linuxcnc 2.9.2 (live) on the intel n100 cpu

Once an inspiring innovator and corporation, I really looked up to them, down to I-don't-know-what the fuck they are doing. Prolly changed silicon for papier-mâché for their np transistors their switching performance is degrading faster than their back pocketed politicians changes their mind.

"It sucks? no worries, increase variants of the same and just change the core naming from cherry to wazoo" ~Intel

That's not what I get with my N100 Firebat T8 Pro Plus, see below
HT and C-states are disabled.
Link to product, I have the N100 16GB version: aliexpress.com/item/1005005234874016.html

Edit, sorry the N100 doesn't have hyperthreading, confused it with another cpu I was working with.
 

Ah. Knew something was up... nothing made any sense at all. Thanks for posting Hakan, u saved another of Intel's badday lol

What I can spot as different vs gambaf is that I run debian off the internal nvme drive, not the live debian.
The only real modification I have made is to increase the megahertz for the memory, it has got DDR5 memory at 4800 MHz but is set much lower i bios. I have also looked around in bios to reduce turbo and power states, but there are not that many such settings. I might have touched one or two such setting but I can for sure not find back which one. The C-states are disabled for sure, that made a big difference. isolcpus=1,2,3 made a difference as well, it has got four real cores.

I'll place an order for a J6426 and a N100, test both, and return/refund the worst. Not entirely sure why I'm being recommended a J6426 over the endless list of other four-cores, neither asked why, but I take it from a seller who specialize in industrial automation HMIs, processors and what not (including linuxcnc).

I'll update.
(man I'll miss that decade long luv affair with that ice-age core-duosaurus)

Mecanix wrote:

Not entirely sure why I'm being recommended a J6426 over the endless list of other four-cores.
 

Obviously has to do with Intel's nothing-fancy core naming sounding like a cool exo-planet. Bet it smells roasted rose petals when glxgears are crunching the silicon over others too (insert rolleyes meme here).. who knows.

**No Affiliation, zero recommendation here, based on luck, waste of time, and money - likely. At your own risk.**

Hakan, could the difference be that you didn't "stress" your machine while testing? And by stress I mean a few glxgears and opening up firefox to play some hd yoututbe in the background.
With all the Intel hate going on here, I'd expect people to compile the kernel in the background with 16 threads, but instead you're letting them off easy with just and idling pc?
I did install linuxcnc on the nvme drive after the initial test on the live version so I could set cpu isolation, which I don't know if you can do on the live version.

When I just launch the histogram without any additional tasks running (glxgears, firefox/youtube), I get similar results to you.
All the tests were done with the following command line as requested:
latency-histogram --nobase --sbinsize 1000 --show

Modified BIOS, under stress, isolated cores:


Modified BIOS, idle, isolated cores:


Default BIOS, isolated, idle:


Default BIOS, not isolation, idle:

 

This is an unreasonable load for linuxcnc but here we go. 9 glxgears, youtube streaming, recompiling a project with files on a network drive.
Youtube video stutters a bit because of the load, so it is at least a bit loaded.
Still inside +- 15 usecs.
 
gambaf, you use isolcpus=2,3 so you still can get that sort of delays, use isolcpus=1,2,3