Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

senttoschool

macrumors 68030
Original poster
Nov 2, 2017
2,626
5,482
You might have missed the fact that the M1 Ultra will present two GPUs as one to the OS. This is a huge hurdle crossed for GPUs.

We've gone from SLI/CrossFire where games and applications had to be hand-optimized to take advantage of multiple GPUs to this. Hand-optimization is ultimately what killed the consumer market for multi-GPUs.

Apple leap-frogged AMD and Nvidia. This means Apple is at least 3 years ahead of AMD and Nvidia given that Lovelace and RDNA3 will not be transparent multi-die based.

This is huge. We all know that the future of CPUs and GPUs is chiplet/tile-based. Monolith just doesn't work anymore when we're trying to scale due to chip defects.

Note that AMD has the Instinct MI200 server GPU that glues two GPUs together but they're presented as two separate, independent GPUs to the OS.
 

bobmans

macrumors 6502a
Feb 7, 2020
598
1,751
Just imagine in a couple years when Apple comes up with some kind of dedicated GPU that’s stoll somehow using the unified memory and high-bandwith with the cpu. The M1 Ultra probably already beats every single dedicated GPU out there which is kind of crazy.
 

crazy dave

macrumors 65816
Sep 9, 2010
1,453
1,229
Yeah that is pretty cool, hadn't thought about that. Intel's Ponte Vecchio will also use chiplets in its GPU but it obviously won't be available to mere mortals whenever it does release (soonish I think?).
 
Last edited:

leman

macrumors Core
Oct 14, 2008
19,521
19,678
Yeah, Apple is the first one to bring “proper” multi-chip technology with ultra-fast interconnect to the market with this as far as I know. Curious to see Intels solution.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.