I set out to research the limits of clock sync precision, effectively chasing the "Last Nanoseconds to UTC."
Is sub-nanosecond UTC sync possible? Literally, no. But practically yes.
This presentation sums up my research on the topic. Having retired from work in financial markets, it focuses on how market participants can use GPS or other GNSS satellite constellations to synchronize their clocks.
In a nutshell, two factors make it literally impossible to sync clocks with precision better than about 5ns:
- The best of the GNSS constellations still have a
standard deviation of 5ns around UTC.
Look for
sigma_GPSin a recent Circular-T from BIPM to check me on this. - A national time lab like NIST may have atomic clocks that can do better, but UTC is a virtual timescale defined after the fact, not in real-time.
So if you're getting time from GNSS instead of from a national lab, you can't literally get closer than 5ns.
However, my homelab tests showed that two high-quality GNSS receiver modules (u-blox F9T) could agree on time to 550ps when tracking a single constellation. So even if that constellation was running ahead of or behind UTC by 5ns or more, clocks synced to it could still agree sub-nanosecond.
This is the state of GNSS clock sync as I see it in late 2025. There are promising recent developments that I haven't tested yet:
- Galileo has added HAS, which may bring down their standard deviation.
- Fugro AtomiChron promises sub-nanosecond sync to their own traceable reference timescale.
I delivered this presentation at the London STAC Summit and New York STAC Summit.
My presentation style is heavy on visuals and light on bullets, with my narration giving the key points verbally. The GitHub audience necessarily misses out compared to the live audiences. However, the live audiences didn't have enough screen time to savor the final slides on Takeaways and Best Practices.
You can view the presentation PDF here on GitHub or Download it for offline viewing.
The tools I used for my research are in another repo.