Internet speed test

What the four numbers mean

Download and upload are throughput: how many megabits per second your connection can move in each direction. Download is the one an ISP advertises, and on most home connections upload is a fraction of it.

Latency is how long a round trip takes — out to the server and back — with no data to speak of in either direction. It is what makes a video call feel immediate or awkward, and it has almost nothing to do with your download speed. A satellite connection can be fast and still feel slow.

Jitter is how much consecutive round trips varied. A steady 200 ms is usable; 40 ms that keeps jumping to 300 ms is what makes voices break up and games rubber-band. It is measured here as the average difference between one trip and the next, so a perfectly steady connection scores zero.

What is measured, and against what

The test runs against the Cloudflare datacentre nearest you, and the page tells you which one. That is deliberate: the number you want is what your line can do, so the far end should be as close as possible. Every byte counted crossed your connection and nobody else’s.

This also means a slow result is a fact about your connection or about the path to that datacentre, not about a distant server having a bad day. If you want to know how a particular game server or website feels, its own distance is a separate question — one this test does not answer.

Getting a result worth trusting

A speed test measures the connection as it is at that moment, including whatever else is using it. For a number you can act on:

• Stop other downloads, backups and streams first — a single 4K stream is a quarter of a typical connection.
• Test over Ethernet if you can. On Wi-Fi you are partly measuring the room, and a 5 GHz connection two walls away can halve.
• Run it more than once. Connections vary by the minute and by the hour, and one number is an anecdote.
• Compare like with like: a result here and a result from another speed test can differ by 10–20% simply because they use different payload sizes and different numbers of parallel connections.

How the figure is arrived at

The download runs for about ten seconds and the upload for about six, transferring continuously, and each figure is the average rate over its window: total bytes divided by total time. That is the sustained speed rather than the best moment, so a brief burst does not become the headline and a brief dip does not either.

The first second and a half is measured but not counted. A connection does not start at full speed — the congestion window has to open, and on a long path that takes a moment — so including the opening would report the ramp-up as though it were the connection. Every speed test discards some opening period for this reason.

Every connection gets the whole ten seconds. A fast line simply moves more data in it — a gigabit connection transfers well over a gigabyte where a 20 Mbps one transfers 25 MB — and nothing cuts the test short for being quick. That matters for comparing results: a figure averaged over ten seconds and one averaged over half a second are not the same measurement, and this way every result on the page is the first kind.

The rankings, and how much to trust them

Each completed test is added to the tables below, so the rankings are built from whatever visitors have actually run recently rather than from published advertised speeds. That makes them more honest about real-world performance and less reliable in every other way, so a few things are worth knowing.

The figures are measured by each visitor’s own browser. There is no way around that — only a server timing the transfer itself could prove a number, and it would be measuring its own sending rather than your receiving. So the tables use medians over at least three tests, which means a handful of odd or invented results cannot move a provider up the list the way an average would.

They are also shaped by who happens to visit. A provider near the top may simply have a few enthusiasts on its fastest tier, and a country with a dozen tests tells you far less than one with a thousand. Read the “tests” column before reading the ranking.

What is stored

Running the test adds one row to the rankings. It contains the four speeds, your internet provider, your country and approximate city, and the datacentre that served the test.

Your IP address is not stored. The provider and the location are derived from it — that is how any site knows roughly where you are — but the address itself is never written down, and there is no column in the table that could hold one. The location is city-level at best and is often wrong for mobile connections and VPNs, where it points at the operator’s gateway or the exit server.

Rows are deleted after 90 days. Nothing here is tied to a visitor across tests: two runs from the same connection are two unrelated rows.

Why this tool needs a server

Nearly everything else in this catalogue runs entirely in your browser. This one cannot: measuring a connection means timing bytes against something at the other end of it, and a browser has no way to conjure one.

The download is random bytes generated for each request and never written down. The upload is random bytes your browser generates for the purpose, counted as they arrive and discarded — nothing of yours is sent, and nothing is kept.