Apparent difference Gecko G213V -- G214 ?

I have a XYZ mill under construction. For over half year now.

The machine uses:

LinuxCNC 2.7.8 under Debian Wheezy 32 bit rtai
5 mm ball screws with single nuts
3 nm at 4.2 A stepper motors, 200 steps/rev 0.9 Ohm 3.8 mH
Gecko G213V motor drives setup at 10 microsteps.
Unregulated power supply comprising transformer, rectifier, capacitor. Voltage 68V idle, 65V at 10A
Mesa 6i25 port interface.

Step and dir inputs to G213V are driven directly from 6i25 stepgen output pins, each with a 350 mm long twisted pair with ground.

I have struggled with this setup losing steps, it seems to lose about 10 microsteps each direction reversal. Equal on X, Y and Z.
Stepping works for some 30 seconds after system powerup, then position starts to slip.

Using a dial indicator to verify movement with this test program
Whatever I have tried so far has not changed the situation, no better, no worse.
* speed and accel change by a factor 10
* step time by a factor 10
* dirsetup dirhold by a factor 10
* total rewiring, step/dir signals on twisted pairs with ground
* pullup resistor to 5V
* inverted step pulse polarity
* small capacitor on signal pin
* RC filter on signal pin
* New motors
* double check linear mechanics for slippage or binding.
* Different power supply.

Today, in desperation, I exchanged one G213V to a G214 driver board.
The G214 does not have any problem in my setup!!! I ran the test program for three hours without problems.

So, what is happening? It is unlikely that i have three G21V that have component broken in the same way.
What differs in input characteristics between the two boards?

Buying two more G214 would solve my issue, but I still hope my G213V can be used. Maybe driving step/dir pins directly from a 6i25 is the issue?

Step polarity? If this was wrong, the setup/hold times would be violated

I have tried with inverted polarity too, forgot to put it into my list.

Since the G213V uses active high step/dir signals and the FPGA cards cannot drive to 5V,
have you tried stiff pullups on the step/DIR signals, say 270 Ohms or so?

The G214 is active low so will get very close to 5V drive but the G213V will only get 3V or so
without a stiff pullup resistor

I have tried 4.7 kOhm pullups, no difference.
Will try with 270 Ohm.

Now I have tried with 270 Ohm pullup to 5V. No improvement, not worse either.

Then I tried with a buffer CD4050B between 5i25 and G213V. No improvement, not worse either.

Then I changed G213V resolution from 10 microsteps to 5 microsteps. It seems to resolve the issue!!
I was able to increase max velocity from 10 mm/s to 50 mm/s so it is not the reduced pulse frequency that is the fix.
Any thoughts about this?

pretty weird
If the 5 ustep mode works that suggests its not a level/timing issue
and really sounds more like a firmware bug in the drive