Raspberry Pi 4

> If you haven't done so already, bring down the wlan0 interface. It might help.

A good hint!
I have done it and now I can do go down to a period of 140.000 in linCNC with
display via vnc.

Inside a "hal-only"-environment, it's possible to drive a stepper with 10kHz
with a step/dir-signal. But I think the timing quallity of the signal will be poor.

lets see

10kHz stepper signal - not as bad as expected. But some deviations in timing....
The above spikes are the durations of the "low" and "high" parts oth the squarewave.
not even so bad.... but if I zoom it:


erste.de/timing-tests.PI4.html

I works well, but has some stability issues. This might be due to thermal issues with the Pi, we have no heat-sink or fan.

andypugh that would be correct, from my experience using a Pi3 with Mesa cards you definitely need an heat-sink
and a fan to keep the cpu temperature as low as possible to avoid the automatic throttle down.
The throttle down mechanism makes the cpu run slower and (perhaps?) suspend execution for short amounts of time
in order to keep the temperature under control.

Also I advise using cpufreq to set the power governor to performance and use maximum clock speed, but this will only give you
any advantage if you're using a heat-sink and a fan to "externally" control the temperature.

I didn't see any stability issues in my testing. No fan or anything, it was just laying on the shelf.
It did use a 5A power supply, although I measured the current to be below 1A, typically 0.85A.

The system will sleep/hibernate/suspend/shut down after "a while" if left idle with no user input, this cured itNow the screen stays on the whole night. And the wired network interface stays up.

wicki does isolcpus=1,2,3 in /boot/config.txt help also for the base tread? It helped a lot for the servo thread.

Power management, speed stepping, virtualization etc will have similar effects on latency as on x86. It's just that ARM will have all those settings in completely different places than x86 where it's set in BIOS.

> does isolcpus=1,2,3 in /boot/config.txt help also for the base tread?

also a nice idea!
now a period of 80.000 ist possible.

analyzing the squarewave is a little bit confusing me:
__--___--____--___

the high-level-timespan is always 80usec (between 78.5 and 81)
thas OK.

but the low-level-timespan differs much more. (beween 160usec an 320usec)
the time-period... thats not nice....
maybe, with a quad-signal instead of step/dir it will be better. but then I need
a quad2stepdir-converter...
an arduino?....
*thinkig*.....

Looks like heat is going to be an issue with the Rpi 4. I noticed mine running pretty hot out in the open with just convection cooling and no heat spreader.

This is easily and cheaply solved with a heat spreader or with a bit more expense and lower reliability with a fan. Proper enclosure design for long life will be important. Heat spreader connected to metal enclosure will work.

There is already a long discussion on the pi forums about this (if you can stand reading it)
www.raspberrypi.org/forums/viewtopic.php?f=28&t=243500

wicki wrote:

>
now a period of 80.000 ist possible.

What is a possible, usable step-rate?

I'm not sure what you are measuring yet. Are you measuring all the duration of all the intervals between pulse edges falling and rising, just rising or just falling or?