Raspberry Pi 4

After some torture


I guess that's it then.

I am all for improvements. Checked the link. We both use the RaspberryPi kernel patched with the official Preempt RT patches. The link uses CustomPi, I used standard raspbian. My immediate thought is that there should not be much of a difference.

One thing I know is that the sd-card I used during these tests was amazingly slow. I will retest early next week with a much faster sdcard, perhaps that slowed things down. Apart from that I can just say I was hoping for a better result and I am willing to test good leads.

Here is a little q&d-howto to install a rt-patch :

erste.de/raspi4-RT-kernel-howto.html

I will also document the needed steps for compiling the linCNC-source on
the RP4 the next days.

compile, latency-test and start of lincnc 2.9 on a PI4 :

erste.de/raspi4-RT-kernel-howto.html#compile_linCNC_on_Raspi4

one axis test: lincnc 2.9 on RPI4 with
Linux raspberrypi 4.19.50-rt22-v7l+ #2 SMP PREEMPT RT Sat Jul 13
09:21:03 BST 2019 armv7l GNU/Linux

a stable 5kHz step/dir signal is generated.
as a quad-signal 10kHz should be possible.

further tests in a 4-axis system will follow.

I tested today again with a new sd card. Well it was better but didn't really solve anything.

If I run latency-histogram for under 1 minute I get similar looking graph like wicki's.
The previous graph I posted ran for 1.5 hour with me surfing and 7? glxgears.

I can't see there is a real difference otherwise. I used the 4.19.57 kernel with the RT-patches for 4.19.50, but it shouldn't matter, the RT-patches don't change that much and the patching was clean with just a few line offsets. wicki used 4.19.50 for kernel and rt-patches. There is a difference of course.

wicki what does the latency-histogram look after, say, 10 minutes?

And does it work running linuxcnc, that is of ocurse the interesting question.
wicki, if you are busy testing I think I'll wait for your results.