Test your internet speed

What the test actually does

It runs three measurements in order, all of them from this browser tab, using the public endpoints Cloudflare provides for exactly this purpose. First a few throwaway requests open the connection, then a dozen tiny ones measure the round trip to the nearest data centre — that is your latency, and how much it varies between requests is your jitter. Warming up first matters: a first request that is still paying for DNS and TLS can report five times the real figure. Then several parallel downloads run for a few seconds, and several parallel uploads after them, with the size of each transfer chosen from the speed already observed so the test adapts to a mobile connection and a fibre line alike.

Two details matter for the number you end up with. The first second of each phase is discarded, because a fresh connection starts slowly and ramps up — including that ramp would under-report a fast line. And more than one transfer runs at a time, because a single connection often cannot fill a fast link on its own. This is the same shape as every credible speed test, and it is why the reading climbs while the test runs and then settles.

Average, peak, and the gap between them

Each speed is reported twice: the average across the whole measured window, and the fastest single second inside it. The average is the honest headline — it is what the connection actually sustained — and the peak is what it managed when everything lined up. A second is the shortest window worth quoting, because anything briefer measures a buffer emptying rather than a line running.

The interesting part is the distance between the two. A few percent apart is a steady connection doing its job. A peak two or three times the average means the speed collapsed somewhere during the test — congestion on a shared segment, a Wi-Fi channel with a neighbour on it, a device that woke up and started downloading, or a provider shaping traffic after the first few seconds. That pattern is worth more to a support desk than either number alone, because it points at something intermittent rather than something broken.

Where the test measured to

The results name the Cloudflare data centre that answered, by the three-letter code of the airport it sits beside, and the straight-line distance from the location your network reports. That answers a question a speed test usually leaves open: whether a test run in Denver was answered in Denver or in Chicago. Cloudflare's network routes you to the site with the best path rather than the closest one on a map, so the two are usually but not always the same place.

The distance is a straight line, not a route — the fibre between you runs along roads and around geography, so the real path is longer. It is still enough to check the latency against, because light in glass travels about 200 km every millisecond, and a round trip covers the distance twice. A data centre 300 km away therefore cannot answer in less than about 3 ms, however good everything else is. If your round trip is far above that floor, the delay is equipment rather than distance; if it is close to it, the connection is about as quick as it can be from where you are.

Reading the four numbers

Download is what most people mean by speed: how fast data reaches you, and what decides whether a stream buffers. Upload is the other direction, and on cable and DSL it is deliberately much smaller — it is the number to look at when video calls freeze at your end or a backup takes all night.

Latency is the round-trip time of a single small message, measured in milliseconds. It is independent of speed, and it is what a connection feels like: a gigabit satellite link with 600 ms of latency is miserable for calls and games, while a modest fibre line at 10 ms feels instant. Jitter is how much that round trip moves about. Steady is what matters — a call with 30 ms of jitter sounds worse than one with a higher but consistent latency, because the audio has to be buffered or dropped.

Why your result may not match what you pay for

The advertised figure is what the line can deliver to your router under ideal conditions, and this measurement is taken at the end of a chain with several places to lose speed. Wi-Fi is nearly always the biggest: signal strength, walls, an old adapter, and a crowded 2.4 GHz band each take their share. Then there is everything else on the network doing its own thing, the overhead every transfer pays, the router's own limits, and on a shared cable segment the time of day.

If the number looks wrong, test it twice: once over Wi-Fi and once with a cable into the router. If the cable is much faster, the connection is fine and the Wi-Fi is the problem. If both are slow, the line or the router is worth a call to your provider — and a result from here, with the latency and jitter alongside it, is a more useful thing to quote them than "the internet is slow".

The other half of what a support desk asks for

A speed result answers how fast; the other two questions are always who and what. Your public address is on the home page, along with the network it belongs to, and your user agent reports the browser and version you are running. Between the three, a support ticket has everything it needs on the first reply rather than the third.

Common questions

How accurate is a browser speed test?

