Re: [chrony-users] PPS refclock on Macs

[James Clark <jjc@xxxxxxxxxx>]

On Thu, Jun 5, 2025 at 11:36 PM Rob Janssen <chrony-users@xxxxxxxxx> wrote:

Do these not have a suitable Intel network controller (225) that allows PPS input?

Modern Macs (other than $7,000 Mac Pro) are sealed boxes, with no IO ports other than Thunderbolt/USB ports and a headphone jack.  It is possible to get a Thunderbolt expansion box with a PCIe slot (e.g https://www.sonnetstore.com/collections/thunderbolt-expansion-systems/products/echo-express-sel-tb3). Into that you could plug an i225-T1 or i210-T1, which allow PPS input. And Apple does have a driver for this (see https://khronokernel.com/macos/2021/11/22/PCIE-ETHERNET.html). But the key problem with this strategy is that macOS has nothing like the PHC subsystem in Linux, so there is no way to get PPS timestamps from the SDP pins.

Otherwise, when you need something better than toy accuracy it is likely better to get an NTP server
(e.g. from ALLNET) and sync over the network.  On a local network that performs better than anything
that involves USB.

One might think so.  But it turns out you can get much better than toy accuracy: the approach using audio pulses achieves a tracking RMS of less than 2µs, although you still need a local NTP server to calibrate the offset, if you want better than 1ms accuracy. This is because (a) USB audio is using isochronous USB and (b) macOS has kernel support for timestamping audio samples.

James