No, its not, and I heard the same nonsense about the Ultra chip.
Physically it is. There is only one edge the 2x set up has to deal with. That one edge doesn't scale to four dies.
The memory layout also is a substantial impediment. The die is just shaped 'wrong' to scale up past two. The problem is the Max die is highly optimized to just get to 2 while simultaneously squeezing onto a narrow laptop logic board and that substantially impedes going further
You could get a CPU only set up to mimic something similar to what AMD does with their chiplets , but the GPU cores are far more NUMA sensitivity issues when coupled to high screen refresh rates. [ AMD just does equally 'worse' NUMA for all the cores. That works for CPUs. It doesn't work so well for GPUs... e.g., AMD's completely
reversed orientation with the GPU chiplets they are using for high end mainstream graphics. AMD has a single edge fused CDNA solution but there are no display controllers on that solution on purpose. (i.e., it isn't a single user , GUI focused GPU. ) ]
In short, I am not talking about the what you might have heard about the Ultra chip at all.
Dogmatically clinging to the laptop optimized dies as super chunky chiplets reeks of a Rube Goldberg attempt to save money by driving up usage of the die because can't afford to do it right (e.g, can't get the green light to get a die that is a far better match to the problem because the volume is judged to be too small ). The problem there is that Apple's expenses may have been contained but the end user cost skyrockets. If it is too expensive and 'practically nobody' buys it , then some technically working solution is not of much productive use.
Pragmatically they need at least to UltraFusion connector edges to make something decent work. And once you do that it will decline as a monotholithic , narrow laptop logicboard solution. Therefore, they need another die. That doesn't mean throwing the P or E core cluster designs out the window, LPDDR5 controllers out the window. , unified memory out the window , integrated TB controllers out windows. It just needs a better disaggregation.
It just means at least one more die in the line up that is more desktop focused die where routinely use multiple chip/die (primarily chiplet focused ) solutions. This wouldn't be a "mac pro only" usage. The Mac Studio has market targeting problems too. If Apple brought back an iMac Pro; same thing.
With better disaggration they could have 3 die solutions at better price points for some users.
P.S. to tie it back a little to the thread's topic area. Intel's four chunky chiplet solution for 3400 isn't winning any prizes for affordable to make , healthy margin returns producing sales either. But at least they were designed to fit well together in dense packed orientation. The Max is not.
www.macrumors.com
