All modern desktop motherboards have very limited expansion IO available compared to a similar class motherboard from a decade ago. Part of this is that all the IO protocols now are so fast you can't just have 8x nvme storage hook-ups (like you'd have 8x sata in the past) in the space of a modern board's controller chip(s) and the CPU. IO has grown faster than processing. And sharing pcie with GPUs makes this even worse.
The market has changed and desktop PCs assembled by humans with the expectation of adding new parts later is relatively smaller. And a smaller consideration is given to them versus the big (server/mobile) markets needs for pcie. This probably won't change but I think an article pointing it out as a bad thing for desktop computers is welcome. Maybe some high end mobo manufacturer (Asus?) can make a real non-server mobo with some actual physical IO (extra pcie 5 switch/bifurcation magic?) for the niche markets.
On what do you base this? The only thing I can think of is that a lot of unpopulated x1 PCIe slots got replaced with the much more useful m.2 slot.
re: HBAs in general, yeah, I'll admit it's weird but I really don't like having to use external active controller cards for reading my storage. I just want my motherboard to do it to keep things simple.
Dealing with that speed increase is causing a whole lot of problems for all the designers in the ecosystem. It's a bit of a nightmare, but people are for sure using it (in your laptops, in the data center).
We ended up adding some specific features/strategies to the autorouter we're building for to handle those speed. It was pretty hard to get right.
Not anymore at least. Top-tier "GPUs" these days don't even have plugs for monitors. You won't see much of an improvement either by maxing out the memory - anything beyond 16GB will make very little change.
You might want to explore RAID-1 arrays to reduce that 9µs random read latency to 4.5µs or less (with more SSDs), but there is little benefit in transferring 8K UHD content faster than the monitor's frame rate allows one to watch it. I no longer see storage like that in the wild (as such things get abstracted away in cloud platforms) but I still remember RAID-1 "readzillas" and RAID-0 "writezillas" with a dozen or more fast disks (we used spinning metal when I last saw them). These are specially useful for back-ups, as they can ingest a truckload of data and spit them to tape at the maximum speed of the tape (avoiding a lot of wear and tear).
So whats the second thing? It's never mentioned AFAICT. Assuming they meant in addition to PCIe host controllers on your CPU die (though could be in a south bridge or whatever) you need a PCIe 5.0 device?
The rest seems to complain about higher power consumption and heat. This is data center shit anyway so power and cooling aren't big issues there. Given enough time these issues will be fixed once consumer vendors starts caring enough to fix it. Or not. Who cares.