They could. If Intel CPUs supported such a protocol. But they don't. So they can't.
What won't ever happen is an M1 PCIe board that lets you run ARM apps or does general purpose work. There is no way for the M1 to efficiently sync memory with a Xeon. Yes, if a Xeon support Infinity Fabric it could sync with something like an M1. But it doesn't.
I would have swear that vega2duo were interconnected via infinity fabric.... in a macpro, so if they could do it over MPX module on amd silicon, I guess they could do it over their own sillicon... It might need an external connector to work....
And any how they still could do it over just plain pcie gen3 x16 like cluster computers linked over 100gb FC .
I’m not referring to multipro clustering, I am refering as clustering ala XGrid.
a m1 is only 14w and the die is very small... basically they could fit 10 M1 on a fullsize pcie card if they want...
a 6900, will never ever do anything else than what it is intended to do, and will never ever have the render quality of cpu rendering...
render the same scene in bender with cpu/gpu and see how it looks...
don’t forget the 2018 mac mini cluster over 10gbe ... apple know how to do clustering, and macmini farm make a lot of sense.
what i am referring to is only a pcie version of that...
once again apple would probably never do that , but i would totally pay 3000€ to have a dual M1 pcie card that shows just like network macs and a kind of « xgrid » app to handle everything... a couple more thunderbolt port and maybe a 10gbe rj45, or a external cfexpress slot to connect small video capture card....
anyhow lets not hijack this post with our crap as anyhow apple will probably never do that....