And every one of those slowpokes is camped out in front of me!! 
That’s just a guess.
I would say that your GPS would be the most accurate - calibration of speedometers aren’t perfect and of course some tires mean that the speed on your cars speedometer is more out as you get faster.
The sat nav / GPS will pretty much always be the most accurate.
It may vary with device - dedicated device vs smartphone app vs built-in in car- but many GPS devices do an element of dead reckoning. (If you don’t know the term, you can google it).
It’s why it doesn’t go ape-crap when you drive into a tunnel, under an overpass, down a street with tall buildings, or even drive around on very cloudy day, any of which will disturb the signal reception.
To do accurate dead reckoning, it needs to keep an accurate running track of speed data. I’d personally shade my preference to the accuracy of the GPS device for that reason.
You can be so mean… Bawahahahaa 
My GPS (Garmin) will sometimes show me going the opposite direction I’m actually moving for a bit after a turn.
I.e., it doesn’t even have the sign of my velocity right.
People put way too much trust in GPS’s. Speed is even less reliable. They aren’t nearly as accurate as people think.
My car’s speedo and odo both read about 3% high. I’ve tested it with two different GPS units plus have done the ‘stopwatch between mile markers while holding a steady 60 mph’ thing.
I seem to remember that legally odo’s can be +3% to -1% off. (I may have the + and - switched.)
I think Honda was caught a few years ago optimizing the +3% so that a 36K warranty would actually run out at about 35K miles. My Kia has a 100K warranty so it will be more like 97K actual miles when it runs out.
I disagree.
What model & age?
Where & how is it mounted?
Do you pay any attention or number of satellites persent in it’s view and their location?
Been flying with, driving with, walking with many different brands & types, did land surveys when there were only 4 satellites available to civilians and they still had ‘jitter’.
Only time I ever had a direction indication wrong was after a quick stop and going in reverse for some distance over inches or making as tight a turn as possible in bad conditions of obstructions ( trees, buildings, lack of available satellites etc. )
I have not used every make & model of everything since day one and in spite of all the bad things Garmin has done over the years, going the wrong way I have never seen.
Steady speed in reasonable conditions is about as good as can be at that cost point.
Maybe you are just unlucky or?? I don’t know but a generic blanket warning of today’s units for auto speed is IMO not correct.
I have no opinion of the built in stuff in cars, who makes them, etc. I can’t afford nor want vehicles of that cost. I have no clue but you did say, Garmin.
If you are a pilot and trust anything totally, every time under every condition in regards to navigation, you are a dead person rushing to your death with all possible haste.
IMO.
But if the speedometer is incorrect, won’t the odometer be incorrect too? Seems there’s an easy way to test that–just drive a known distance, never mind the speed. If the odo has a different distance than the known distance, your speedometer must be off too.
On a few of my vehicles that I’ve tested, while the speedo might be off, the odo is dead on. They’re just electronic calculations these days. My last motorcycle had an adjustment field in the firmware that could change the speedometer reading on a percentage basis, but that setting didn’t affect the odometer.
On older systems, the od was a simple geared system off the transmission or front wheel of some motorcycles. The speedo needle gauge incorporating a spring and spinning induction drag system. 2 places that require very delicate and fine adjustments mass produced by the millions of millions. Kinda splains it for me.
New digital systems are magic to me.
Has digital or GPS replaced the rear bumper calibrated bicycle tires of hard core rally meets??
Just a small thing to watch out for. Depending on how your GPS does it’s internal calculations, displayed speed may not be accurate when going up or down hill.
The basic GPS calculates position on an idealized version of the Earth’ surface, the geodesic. This does not take into account local thing like mountains 
Of course the GPS can determine your height above the geodesic i.e. altitude, but this is notoriously inaccurate. So even if the internal speed calculation is more sophisticated and uses the rate of climb to allow for the driving up the “hypotenuse”, it may still be out a bit.
So perform your calibrations and comparisons on a reasonably level section of road.
Unless you’re on a 4WD jeep trail in the Rockies, the slope of the road is not likely to be severe enough to introduce a very large error. you can find sections of 8% grade on I-70 west of Denver, i.e. this is pretty steep but also pretty rare (you may find steeper grades on some obscure streets, but not on stretches long enough to assess or care about GPS speedo accuracy).
For an 8% grade, your horizontal velocity is 99.7% of your velocity along the road grade.
Not necessarily. If the odo is calibrated to be accurate to +/-2%, and the speed is calibrated to be accurate to +0%/-4%, then there’s plenty of room for them to be different.
If the GPS uses doppler - which AFAIK they do, they work out the vector, and thus can give you a correct speed. GPS also provides a direct altitude (which is relative to the geoid, but not as accurate as the lat/long) so it can still calculate a proper speed based upon 3D coordinates. But what an individual unit does is perhaps another matter.