How this speed test measures your connection
A speed test is simple in principle — move a lot of data and time it — but getting a number that matches what your line can really do takes care. This one follows the approach used by the best modern tests, Cloudflare’s and Netflix’s fast.com:
- Idle latency first. Around 20 tiny requests measure the round trip to the server while nothing else is happening. The median is your latency (ping); the average change between consecutive probes is your jitter.
- Parallel streams, growing sizes. Download starts with two small 100 kB requests, then — as the speed becomes clear — opens up to eight parallel streams and grows each request to roughly 1.5 seconds’ worth of data, up to 50 MB. That is what lets it fill a 1 Gbps full-fibre line, while a 4G phone or a 3 Mbps ADSL line simply sees small requests and fewer streams.
- Warm-up discarded. TCP starts slowly and the stream count ramps up, so the first part of each phase (up to two seconds) is thrown away.
- Percentile, not peak. Bytes are counted as they arrive and turned into overlapping two-second rates. The headline figure is the 90th percentile of those rates, the same statistic speed.cloudflare.com uses; the sustained average is shown alongside it, and every figure is in Technical details.
- Latency under load. While data is flowing, probes keep running. How much latency rises is bufferbloat — the reason calls stutter when someone else starts a download.
Each phase stops once the result has been stable for a few seconds, at 12 seconds, or when the data limit is reached. In the default Auto mode, latency probes and download data come from our own Cloudflare edge function (/api/speed), served from the Cloudflare data centre nearest to you — London or Manchester for most UK users. Uploads go to Cloudflare’s public speed.cloudflare.com endpoint, because in our testing it accepted upload data noticeably faster than our function can, and an upload test must never be limited by the server. You can force either server to cross-check, and the page falls back to Cloudflare’s endpoints automatically if ours do not answer.
What download, upload, latency and jitter mean
Download governs streaming, web browsing and how long updates take. Upload matters for video calls, sending large files, cloud backup and live streaming — and on FTTC and cable it is often a tenth of the download. Latency is the delay before anything happens, and matters more than bandwidth for gaming and calls. Jitter is how much that delay wobbles.
| Activity | What it needs |
|---|---|
| Netflix 4K / UHD | 15 Mbps per stream (HD 5 Mbps, SD 3 Mbps) |
| Teams / Zoom group video in HD | About 3.8 Mbps up and 3 Mbps down, latency under 150 ms |
| Online gaming | Only 3–6 Mbps, but latency under 50 ms, low jitter and little bufferbloat |
| Working from home | 10 Mbps up or more makes cloud files and calls painless |
| Downloading a 100 GB game | About 13 minutes at 1 Gbps; 7½ hours at 30 Mbps |
The bufferbloat grade uses the extra latency seen under load: A+ below 5 ms, A below 30 ms, B below 60 ms, C below 200 ms, D below 400 ms, F beyond. Grades C and worse mean calls and games lag whenever the line is busy. The usual cure is a router with Smart Queue Management (SQM, fq_codel or CAKE), set slightly below your line speed.
UK broadband speeds, Ofcom figures and your rights
Ofcom’s Connected Nations 2025 report puts the average UK fixed-line download speed at 285 Mbps (July 2025, up from 223 Mbps a year earlier). That is a mean, pulled up by the growing number of gigabit full-fibre lines, so most homes sit below it: a typical FTTC or SOGEA line from BT, Sky, TalkTalk or Vodafone manages 30–80 Mbps down and 10–20 Mbps up, Virgin Media’s cable and full-fibre packages range from about 100 Mbps to over 1 Gbps, and ADSL rarely beats 10 Mbps.
The Universal Service Obligation gives every home the legal right to request a “decent” connection of at least 10 Mbps download and 1 Mbps upload, provided by BT (or KCOM in the Hull area), subject to a cost threshold. If your results are consistently below that on a wired connection and no better service is available at your address, check your eligibility with BT.
If you are simply getting less than you pay for, Ofcom’s voluntary Codes of Practice on broadband speeds — signed by the major providers — require them to give you a minimum guaranteed speed when you sign up. If speed falls below it and the provider cannot fix it within 30 days, you can leave without penalty. Keep a record: test on a cable at different times, and export your history as CSV to support your complaint. Your provider will usually want its own test run too.
Getting an accurate result — and why speed tests disagree
- Use a cable. Wi-Fi is the most common bottleneck. A phone two rooms away from the router may see a third of the line speed.
