You're a processor manufacturer, and you need to release a more powerful model next year than the one you currently produce. How do you do it?
The classic answer is multifaceted.
It includes, among other things, using a smaller lithography node in the manufacturing process, adding more cores, increasing the clock speed (while keeping temperatures reasonable), and expanding the cache memory.
With Apple Silicon processors, Apple took a similar approach when it released the improved "Pro" and "Max" variants of the M1 chip in 2021.
But the truly interesting innovation came later — in March 2022, with the announcement of the M1 Ultra.
-
Using multiple processors in a single system is not a new method, and it has long been common practice in the server world.
To use multiple processors in the same system, you typically need a motherboard with multiple CPU sockets, each housing a separate processor.
Apple took a different approach. It connected two M1 Max processors together to create a single Ultra chip.
The processors are connected directly to each other via a bridge called UltraFusion.
This bridge links the two processors through 10,000 data pathways, with a data transfer rate reaching a staggering 2.5 terabytes per second! (Not bits — bytes!)
Connecting the processors directly to each other enables exceptionally high combined performance, while also reducing energy consumption and latency.
In the video:
Two M2 Max processors connected via UltraFusion to create a single M2 Ultra chip.