Mesa hm2/hm2_7i96s.0: error finishing read

I have added a total soft error count pin to hm2_eth as a debugging aid:

hm2_CARD.0.packet-error-total

PCW wrote:

With 5.10 on my Elite 800 G1, I get only rare spikes, maybe once ever couple of days, judging from 
the tmax numbers. There is definitely video  interference but no huge spikes like you are getting.


 

Thats interesting, What spec is the Elite 800, is it one with discrete graphics? as there looks to be quite a lot of model options.

Its a Elite 800 G1 with a Intel 4570S (2.9 GHz) CPU and integrated Intel graphics

Strangely enough the 4.19 kernel has similar performance as 5.10 on the 800
but 4.19 is _much_ better on my 8300 (allowing 4 KHz servo threads)

Probably out of left field but was comparing notes off forum with another user who has used the Odroid H2+ like me. This was discontinued due to chip shortages but has now been replaced with the H3 and H3+
He pointed me to this post on the Odroid wiki demonstrating the kernel drivers are 15-30% slower than the native realtek driver. He said that he experinced following errors while jogging with the 5.10 kernel which was resolved after compiling the realtek 8125 driver!
wiki.odroid.com/odroid-h3/hardware/insta...net_driver_on_h3plus

I think its time to get a couple of Odroid H3 PC's. JUst not sure on the difference between the H3 and H3+ and if the different CPU is necesary for LCNC.
www.hardkernel.com/product-category/odroid-board/x86/

He pointed me to this post on the Odroid wiki demonstrating the kernel drivers are 15-30% slower than the native realtek driver. He said that he experinced following errors while jogging with the 5.10 kernel which was resolved after compiling the realtek 8125 driver! wiki.odroid.com/odroid-h3/hardware/insta...net_driver_on_h3plus
 

Thats interesting, i did try the r8168-dkms driver on 5.10 but it did not seem to make much difference on my hardware.

I have also now tried the 6 series kernel that is in experimental and that has same issues on my hardware, (yet to test on all 3 boxes).

I am compiling a spreadsheet of all my testing which i will share soon, I want to try and find some kind of pattern. I think next steps is we need to take a case to the preempt_RT patch maintainers and see if they can shed so light on this, issue will be i doubt any of them use linuxCNC or are likly to have a mesa ethernet board to reproduce the issue, I dont know enough about the workings of the preempt rt kernel to know how else we can compare/profile the versions in a way that might be relevant to this issue.

I don't seem to have an issue on my 8300 USDT and a 5.18 kernel
(This is after a couple hours with a youtube video running)