Thanks. Looks like the default digitalWrite() has too much overhead. I just tried measuring the uno with the OP code and I topped at 116kHz. The nano that I measured before topped at 70kHz. But seeing there are many different timers I guess it depends on which pin too?
For the sample code you supplied using the TimersHelpers.h library, I downloaded it and tried the sample code. There was an error on debug saying TImer0 is not defined?
If I want to create 50% duty cycle signals at the highest frequency possible (essentially just a normal periodic square wave). Can that be done up to 4MHz (you mentioned 4MHz can only choose between 33% or 66% duty)?