John Siracusa1d ago
To better days…
Show threadAaron1d ago

@siracusa the biggest disappointment about this is that Apple never did bother to solve their RAM ceiling problem in Apple Silicon Macs. I was hoping to see some clever solution that added slotted memory that had a direct connection to the rest of the SOC to be super fast and low latency.

Hell, I would have settled for a version that was architected like the Intel Mac Pro with slotted RAM and higher capacity at the cost of latency.

Instead Apple just didn't bother to try to solve it.

Show threadTorb1d ago
@harpaa01 512Gb RAM is a too low ceiliing?
Show threadRegis Freyd1d ago
@torb @harpaa01 keep in mind that for sound designers in heavy production, for instance, they need 1tb of RAM at least to have a snappy setup.
Show threadzed
@regis @torb @harpaa01 you’re going to need to provide a source for this because this is not believable.
Mar 26 at 11:07pmWeb
Show threadAaron1d ago
@zed I forget whether it was part of the 2019 WWDC keynote or in some supplemental materials released that same week but there were interviews with audio professionals who were delighted to finally be able to have Logic sessions where they could actually edit a full recording with dozens of different instruments on their own tracks, and this was something they were finally able to do because of the RAM levels. MKBHD discussed this when he reviewed the 2019 Mac Pro as well.
Show threadzed18h ago

@harpaa01 I appreciate you trying, but some people desiring high amounts of ram doesn’t back up the claim that “sound designers in heavy production” need 1TB of ram.

The 1TB of ram being the baseline for “need” is the real sticking point.

Show threadcocoviper7h ago

@harpaa01 @zed
What I’ve heard about this is it’s based on older pro app architectures that didn’t stream files in very well from the disk. But that’s not really a limit anymore (both because of better designed apps and ever faster SSDs).

Agreed though it needs an actual test with empirical data. The only 1TB RAM single user use case I’m aware of is local SOTA AI.