The PC industry has been scroogling users on I/O since the late 2000's, the main entity to blame is Intel with its vainglorious plan to take over the complete BoM for PCs. Now AMD is at the center of the phenomenon.
This wasn't an unreasonable worry for nvidia to have. They'd just lost preferred chipset status on AMD platforms when AMD purchased ATi. Intel had released some lower end platforms with only 1x pcie connections (nvidia had responded with nvidia Ion which could work over such a tiny link) and AMD was talking big about their APUs.
It was a reasonable fear back then that Intel and AMD, with a tightened grip on their platforms and their own integrated solutions competing in the same space, might choose to just cut off nvidia's air supply by flooding the market with chipsets that didn't have the connectivity Nvidias cards needed.
What does this mean? (I'm familiar with the Microsoft ad campaign, but that doesn't make sense to me in this context.)
Here's a simple example.
I found an old i3 PC from Intel in my house that was left by one of my son's friends. The CPU is artificially limited to 16 PCI lanes.
You might think you could plug a discrete GPU in, it only uses 16 lanes, but no, some of those lanes are consumed by storage, USB ports, super I/O, etc.
So this computer is e-waste now because it can't be upgraded to keep up. This kind of delinquency is only possible because Intel has pushed barely-functional integrated "GPU"s.
Back in the 1980s and 1990s you had to know some rules about how interrupts were assigned to slots to build a working PC. Since the early 2000's, PC builders face a number of barely documented rules about how PCI lanes are assigned which boil down to "i3 and i5 are e-waste, buying a budget or mid-range motherboard is a waste of money because upgrading your machine will be a matter of 'you can't get here from there'"
I personally did "x16-x1 mod", ie removed the plastic from the connector so a x16 video card could be installed in a x1 slot (and free up the x16 slot for a network or RAID card, don't remember atm). Video card worked fine. It's up to PCI-E standard to actually use any amount of lanes available.
https://en.wikipedia.org/wiki/File:PCIe_J1900_SoC_ITX_Mainbo...
> scroogling
So the PC acquired a very diverse set of cards and was able to do heavy processing in a very diverse set of roles - from medical equipment, cash registers, MIDI/music composition, graphics, publishing, sound, networking, etc. No business was untouched by the PC in the 80's--the ability of anyone to make hardware for the platform was a significant part of that being as cheap as possible.
PC hardware seems to be slowly moving towards being like a cell phone - everything onboard/builtin, ports picked by the manufacturer (you will get 1 USB-C port and like it), and if you want to add something, your primary option is a USB-type port of some sort - with its own world of firmware, controllers, etc.
Look at a smartphone and a modern ODM-designed cheapo laptop motherboard - they're very similar.
If motherboards stop being expandable it might mean the PC hardware ecosystem isn't going to be able to innovate except at the behest of high-capital firms like Intel, AMD, etc. That might be OK if we are truly at the zenith of what's possible with a PC, and really want to give the keys to the kingdom to those firms.
And who knows - I know somewhere in there USB4/Thunderbolt on some level is able to move PCIe traffic, so maybe it won't be so bad.
No, we don't, and they've already shown we can't.
Intel kept us at 4-core shit-tier processors for a decade, and the moment AMD managed to get a leg-up on Intel, they killed their HEDT platform because it competed with their server platform in PCI Express lanes and turned their back on all the gamers that brought their sorry ass back from the brink of death.
Both of them have been shown to be greedy, opportunistic shitbags of companies that cannot be trusted.
Back when the USB1 spec came out it said you could plug 127 devices into a port through a hierarchy of hubs.
With three different laptops and a plethora of USB 3 hubs I found the system would support only a limited number of devices in the tree. If you plugged in too many devices (somewhere between 3 and 5), plugging in an new device would cause an existing device to drop out. It's just annoying if it is a keyboard or mouse that gets dropped out but if it is a storage device data could get corrupted.
I looked at the USB 3 spec and couldn't find any guarantee that there was any number of devices you could plug into hubs and have it work.
It's a controller and/or BIOS limitation[1][2]. The complication is that while USB devices can be addressed using 7 bits = 127 devices total, each device usually creates more than one endpoint, and each endpoint consumes controller and system memory resources. The BIOS allocates the memory, and the amount is apparently hardcoded (guess a setting would be too difficult). If you have many USB 3 devices with a lot of endpoints, that memory runs out quickly.
In addition, each endpoint reserves some bandwidth, so the uplink needs to be able to provide that bandwidth.
[1]: https://community.intel.com/t5/Embedded-Intel-Core-Processor...
[2]: https://borncity.com/win/2019/09/06/windows-10-not-enough-us...