Speed tests disagree because each one measures a different path in a different way. They use different servers in different places, push the connection with different numbers of parallel streams for different lengths of time, and turn the samples into one number with different math. So two honest tests can give noticeably different numbers on the same line, and neither is lying. They're measuring slightly different things.
They test to different servers
Every speed test measures the path from your device to its own server, and those servers sit in different places on the internet.
- Speedtest by Ookla uses a network of servers that are owned and operated by ISPs and mobile carriers. By default it picks a nearby server with a fast ping. That server is often inside your own provider's network, which is close to a best case.
- FAST.com runs downloads from and uploads to Netflix's servers, the same ones that deliver Netflix video.
- Cloudflare's test measures to Cloudflare's own network.
- SpeedMetric runs in your browser against Cloudflare's nearest data center.
A server inside your ISP's network skips the links between providers, where congestion often happens. A content-delivery server shows what real services can deliver to you. Ookla says so itself: many sites and streaming services keep their content on servers farther away than its test server, which can mean slower speeds and higher ping from those services.
They push the connection differently
A single connection rarely fills a fast line, so most tests open several at once. How many, and for how long, changes the result.
- Ookla scales the number of connections to multiple servers to saturate links up to 10 Gbps.
- Cloudflare deliberately doesn't try to saturate your connection. It sends a series of fixed-size data blocks to reflect more ordinary use. Small transfers can't reach full speed, because each one is partly limited by latency.
- FAST.com runs a series of downloads and uploads and reports the maximum speed your connection can deliver.
- SpeedMetric opens several parallel connections for up to five or six seconds per direction and ignores the first fraction of a second, while connections are still ramping up.
So a test built to find your connection's ceiling will usually read higher than one built to mirror everyday browsing. On a gigabit plan, that gap can be large.
They calculate the final number differently
A speed test collects many samples and then reports one number. Some report something close to the peak. Others average everything after a warm-up. Others combine block-by-block measurements. SpeedMetric divides the data it transferred by the time it took, after a short warm-up, rather than reporting a momentary peak.
None of these methods is wrong. A peak-style figure answers "what can this line do?" An average answers "what did I get over these seconds?"
Ping and latency differ even more
Latency numbers vary between tests more than speeds do, because tests measure different kinds of latency:
- Idle ping is the round trip on a quiet connection. SpeedMetric reports idle ping and jitter.
- Loaded latency is the round trip while the connection is busy. FAST.com shows unloaded and loaded latency, and Cloudflare and Ookla also measure latency under load. SpeedMetric doesn't.
If one test shows 15 ms and another shows 80 ms, check whether one of them is reporting latency under load. That gap is useful information in its own right. What download latency means explains why.
Your side of the test changes things too
Some differences have nothing to do with the test itself:
- Wi-Fi vs Ethernet. Wi-Fi results move around from minute to minute. Comparing two tests over Wi-Fi mostly measures your Wi-Fi.
- Browser vs app. Browser tests carry some overhead, and on multi-gigabit lines a slow device can become the limit.
- Timing. Tests run a minute apart see a different network. Someone else in the house may have started a download in between.
- VPN. A VPN routes every test through its own server, so all of them measure the VPN as well.
Which test to trust, and for what
Use the test that matches your question:
| Your question | Best approach |
|---|---|
| Am I getting the speed I pay for? | A test that saturates the line, on Ethernet, at a quiet time. Compare it with why tests fall short of the advertised speed. |
| Why does Netflix buffer? | FAST.com, since it uses Netflix's own servers, plus a test on the TV itself. |
| Why do calls or games lag when the house is busy? | A test that reports loaded latency, alongside idle ping. |
| Is my Wi-Fi worse in this room? | Any test, as long as you use the same one in each room. |
For tracking your connection over time, the method matters more than which brand you pick. Use the same test, the same device and the same connection type, and run it three times and keep the middle result. How to test internet speed properly walks through the setup, and what a speed tester measures covers what each number on the result screen means.
Common questions
Why does the same speed test give a different result every time?
Your connection changes from second to second. Other devices start and stop, Wi-Fi conditions shift, and the test server's load varies. As a rough guide, a 10% swing between back-to-back runs is nothing to worry about. Run three tests and use the middle one.
Is a speed test app more accurate than a browser test?
On very fast connections, often yes. A browser adds overhead and can be held back by a slow device. On typical home plans, a browser test on a decent computer usually lands close to an app.