Atemu

Atemu, 1 year ago

Why bother with such micro optimisations when the purpose is to be used extremely infrequently for compatibility reasons?

Atemu, 1 year ago

there’s a different nvidia driver for each kernel version. Already a stupid design

That’s not a stupid design at all. A nvidia kernel module artifact is only compatible with exactly one kernel ABI. Thus you need one binary nvidia package for each kernel you ship.

Arch also has one package for every kernel ABI they ship: nvidia and nvidia-lts.
Though it should be noted that their design assumes that these two ABIs are the only possible ABIs which isn’t strictly the case as the zen, hardened or RT variants may sometimes lag behind their regular counterpart. That’s a stupid design if anything as it increases the friction of kernel ABI upgrades as a kernel package maintainer.

We at NixOS also ship the nvidia module for each of our ~50 kernel variants; all major versions of the Nvidia module compatible with that kernel in fact.
The only possible way to access these nvidia kernel modules is via a certain kernel’s linuxPackages attribute set that contains all packages that rely on a kernel ABI such as kernel modules or packages like perf. That’s good design if you ask me but I’m obviously biased ;)