I think we’re likely to see some kind of tiered storage (maybe still 8GB of GDDR5, but some additional memory running at slower speeds that’s volatile and then segmenting the SSD into a bigger, slower part and a smaller really fast part.
The technology all exists to do this and it’s ideal for the kind of local inference Apple wants to push.
The technology all exists to do this and it’s ideal for the kind of local inference Apple wants to push.