Working with the feature cache

In simple terms, the feature cache is a local copy of the dataset.

It enables an increase in performance - and from experience and observation of customers: it works well.
It speeds up map refresh and improves snapping on a large map extent. In extreme situations, where latency or response times from the database are too long (temporary network problems, ...), the feature cache is often the only way to still be able to snap.

These possible benefits are available when the feature cache is enabled.

As always, if you want to save, you have to invest something. Here it means that creating the feature cache takes time, of course.
In return, the search, the snapping, ... within the "boosted" map is much faster.
The larger the amount of data, the longer it takes to build the feature cache. Because at some point this is no longer worthwhile, there is the checkbutton for the minimum scale.

The feature cache is always available via the corresponding toolbar.
With auto cache you switch it to continuous operation. The effect is that the rebuilding of the graphic after changing the map extent takes longer than without feature cache, but many actions are then faster.

Whether the feature cache is worthwhile for an map extent depends on the situation. You can safely find out for yourself and gain experience.
Whether the auto cache (continuous operation) is worthwhile also depends on whether you prefer to work in a constant map extent (continuous operation could be worthwhile), or whether you frequently change the map extent (creating the feature cache could eat up the advantage in the actions).

The feature cache is cleared when the map extent changes. (And rebuilt if auto cache is used.)
The feature cache is also flushed when feature operations are "disturbed". This means actions like canceling the function and go back/undo.

For more information with screenshots, see the Esri documentation:
Working with the feature cache