PCIe is too physically large, Oculink is too thick and doesn't have enough pins for power, and Thunderbolt has protocol overhead and throughput limitations.
It seems to me that oculink port is still a good option for side module, although it's unrealistic. I hope ltt lab create custom gpu module from the desktop 4090: built-in power, water cooling and of course rgb :)
Yeah, sorry, no. Thunderbolt does have overhead and limitations, but imo those limitations don't justify making a new standard, even if it's more open than the others. Maybe you could say more about those limitations and convince me?
For instance, Thunderbolt bandwidth is limited to 40 Gbit/s. The Framework adapter for GPU is PCIE x8, which gives 128 Gbit/s.
(not affiliated with the Framework company, only as a owner of two 13" Framework laptops; so I might be wrong/incomplete about the limitations implied by the OP)
There are companies putting nonstandard PCIe ports on the outsides of laptops, and compared to that an internal custom port is much less egregious. 40Gbps is just not enough to run a GPU without significant performance loss. Especially when 40Gbps is actually 32Gbps of PCIe and on older controllers there's an arbitrary limit of 22Gbps.
I really hope thunderbolt 5 in asymmetrical mode can devote the entire 120Gbps of outgoing data to PCIe, and doesn't have some stupid limit like 64Gbps.
And when a new standard amounts to "arrange the pins like this", it's not a very big deal.
