Internet speed test

How the measurement is taken

• Latency is the fastest of twelve HTTPS round trips to an endpoint with no body, not the average. The average measures the worst queue you happened to hit; the minimum measures the path itself. Two uncounted requests run first so a cold server does not land in the sample. It will read a little above an ICMP ping, because it includes the server answering rather than only the network.
• Jitter is the mean absolute difference between consecutive round trips. That is the number that decides whether a call sounds clean, and it is independent of how much bandwidth you have.
• Throughput is measured over several parallel connections. A single TCP stream is limited by the bandwidth-delay product, so testing one connection at a time understates a fast line on a long path.
• The first 1.5 seconds of each stage are thrown away. TCP starts slow and ramps up. Including the ramp measures congestion control rather than your connection.
• The payload is random. A stream of zeroes compresses to nothing and would report a speed that never happened.

Reading the numbers

HD video needs roughly 5 Mbps per stream and 4K around 25 Mbps, so a household watching on two screens is comfortable at 50 Mbps. Video calls need very little bandwidth and care enormously about the other two figures: keep latency under 150 ms and jitter under 30 ms and a call holds together. Competitive gaming wants latency under 50 ms, and no amount of bandwidth substitutes for it.

A result below the speed on your bill is normal. Advertised figures are the line rate to your provider, measured on a wired connection with nothing else running. Wi-Fi, the distance to the test server, peak-hour congestion, an ageing router and every other device in the house all take a share. Within about 80 percent of the advertised number over Wi-Fi is a healthy connection.

Getting a result you can trust

1. Use a cable if you have one. Wi-Fi is the bottleneck in most slow results.
2. Stop other downloads, backups and streams first, on every device on the network, not just the one you are testing from.
3. Run it three times and take the best. Any single run can land in a moment of congestion.
4. Test at the hour you care about. A line that reaches its rated speed at 3am and collapses at 8pm is a congestion problem, and only the second measurement is worth reporting to your provider.

Testing what a VPN costs you

Run this test with the VPN off, note the three numbers, connect, and run it again. The difference is the real cost of the tunnel on your line, which is more useful than any provider's marketing figure.

Expect to lose some throughput. Encryption costs CPU time and the traffic takes a longer path, so a fair range is roughly 10 to 30 percent on a nearby server, and considerably more on a distant one. Latency rises by roughly the round trip to whichever server you picked, so connecting to another continent adds a large fixed penalty no amount of bandwidth hides. Picking the closest server recovers most of both.

If the drop is far worse than that, the usual causes are a distant server, an overloaded one, or a protocol fighting your network. It is also worth confirming the tunnel is actually carrying everything: run the DNS leak test and the WebRTC leak test, since a connection that only half works can look fast for the wrong reason.

Data usage and privacy

The download stage runs for about eight seconds and the upload stage for about eight more, so the amount transferred scales with your speed. On a 100 Mbps connection that is roughly 100 MB down, plus however much your uplink manages in its eight seconds. Each direction stops at 256 MB regardless of how fast the line is, so a gigabit connection finishes its stages early rather than moving a gigabyte. Do not run it on a metered mobile plan without checking your allowance.

Every byte goes to and from endpoints on this domain. No third-party speed test service is embedded, no result is stored, and no identifier is set. The usual embeds carry advertising trackers, which is not something a site whose product is a no-logs VPN can credibly ship.