Pressure Advance for direct drive

Is it worth using Pressure Advance on Direct Drive systems, I have an E3D Titan, if so what setting would be best?

Try it and see. I expect it will help, but you will need a low value of pressure advance, maybe in the range 0.02 to 0.05.

Running a wadederivative direct drive + RRF1.18beta2 and printing at around 50mm/s, I saw a noticeable improvement last week with pressure advance values in the range dc42's suggesting.
I'm currently running M572 D0 S0.025 in each gcode startcode block  i.e. not in RRF config.g. I read a good reason for it, but forget. Anyone care to remind?
The nice thing about tweaking pressure advance is M572 can be issued while printing, making it super easy to experiment  thank you dc42
I started with the rather extreme value of S5.0 just to check pressure advance was visibly functional. That resulted in a crude chamfer (approx 2mm) being applied to external square corners, S0.05 resulted in very slight but noticeable rounded corners and right now I'm happy with S0.025.IIRC S0.05 actually made for 'nicer' external perimeter corners than S0.025, but also resulted in gaps beginning to appear between infill and perimeters/outlines  perhaps more perimeter/outline overlap might solve this (currently its 15%).

@hairy_kiwi  do you have photos of printed parts with and without M572?

What kind of model is good for testing pressure advance?

What kind of model is good for testing pressure advance?
I'd guess the effect would be more noticable with longer print (extrusion) moves between changes of direction.

@hairy_kiwi  do you have photos of printed parts with and without M572?
No, sorry. I'll post some when I get a moment.

What kind of model is good for testing pressure advance?
A solid infill 20  30 mm square x 3 mm height block ought to be of sufficient size to visually assess all the nuances of changing pressure advance. You may even need less zheight than 3mm to reach a conclusion. I would also suggest using the halfsplitsuccessiveguesstimatetechnique* to derive your preferred value, rather than changing the value in small increments; you should find it much easier to see any changes that way. Until last week I was using a single 50 x 50 x 1.8 mm block with 0.3mm layer height and solid infill while setting up my machine. I found it a useful size block for checking bed flatness, Esteps/mm, jerk and acceleration and did also use it for setting pressure advance, but its size is probably overkill if you only want to set pressure advance.
*There's bound to be a bona fide mathematical process name for what I'm describing, but I can't find it so I've mashed one together: the halfsplittechnique (from electrical faultfinding) with successiveapproximation (from mathematics, for finding roots of equations). 'Approximation' is further replaced by 'guesstimate', because at the end of the day it's all a somewhat subjective process.
Good luck!Have fun 
Thanks both, I'll start experimenting.

*There's bound to be a bona fide mathematical process name for what I'm describing, but I can't find it so I've mashed one together: the halfsplittechnique (from electrical faultfinding) with successiveapproximation (from mathematics, for finding roots of equations). 'Approximation' is further replaced by 'guesstimate', because at the end of the day it's all a somewhat subjective process.
I think you mean binary search. I remember using this technique when i was a student, to determine the appropriate amount of chilli to use when cooking chilli con carne. It only took me 3 meals to get it right.

What kind of model is good for testing pressure advance?
*There's bound to be a bona fide mathematical process name for what I'm describing, but I can't find it so I've mashed one together: the halfsplittechnique (from electrical faultfinding) with successiveapproximation (from mathematics, for finding roots of equations). 'Approximation' is further replaced by 'guesstimate', because at the end of the day it's all a somewhat subjective process.
Good luck!Have funBack when I was a bench tech we called it "divide and conquer".
Good job with the naming convention!
I looked it up in a class training book from the '70's that I still have ( I'm not a pack rat ) and it was called the "halfsplit rule" there. 
further Q on this, what E Jerk value would be a good starting point, its currently set to 20. I run a 0.9 deg stepper on the titan extruder

I think you mean binary search. I remember using this technique when i was a student, to determine the appropriate amount of chilli to use when cooking chilli con carne. It only took me 3 meals to get it right.
That's it dc42! Many thanks for enlightening