For a normal home connection it is close — within a few percent of what a desktop app reports — because the limit is nearly always the line itself rather than the browser. It becomes less reliable at the top end: past roughly 500 Mbps a browser tab, your Wi-Fi adapter, or an old phone can run out of headroom before the connection does, and the number you see is then measuring your device. Running it on Wi-Fi measures the Wi-Fi as much as the internet, so plug into the router if you want to test the line on its own.

Why is my speed lower than what I pay for?

A quoted speed is the maximum the line can carry to your router in ideal conditions, and several things sit between that and this measurement. Wi-Fi is usually the largest: distance, walls, and a crowded 2.4 GHz band routinely halve it. After that come the other devices sharing the connection, the protocol overhead every transfer pays, the age of your router, and the time of day on a shared cable segment. A result within about 20 percent of the advertised figure over Wi-Fi is normal; a result at a small fraction of it is worth investigating.

What are download, upload, latency, and jitter?

Download is how fast data comes to you, which sets how quickly a page, a video, or a game arrives. Upload is how fast data leaves you, which matters for video calls, backups, and sending large files — on most cable and DSL connections it is far smaller than the download figure, by design. Latency is the round-trip time for a single small message, and it is what makes a connection feel instant or sluggish regardless of its speed. Jitter is how much that round trip varies from one message to the next, which is what turns a call choppy even when the average looks fine.

Where does this test measure to?

To the nearest Cloudflare data centre, using the same public measurement endpoints Cloudflare's own speed test uses. That is deliberately close to you: it measures the part of the path you can do something about — your device, your Wi-Fi, your router, and your provider's network — rather than the state of a server on another continent. It is also why the result is a fair upper bound: no real download will be faster than this, though many will be slower for reasons that have nothing to do with your connection.

Which Cloudflare data centre am I being tested against?

The one that answered, which the results name by the three-letter code of the airport beside it — DEN for Denver, LHR for London — along with the straight-line distance from the location your network reports. Cloudflare routes each request to the site with the best path rather than the nearest one on a map, so a test from Denver is usually answered in Denver but can land in Chicago or Dallas if that is where the better route goes. The distance is worth reading next to the latency: light travels about 200 km through fibre every millisecond and a round trip covers the distance twice, so anything much above that floor is equipment rather than geography.

What is the difference between the average and the peak?

The average is the speed sustained across the whole measured window and is the number to quote; the peak is the fastest single second inside that same window. A steady connection shows the two within a few percent of each other. A peak that is two or three times the average means the speed collapsed part-way through — congestion, Wi-Fi interference, another device waking up, or a provider shaping traffic after the first few seconds — and that gap is a more useful thing to report to a provider than a single slow figure, because it shows the line can go faster than it did.

How much data does the test use?

A standard run transfers up to about 125 MB — roughly 100 MB down and 25 MB up — but only a fast connection actually reaches that, since each phase stops on a timer as well as a data cap. A slow connection may move only a few megabytes in the same window. Light mode caps the whole run at about 20 MB for anyone on a metered or mobile connection, at the cost of accuracy on very fast lines, which need a longer transfer to measure properly. Either way, the exact amount used is shown with the results.

Why does the number climb while the test runs?

Because connections do not start at full speed. A new transfer begins cautiously and doubles its rate every round trip until it finds the ceiling — the mechanism that stops the internet collapsing under congestion — so the first moments of any measurement are genuinely slower than the line. The reading you see climbing is that ramp happening live. The final figure discards the first second and several parallel transfers are used, which is what it takes to fill a fast connection from a browser.

Is anything about my connection uploaded or stored?

The measurement transfers are meaningless payloads — zeroes going up, zeroes coming down — sent to Cloudflare's public endpoints, and nothing about your result is stored, logged, or sent anywhere by this site. Your results exist only in the page in front of you and disappear when you close the tab. The network name shown alongside them comes from the same lookup the home page uses, which reads your request and answers it without keeping a copy.