There are two methods in use. Honeywell uses a world model (probably WG84-based) while Litton (now Northrop-Grumman) use (or used to use) a Swissair proprietary look-up table. Both produce the variation for a given Lat & Long.
Both manufacturers produce software updates that follow mag var changes but there will still be differences compared to whatever standard is used by the local NAV aid authority. As an example when we were doing HUD trials for CAT IIIa at our local field we had nearly 3 degrees difference between IRU and the Jepp chart ILS course.
Another way it manifests itself is that AHRS-equipped aircraft (that actually sense the local magnetic field) do not follow the same takeoff path as IRS-equipped ones - but the AHRS guys get the complaints from ATC.