Seems Nvidia isn't the only one beefing up GPU to GPU communication speeds :
https://www.guru3d.com/news-story/radeon-vega-20-will-have-xgmi-interconnect.html
Seems AMD will be unveiling a new high speed communication bus on Vega 20, similar to what Nvidia has done with their new bus protocol, and this one is based on infinity fabric which is the same as what their thread ripper CPU's use to communicate between the active CPU dies, so it's a bus with a 50 GB/sec transfer speed between 2 active dies at least...… Both AMD's and Nvidia's are much faster than using the PCI-e bus, which tops out at 16 GB/sec for 3.0 and doubles for PCI-e 4.0 to 32 GB/sec so that put to rest the need to have a faster PCI-e bus in future motherboards......The heaviest thing one can run given the amount of data involved is a video card by far, and the rest is easy by comparison.
https://www.guru3d.com/news-story/radeon-vega-20-will-have-xgmi-interconnect.html
Seems AMD will be unveiling a new high speed communication bus on Vega 20, similar to what Nvidia has done with their new bus protocol, and this one is based on infinity fabric which is the same as what their thread ripper CPU's use to communicate between the active CPU dies, so it's a bus with a 50 GB/sec transfer speed between 2 active dies at least...… Both AMD's and Nvidia's are much faster than using the PCI-e bus, which tops out at 16 GB/sec for 3.0 and doubles for PCI-e 4.0 to 32 GB/sec so that put to rest the need to have a faster PCI-e bus in future motherboards......The heaviest thing one can run given the amount of data involved is a video card by far, and the rest is easy by comparison.