Pulseaudio used features of sound cards (most prominently the hardware read pointer) that ALSA/dmix alone never used.

ALSA/dmix could allow you to get the same power savings as pulseaudio if you set the hardware ring buffer size to, say, 2 seconds.

And that would be fine of you were just playing some music, but if you were also chatting and wanting to get prompt notification sounds they would always be delayed between 0 and 2 seconds depending on where the hardware read pointer happened to be when the system tried to play a notification sound.

ALSA/dmix could also allow you to set a tiny buffer size. Then your music would play, and your notification sounds would play promptly too. But if you were just playing music your CPU would never be able to go into the lower power sleep states because it would need to wake up every centisecond to service the tiny ring buffer.

That would kill your battery life.

Pulseaudio's (terribly named) "glitch free" audio feature was the first solution for Linux that allowed you to get power savings and low-ish latency. Your mp3 player filled up the ring buffer once every two seconds, and if a notification came in pulseaudio would look at where the hardware read pointer was, take the contents of the buffer that was just about to be read, and mix the notification sound into it, writing the newly mixed sound data to the buffer just before the sound card read it.

So, from the user's perspective nothing interesting seemed to happen, but they get better battery life and things like notifications or game sounds work like they expect them to.

ALSA drivers would commonly advertise support for accurately and precisely reporting the position of the hardware pointer, but since nothing actually used that info before, many drivers gave incorrect results, which would only cause problems when using pulseaudio.

Does pipewire interface directly to drivers or is it user-space magic on top of alsa like pulseaudio?