Asrock Q1900M: Unexpected Realtime Delay, but Good Latency Values

So i bought a Asrock Q1900M to replace a failing c2d machine, since i have this board in a nas and it gets good latency values.

Unfortionatly despite the good values, see attachment, running linuxcnc gets an instant Realtime Delay error. I am running a 45us base period and a 1ms servo period.

The test below was run with some glxgears 2 dd's for io and a wget to load network. I cant seam to repoduce any latency excursions with the latency tests.

Any ideas?

Did you disable hyperthreading, TPM, power saving, pci-e power management, turbo, etc in BIOS?

the J1900 dosent have hyperthreading, nor dose the board have tpm, i disabled all the low power c states, onboard audio, the igpu (machine is headless).

If i could somehow repoduce the latency excursion in latency-test/ or the histogram that would be great, but i cant get it to happen.

You can use latency-plot to monitor latency.
Headless, that might easily cause latency issues, i would never use that on a production machine as it adds several layers of potential issues.

latency-plot is really just a different way to query the same information, and it looks great too.
The problem being that all the tools see ~4us jitter max no matter what i do but linuxcnc itself fails instantly.

I dont think headless is any problem, the previous old c2d machine was run headless too.

If close and run LinuxCNC again, does it show the warning again?
If yes, most probably something with the video subsystem going over network causing delays.
Very hard to pinpoint, but try running LinuxCNC from a terminal, should have more info.

Yeah Instantly anytime linuxcnc is run.

I also just tested locally (not headless) and the result is the same.

Did you try from terminal? Is there any useful info there?

theres no usefull info in cmdline output.

I reimplemented latency-test directly into linuxcnc using the timedelta hal module and pyvcp.
And now i can see that the latency excursion occurs once exactly every 60 seconds but only while linuxcnc/axis itself is running never when only latency-test or any of the other latency testing programs are run.

linuxcnc just has to be running, it can be totaly idle and the excursion will occure.

Very strange and i dont know what to try from here

The problem with the latency tests is that they just test dispatch latency.
it's likely what you are seeing is some kind of resource contention.