Installing LinuxCNC On HP T610 Thin Client

There are quite a few differences in setups.

For instance the kernels are different, OS are different, WIndow managers & Desktop environments are different, even the kernel command lines are different.

Actually the differences are so much so that it is no wonder there are differences in latency figures. You are not really comparing apples with apples. There is nothing really strange about it.

The latency histogram clearly shows you are running a base thread.

Hi

ALittleOffTheRails wrote:

Have you got any testing to back that up ?

I’ve seen extremely great results with an dual core Athalon with 4Gig ram running the latest Buster ISO. Yep this is a 2008 or there about setup. The test was also with a base and servo thread.

I feel this is getting into “Windows Guru” territory by just proclaiming “add more ram” to improve latency.

It would be interesting to see your test results with various ram quantities on the same setup.
 

2 GB has been too low in many cases. I am No windows Guru. Last Windows install on a PC owned by myself dates back to about 2010. 
If, then please play with amount of ram yourself and post results in a seperate thread. You WILL see a considerable difference, once there is ENOUGH Ram. I use old PCs for my CNC. If I have to add Ram, I have the bad habit max it out and forget about that issue. Costs little, brings a lot => My opinion.
Greez
chris

That's not how things work. It's up to those who make the claim to prove the point.
So no I'll think I'll pass trying to prove your point.

But what I have done is bought one of these little buggers to play with and see what I can get out of one.

As for the Windows Guru thing.

The usual answer to anything in windows is "apply all s\w updates, install latest drivers" not matter what the issue. I dare say that would be the answer if your windows pc burst into flames whilst powered off sitting in the middle of a grassy meadow.

A couple of suggestions to look into.

[1] Consider adding idle=poll to the kernel command line.   I decided not to use this as it makes things run very hot and keeps it that way even when no load in place.  However if you have good cooling it might not be a concern.

[]2 I used a "hog" script to drive the cpu hertz on the cpu 1 (not cpu 0) to max Hz.  A quirk of the AMD arch is that this will also drive CPU0 to max Hz as well when CPU1 is driven to max.  The effect of this is that you will likely see improved latency,

See forum.linuxcnc.org/18-computer/39533-ins...ient?start=20#206996 for these.

I used path #2 as it means that when you shut off linuxcnc it will drop the cpu load and give things a chance to cool. Depending on your usage pattern this may  or may not be useful.
But with the setup i had (4GB ram, Miny 19.3 and the RT kernel noted in the posts) this hog script approach was FUNDAMENTAL to getting acceptable latency.

Cheers - James.
 

Joco's latency script was run with no base thread

For completeness here is the servo only latency-histogram - as can be seen very little difference to Joco's script and base thread 

Thanks James - I tried the idle=poll method and have a servo around 193.
Ran the hog.sh script and i get the following in the terminal window -
pid 3745's current affinity mask: 1
taskset: failed to set pid 3745's affinity: Invalid argument