Which LinuxCNC HW for 2026+?

When you get a following error, is there a dripped packet?
that is, is hm2_cardname.0.packet-error-total 0?

What are your following error limits?

Can you post a hal/ini file set for a system that exhibits these issues?

I ask because getting following errors and not network timeout errors, that is:
"error finishing read"s is odd and suggests a setup error of some kind

If your PC has a Realtek Ethernet interface, have you installed the DKMS driver?

(the stock Realtek drivers are hopeless latency wise)

If you have in Intel MAC, have you turned off IRQ coalescing?

Have you pinned the Ethernet IRQ to the last processor?

Some systems require disabling  EEE or PCIE power management als

PCW wrote:

Pining the Ethernet IRQ to a non-isolated core  would have the same effect
as not pinning the IRQ at all (especially if irqbalance is disabled as it seems
to be in the standard Debian LinuxCNC distributions)

Thanks, there is quite a few recomendations to keep it on a non isolated core including this one from Intel:
eci.intel.com/docs/3.0.1/development/optimizations.html
Their repo gave me nothing for Trixie.
I wonder if it should be on another isolated core and the servo thread pinned to a different isolated core?
 

My guess is that their suggestion is to improve latency  of everything except
Ethernet on their isolated core by keeping Ethernet IRQs off the critical CPU.


 

PCW wrote:

My guess is that their suggestion is to improve latency  of everything except
Ethernet on their isolated core by keeping Ethernet IRQs off the critical CPU.



 

Yes, some said that latency spikes from the NIC housekeeping etc could interfere with the RT task.

Kudos go out to all who made this possible:  
60000+ probes for now, during the last 24 hours, with an RPi5, 7c81 and the brand new 2.9.7.  
Zero issues, so far.    \o/

While probing, this thing now also runs a VNC server in the background,  
I am connected via VPN; it’s streaming music and also downloading a random  
10GB image file in a web browser.  

Pretty good latency too. Comparable to a decent RTAI setup, imho.  

Still some extra steps, as described here,  
  forum.linuxcnc.org/49-basic-configuratio...rol-is-choppy#332726  
and this "xset -r“ drives me crazy, bc I have to use the keyboard until I managed to get the handwheel working.

 

Thank you very much for your suggestions.
Seems like there is indeed a lot of fine tuning or trial and error necessary.

Even with limits set to 99999, this happens; and on one system, I also got the probe error, although my „probe flipflop“ was held in reset.

So far, my first working setup >2.8.4 is the RPi/7c81 combo,
but with that working, I hope to find some time to learn more about your proposals for the Ethernet solution.


Thx,
AS

Those figures look pretty good, where did you get your kernel from or did you build it yourself.