The problem

Ask anyone where their servers get time from and the answer is almost always pool.ntp.org, time.google.com or time.cloudflare.com. Those work, but they are not official time. The pool is volunteer-run machines of unknown quality, and Google famously smears the leap second, which means it deliberately serves time that is wrong by up to 500 ppm around a leap event — fine for a fleet that must never see 23:59:60, not fine if you want traceability to UTC.

The actual authoritative sources are the national metrology institutes (NMIs): PTB in Germany, NPL in the UK, NIST in the US, and dozens more. Each maintains its national realization of UTC — UTC(PTB), UTC(NIST), and so on — and many of them run public stratum 1 NTP servers fed directly from those clocks. The problem is that there is no single place where they are listed. Every institute buries its NTP hostnames somewhere different on its website, and half the “official NTP server” lists floating around the internet are stale copies of each other.

Except there is a single authoritative source, it’s just not very discoverable.

The source: BIPM

The BIPM — International Bureau of Weights and Measures, the organization that literally computes UTC — publishes an annual report with a Time Dissemination Services section, where each member laboratory declares what time services it offers, including NTP:

It’s a PDF, organized per laboratory, mixed in with details about telephone time services and radio dissemination. Useful as a reference, useless as an operational list. So back in April 2024 I went through it and extracted every public NTP server into a gist, and I’ve been maintaining it since:

List of Official Stratum 1-2 public Time Dissemination Services over NTP

How the list is built

The rules I settled on while extracting:

  • Public servers only. Several institutes (Australia, Luxembourg, Costa Rica, Panama) offer NTP but require you to request access first. Those are out — the list is for servers you can point chrony at today.
  • One row per institute, not per country. Some countries have several independent services (Argentina has three, Brazil, Poland and Belgium two each), sometimes at different strata with different access policies. The July 2026 rework converted the original per-country list into a per-institute table for exactly this reason.
  • Stratum and access policy in the notes. A few institutes run free stratum 2 while keeping stratum 1 restricted or paid (MIKES in Finland, Observatoire de Paris, ORB in Belgium, DMDM in Serbia). Worth knowing before you wonder why you can’t reach the stratum 1 hostname.
  • (unconfirmed) markers. Entries carried over from the 2024 report that the laboratory didn’t confirm for the 2025 edition are tagged rather than silently kept or dropped.
  • A changelog at the bottom. Time server hostnames churn more than you’d expect — see below.

There’s also a small extras section for well-known public services that are traceable to an NMI but aren’t the NMI’s own entry in the report: Netnod’s ntp.se (Swedish official time, NTP + NTS) and Brazil’s ntp.br.

What the 2025 refresh turned up

Re-checking the whole list against the new edition in July 2026 was the interesting part. Diffing two editions of a PDF by hand finds real problems:

  • Hostname churn is real. Poland’s AOS moved vega.cbk.poznan.pl → ntp.cbk.poznan.pl, Belgium’s ORB renamed both servers to ntp-public-1/2.oma.be, Sweden’s RISE retired the entire sptime.se domain, Vietnam and Ukraine changed IPs. Any list of NTP servers that isn’t re-verified yearly is quietly rotting.
  • Some services disappear entirely. Greece’s EIM now reports “No Internet Time Service offered”, and Brazil’s PADS-UFRJ dropped out of the report.
  • My own list had bugs. The Colombia row had Bosnian hostnames in it — a copy-paste error that had survived two years. Norway’s institute name was mistranslated as the “Correctional Service” (Kriminalomsorgen vs Justervesenet — one of these disciplines clocks). Maintaining a list means auditing it, not just appending to it.
  • South Korea’s public server is not what it seems. KRISS’s public ntp.kriss.re.kr is stratum 3 and cloud-hosted; the stratum 1/2 servers are not for public service. The report says so, most lists don’t.
  • NTS is spreading. PTB, BEV (Austria), DFM (Denmark), RISE (Sweden), ROA (Spain) and Netnod now offer Network Time Security alongside plain NTP. If your clients support it (chrony ≥ 4.0 does), there’s no reason not to authenticate your time.

Finally, everything got a reachability spot-check with sntp — five entries (ntp.inm.ro, time.saso.gov.sa, Vietnam’s 14.232.244.11, master.ntp.mn, sltime.org) didn’t answer from Western Europe at check time. They’re kept, noted, since some NMIs filter by geography (USNO’s servers, for instance, only answer from North America).

Using it

Pick three or four institutes near you, prefer NTS where offered, and you get an upstream set with known provenance. For a server in Western Europe:

# /etc/chrony/chrony.conf
server ptbtime1.ptb.de iburst nts
server ptbtime2.ptb.de iburst nts
server ntp.metas.ch    iburst
server ntp1.npl.co.uk  iburst

Four sources across three countries, all stratum 1, all fed from national realizations of UTC — and if chronyc sourcestats shows one of them drifting, you know exactly which institution to raise an eyebrow at.

The gist gets refreshed when the BIPM publishes a new edition. If you spot a dead server or a missing institute before I do, leave a comment on the gist.