This is very fresh, about 1 month old, context is that the latest nvidia 435
driver added support for render offload and driver controlled dynamic runtime
pm (turing only).
So I don't expect Nvidia to talk about bugs only existing in old kernel
versions.
It seems to me that the mentioned commit from the nvidia docs breaks runtime-pm
on the nvhda device but the subsequent commit refered to by Przemysław Kopa
doesn't fix it.
As to what "ongoing works on the audio driver" means, this is out of my
knowledge, would need to poke nvidia staff about it.