>The key hardware fact: The CM4 uses the BCM54210PE Ethernet PHY which has full IEEE 1588v2 hardware timestamping support. This is what makes sub-microsecond PTP possible on the CM4. The BCM54210PE timestamps packets right at the wire inside the PHY, before the packet touches the kernel network stack. This eliminates the jitter that comes from kernel scheduling and interrupt handling.
Very cool feature I wasn't aware of because I'm not a true Chime Head.
RevK's recent stratum 1 server hardware [0] is worth a look. As are the notes [1] that someone wrote up after using it in the UK NTP pool for a couple of weeks.
I saw guides for how to do it with the pi5, but I couldn't get it to lock to the pps reliably and chrony said I had way more jitter on pps and nmea than there should've been. Might try following this one minus the difference in pps
are you trying to use gpio for the pps? feeding the pps into the NIC clock is what gets you the great accuracy, doing it with gpio on the main processor will have much worse jitter
Obviously you really want a tank/vacuum master clock, but they are expensive and big. this GPO Type 46 however is small and more importantly gives off a pleasingly happy ticking noise. (tickety-a-tock, tickety-a-tock, then every 15 seconds a Clonk as the magnets fire)
Very cool feature I wasn't aware of because I'm not a true Chime Head.
[0] https://www.revk.uk/2026/07/time-for-something-new.html
[1] https://codeberg.org/RevK/ESP32-GPSNTP/src/branch/main/RevK%...
Just kidding. Learned a bunch here after I understood the actual topic.
I will now commence shitposting, so please look away now.
Booo use a pendulum like a real time lord: https://gentsofleicester.com/project/alanb/
Obviously you really want a tank/vacuum master clock, but they are expensive and big. this GPO Type 46 however is small and more importantly gives off a pleasingly happy ticking noise. (tickety-a-tock, tickety-a-tock, then every 15 seconds a Clonk as the magnets fire)