It seems that Apple has kept a intently-guarded mystery pertaining to its new M1 Max silicon. New pics of the underside of the chip reveal that it may perhaps essentially have an interconnect bus that permits Multi-Chip-Module (MCM) scaling, allowing the corporation to stack together a number of dies in a chiplet-dependent design. That could final result in chips with as several as 40 CPU cores and 128 GPU cores. Apple has still to affirm provisions for chiplet-centered designs, but the M1 Max could theoretically scale into an “M1 Max Duo” or even an “M1 Max Quadra” configuration, aligning with persistent experiences of many chiplet-based M1 designs in the long term.
Apple has managed to impress the environment not the moment but twice already with the performance of its Arm-based mostly M1 CPUs. The company’s most recent M1 Max chip is a force to be reckoned with by alone – the chips’ gargantuan 57 billion transistors enable Apple to scale up to 10 CPU cores and either 24 or 32 GPU cores (dependent on the configuration you get), all in a solitary 5nm chip. Introducing lodging for a chiplet-based mostly style would theoretically multiply compute resources, and thus overall performance.
You men looking at this or am I just nuts? The true M1 Max die has an total hidden area on the base which was not shown at all in Apple’s official renders of the M1 Max die. Just flip another M1 Max and connect it for an M1 Max Duo chip. Then use I/O die for M1 Max Quadra. https://t.co/McWmofJAls pic.twitter.com/JogRwUGvF6December 2, 2021
The interconnect bus would permit Apple to scale its chips by “gluing together” the ideal range of M1 Max chips. But, of program, it’s not only a subject of flipping 1 M1 Max chip and aligning it with the next a single Apple would nonetheless have to use precise interposer and packaging options for a chiplet-primarily based layout.
Interestingly, Apple’s M1 Professional chip (which fits concerning the M1 and M1 Max SoCs) lacks the interconnect bus — it really is truly located in the prolonged half of the M1 Max (a beefier variation of the M1 Professional). This likely usually means that Apple only expects people that will need the more graphics compute ability in its M1 Max (these kinds of as graphics or tv studios) to involve even further efficiency scaling through this chiplet design philosophy.
Marrying two 520 mm^2 Apple M1 Max dies in an “M1 Max Duo” chip could provide up to 20 CPU cores and 48 or 64 GPU units. It would also demand an acceptable doubling of the system’s memory to 128 GB. Memory bandwidth need to also scale in this sort of a process, up to 800 Gb/s. That appears doable inside the latest M1 Max design, though the 10,040 mm^2 of Apple silicon would be much more high-priced, of program.
Going for the “M1 Max Quadra” option with 40 CPU cores and 128 GPU cores would be even far more complicated. Maybe an additional I/O die, as the supply implies, is the right resolution, but choices abound. Apple could also maintain sufficient inter-die bandwidth by way of an I/O know-how akin to AMD’s Infinity Cloth. Irrespective of whether or not the bigger chips would require an I/O die continues to be an open dilemma, as other leaks have suggested the style and design will be expanded in a monolithic style and design.
The moment all over again noting that Apple’s advertising and marketing diagrams do NOT accurately represent the ground system of its chips, listed here is a significantly less fanciful interpretation of the Pro Mac SoC rumors.Presenting Jade-C: The making block for Professional Mac SoCs. (M1 included for scale.) pic.twitter.com/Lp8ZBDeLiuMay perhaps 21, 2021
It’s ultimately unclear how Apple would pick to tackle memory bandwidth scaling – and any option would have very improved platform improvement charges all over. But then once more, these theoretical “M1 Max Duo” and “M1 Max Quadra” products and solutions would cater to a current market that cares additional about efficiency and power efficiency than cost.