logoalt Hacker News

_factortoday at 12:50 AM1 replyview on HN

They exist to partition capability so that enterprises can’t connect all of their peripherals and some ECC memory to get the same functionality for 1/10 the price. It’s not a physical limitation.


Replies

greycoltoday at 4:05 AM

Obviously market tiering is part of it and you can play tricks with north and south bridge and pcie switches (which adds cost), but a ryzen board that advertises a pcie 5.0 x16 gpu slot and 5.0 x4 m2 slot only has 4 lanes left to work with from the cpu (i.e the cpus only have 24 usable lanes). Which while you can play with generations to get more lanes it's effectively still 16gb/s. That needs to cover network, extra m2 slots, usbs, as well as the extra PCIe slots.

I don't mind having to work within those physical limits but I do want to be able to search for boards that support N components. i.e 1x 4.0x8, 2x 3.0x8, 4x 5.0x4 . But the best you can search for is physical sizes of pcie slots and then dive into a spec sheet for each one, only to find that the 6 x16 slots only have 1.0x1 of bandwidth each.