Most games can run fine on Thunderbolt external GPUs, and that's a PCIe 3.0x4 link (roughly) that, on most Intel implementations, is also bottlenecked by the ch

Most games can run fine on Thunderbolt external GPUs, and that's a PCIe 3.0x4 link (roughly) that, on most Intel implementations, is also bottlenecked by the chipset[0]. Game assets are typically static, so available bandwidth mainly impacts how fast you can stream data into VRAM. Of course, with things like DirectStorage and consoles all embracing SSD-to-GPU streaming and hardware decompression, game developers might actually bother to actually use the extra bandwidth they've been given. However, that's still a way's off and definitely not a requirement to enjoy most games today. The reason why PCIe 4.0 actually became a thing was because of enterprise storage arrays. M.2 slots and U.2 connectors don't have enough pins for 16 lanes, and using up so many of those lanes for one device makes no sense if you need to stick 10 or 20 of them in a server. That's also a use case that doesn't really make sense on AM4, unless you have a bifurcation[1]-capable motherboard or are spending way too much money on M.2 carriers with PLX chips in them. [0] AFAIK, Intel wanted Thunderbolt direct-to-CPU but there was some weird driver/certification nonsense with Microsoft or something, and going through the chipset apparently made it easier for vendors not named Apple to support it. I don't remember the details. [1] The ability to drive multiple PCIe devices off the same slot by splitting the slot's lanes. Most M.2 carrier boards are wired up to work this way because proper PCIe hubs are absurdly expensive... because the entire market for such chips are just storage array vendors.