- Check the device. Many laptops, TV boxes and powerline adapters have 100 Mbps Ethernet ports, and older Wi-Fi 4 or 5 devices cannot reach gigabit speeds. A slow phone or browser extension can also cap the result.
- Stop other traffic. Game updates, cloud photo sync, 4K streams and backups on other devices all share the line.
- Turn off VPNs — they add distance and encryption overhead. Check with what is my IP that yours is really off.
- Test more than once, including the 8–10pm peak, when congestion is most likely. The history table makes the comparison easy.
Different tests give different numbers for honest reasons. Speedtest.net uses a server run by a provider near you; fast.com uses Netflix’s own servers; Google’s M-Lab test uses a single connection, which reads lower on long or lossy paths; this test and Cloudflare’s use the nearest Cloudflare data centre. Methods differ too: peak, average or percentile. On mobile, 4G and 5G speeds swing from second to second, so the percentile and average here can sit further apart than on fibre.
For other checks on your connection, the network tools include a DNS lookup for when websites seem slow to start loading.
What a browser can and cannot measure
A web page can only use HTTP over TCP or QUIC, so some things are out of reach, and we would rather say so than invent a number. True packet loss needs raw UDP or ICMP, which browsers do not allow; lost packets are silently resent by TCP. Instead the test reports stalled probes — latency checks that took over three seconds — which only happens when loss or dropouts are serious. Ping here is an HTTP round trip to the edge server, typically a millisecond or two above an ICMP ping.
A speed test uses real data. The Standard setting caps it at about 500 MB (a 50 Mbps line uses roughly 75 MB); Light caps it near 50 MB; No limit runs a full 15 seconds each way and can use several gigabytes on fast fibre. The test never starts by itself, a live counter shows what has been used, and on phones that report mobile data the page asks before running anything beyond Light.
Nothing about your test is stored on our servers: the edge function generates random data, or counts and discards what you upload. Your history lives in this browser’s local storage and can be exported or cleared at any time.
Frequently asked questions
What is a good internet speed in the UK?
For most homes, 30–80 Mbps download handles several HD or 4K streams, video calls and gaming at once. Families who download big games or work from home benefit from 150 Mbps or more, and full fibre offers up to 1 Gbps or beyond. Upload matters too: 10 Mbps or more makes calls and cloud backups painless. Anything below 10 Mbps down or 1 Mbps up is under the UK legal minimum.
Why is my speed lower than the package I pay for?
Package speeds are advertised as the median peak-time speed for that product, so half of customers get less. Wi-Fi is the most common cause of a low result, followed by old devices with 100 Mbps network ports, other devices using the line and evening congestion. Test on a cable first. If it is still below the minimum guaranteed speed your provider quoted, you can complain under Ofcom’s Codes of Practice.
What is bufferbloat and how do I fix it?
Bufferbloat is the extra delay added when your router or modem queues too much data during a big download or upload. Video calls break up and games lag whenever the line is busy, even though the speed looks fine. The fix is Smart Queue Management (SQM) — fq_codel or CAKE — found in many gaming and mesh routers and in OpenWrt. Set it slightly below your measured speed.
How much data does a speed test use?
It depends on how fast your connection is, because the test moves as much data as the line can carry for several seconds. On Standard, a 50 Mbps connection uses about 75 MB and the cap is about 500 MB. Light stops at roughly 50 MB, which suits mobile data. No limit on gigabit fibre can use two to three gigabytes. The counter on the page shows the exact amount used.
Why do speedtest.net, fast.com and this test give different results?
Each uses different servers and a different method. Speedtest.net often picks a server inside your provider’s network, fast.com uses Netflix servers, and this test uses Cloudflare’s nearest data centre. Some report the peak, some the average; we show the 90th percentile and the mean. Results within 10–15% of each other are effectively the same. A consistently much lower figure on one test points to a routing problem.
Is 10 Mbps the legal minimum broadband speed?
Yes, in effect. The Universal Service Obligation gives UK homes and businesses the right to request a connection of at least 10 Mbps download and 1 Mbps upload, delivered by BT or by KCOM in Hull, if no such service is available and the build cost is within the threshold. It is a right to request an upgrade, not a guarantee from your current provider.
Can this test measure packet loss?
Not directly. Browsers only allow HTTP over TCP or QUIC, and TCP quietly resends lost packets, so a web page never sees them. Measuring true loss needs UDP, as in games and calls. We report stalled probes — latency checks taking over three seconds — which reveals serious loss or dropouts. For precise figures, use your router’s diagnostics or a command-line tool such as ping or mtr.