It’s more that if you have a purely sequential workflow, you get access to less than 2% of the CPU on a 64 core machine.
And even if you’re a bit parallel, Amdahl’s law smacks you pretty hard here.
Plus you get way more thermal room.
Plus you can disable SMT.
And there is always more than 1 process running if you take into account kernel threads.
So I would say way quite a lot more than 2%. More like 10%.
That's still essentially single digit percentages. And if is really in the low double digits, you're back to single digits when you go to 80 cores.
On the positive side, I don't think multi-user systems would be where they are today if we hadn't been staring at a future full of hardware that can only be fully utilized by embarrassingly parallel tasks.
It doesn't help that the default behaviour in Windows is just stupid:
"By default, an application is constrained to a single group, which should provide ample processing capability for the typical application."
This is the new "640KB should be enough for everybody", except now it is "64 threads should be enough".
See:
https://docs.microsoft.com/en-us/windows/win32/procthread/pr...
https://www.anandtech.com/show/15483/amd-threadripper-3990x-...