macs for well over a decade. This ain’t windows.

Not sure if that is missing a sarcasm emoji or not.

Mac Pro 2019 is getting upgrades in 2029? ( never mind device bought in 2022 getting support to 2032 ) Probably not. Intel support is dropped now. Security fixes likely to be few and far between. Rosetta 2 support for mainstream apps in 2029? Dead on systems still getting OS upgrades.


Meanwhile Windows 86->Arm translation support lanished for long while because was 32 bit targeted only.



P.S.

%E2%80%8Emacos-support-timelines-2023.%E2%80%8E003.png




Apple’s standard legacy/ obsolete clock is 5 + maybe 2 years after stop selling . The standard churn on most Mac models is 1 year. So that is 8 years. Then standard practice is two years of security updates. So Maybe 10 if lucky.

Are their folks who did firmware hs Kerry to “cheat” the time windows that Apple provided ? Yes. But somewhat farcical to label that “support” when the official support status is obsolete.
 
Last edited:
  • Like
Reactions: platinumaqua

Apple's M7 Ultra Chip Designed to Match a 2019 Mac Pro Feat​

In other words, thanks to Tim Cook, Apples top-of-the-line chip that will be released in 2028 will finally be able to do something that Apple’s top-of-the-line chip released eight years earlier in 2019 was able to do.

“Can’t innovate anymore, my a**!”
 
Apple doesn’t drop support for any Macs for well over a decade. This ain’t windows.
I guess you forgot the old days when certain Macs were dropped from updates over the lack of FireWire. More recent examples have some random Macs support dropped because of the year introduced (and another Mac with the same chips still getting upgrades).

M1 series' days are getting numbered too, no exception to not getting support for their latest and smartest AI in Golden Gate even if someone got the Pro/Max/Ultra versions of chips with extra cores and RAM.
 
In other words, thanks to Tim Cook, Apples top-of-the-line chip that will be released in 2028 will finally be able to do something that Apple’s top-of-the-line chip released eight years earlier in 2019 was able to do.

“Can’t innovate anymore, my a**!”
Except that the new chip and its integrated RAM will be literally 20x faster.
 
  • Like
Reactions: eldho
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.