I hear myself a fraction late through the software and cannot play in time, what is the actual fix?

You are missing a feature rather than a setting, and it removes the trade-off entirely: direct monitoring.

Almost every audio interface can send the incoming signal straight to your headphones in hardware, before it ever reaches the computer. There is no round trip, so there is no delay - it is effectively instant regardless of your buffer size.

It is usually a knob labelled something like mix or blend, or a button labelled direct or monitor. The knob blends between what is coming into the interface and what is coming back from the computer.

How to use it:

Turn the blend toward input so you hear yourself directly.

Mute the track in the software or turn off its input monitoring, or you will hear yourself twice: once instantly and once late, which is worse than either.

Raise the buffer as high as you like. Once you are not monitoring through the computer, buffer size stops mattering for playing. This is the part people find counterintuitive.

The one thing you lose is hearing effects on your voice while you sing, since the direct signal is dry. For most people that is a fair trade, and there are ways around it if reverb helps your performance.

The habit that follows: use two different buffer settings and switch between them.

Low while recording, if you must monitor through the software - for example when you want to hear an effect. Small buffer, fewer plugins active, accept that the machine is working hard.

High while mixing. Latency is irrelevant when nobody is playing, and a large buffer lets you run far more plugins without crackling.

Most software lets you change this in two clicks and a lot of people never touch it after installation, then conclude their computer is not powerful enough.

On the crackling at low buffers specifically, since that is your other symptom: it is not always raw processing power:

Plugins on the input path are the usual cause. Anything on the track you are recording has to be processed within the buffer window.

Wireless and power settings cause more audio dropouts than people expect. A laptop on battery in a power-saving mode will crackle at buffer sizes that are fine when plugged in.

Sample rate matters too. Recording at a very high rate doubles the work for no benefit most people can hear.

One more thing that catches people once direct monitoring is sorted, because it produces a similar-feeling problem with a different cause.

If you are playing along to already-recorded tracks and something feels slightly out, look at plugin delay compensation. Some plugins - particularly lookahead limiters and certain analysers, introduce delay, and the software compensates by shifting everything else. That is correct for playback and it means what you hear while overdubbing can be shifted relative to what you are playing.

Symptom to recognise: takes that felt right sit visibly early or late on the timeline afterwards, consistently.

What to do: turn off or bypass heavy plugins on the master bus while tracking. You can put them back for mixing, and the take will land where you played it.

And the general principle for the whole tracking stage: make the monitoring path as short and as boring as possible. Direct monitoring, minimal plugins, nothing clever on the master. Save all the processing for when nobody has an instrument in their hands. Nearly every latency complaint I have seen is somebody trying to mix and record at the same moment.