GPS accuracy is also affected by resolution, as is any analogue to digital conversion. Resolution here comes in two main flavours, firstly the number of satellites that have sight of you (pretty sure full resolution here is still restricted to US Military and everyone else gets at best 50%, often less) and secondly sampling frequency. Sampling frequency is how often the GPS checks for satellites and (obviously?) the longer the periods between ‘knowing’ ‘precise location’, the more time it’s just guessing where you probably are based on the last thing it ‘knew’ and your direction and speed as calculated from the pings to date.
Increasing sample frequency is a major battery hog, and most devices are trying to balance acceptable accuracy vs desirable run time. Some devices give you control here and allow you so specify sampling frequency and therefore increase accuracy at cost of run time (Suunto watches I think still do this) but most leave you in the dark about it.
When correctly calibrated and operated, the direct reading analogue approach is always more accurate, but as demonstrated further up the thread, GPS is likely more accurate than a rotation and frequency counter that hasn’t been correctly operated or configured. In a situation such as a car, where it’s far less likely you can control the speedo configuration, I’d probably trust GPS over dash but wouldn’t expect either to be ‘correct’.