Run the M-Lab Speed Test Used in Google Search
This page runs M-Lab’s open NDT7 test, the test behind the speed check in Google Search, directly on M-Lab’s servers. You get download, upload and latency, and you can compare them with a Cloudflare test in one table.
Google is a trademark of Google LLC. This page is not affiliated with Google. The test runs on Measurement Lab (M-Lab) servers.
M-Lab NDT7 speed test
Full test: up to 10 seconds each way with no data cap, like M-Lab’s own client. At 100 Mbps that is about 125 MB each way.
Tick the M-Lab box below, then press Run M-Lab test.
The M-Lab test could not finish
- Read the note about M-Lab above and tick the box if you agree.
- Pick a test size above and press Run M-Lab test.
- If you like, run a Cloudflare test to compare the two.
- Download All data received over one connection, divided by the whole test time, including the slow start. This is how M-Lab counts it. — Mbps
- Upload Data sent over one connection, divided by the test time. When the server reports how much it received, that count is used. — Mbps
- Latency The shortest round trip the M-Lab server’s TCP connection saw during the download (TCPInfo MinRTT). It is measured by the server, not with ICMP ping. — ms TCP minimum
- TCP retransmissions Share of download data the server had to send again. It hints at congestion or loss on the path, but it is not the same as packet loss. — % of data sent
The chart shows the running average speed, in Mbps, as M-Lab counts it during each direction.
Test details
- M-Lab server
- —
- Server location
- —
- Connections
- 1 for download, 1 for upload
- Round trip (TCP, latest)
- —
- Data used
- —
- Test duration
- —
- Date and time
- —
M-Lab picks the nearest server that has room for a test. Its servers usually sit outside your provider’s network.
Compare with a Cloudflare test
This runs our own speed test against Cloudflare’s nearest server in Data saver mode, using up to 37 MB. It never runs at the same time as the M-Lab test.
Not run yet.
| Measure | M-Lab (NDT7) | Cloudflare |
|---|---|---|
| Download | — | — |
| Upload | — | — |
| Latency | — | — |
| Latency method | TCP minimum, seen by the server | Median of 20 HTTPS requests |
| Speed method | All data ÷ whole time | Average after connection warm-up |
| Connections | 1 each way | 1–2 each way |
| Server | — | — |
| Data used | — | — |
A difference between the two columns is normal. Why M-Lab and Cloudflare results differ.
How to read your M-Lab result
Look at download first, then latency. Download and upload are averages over the whole test on a single connection, so they are often lower than your plan speed on fast lines. Latency here is the shortest round trip the server saw, which is close to the delay of your path when it is not busy.
| Number | What it is | What to look for |
|---|---|---|
| Download | Data received ÷ test time, one connection | Compare it with the Cloudflare column, not only with your plan |
| Upload | Data sent ÷ test time, one connection | Matters for video calls and sending files |
| Latency (TCP minimum) | Shortest round trip during the download | Higher than Cloudflare is normal when the M-Lab server is farther away |
| TCP retransmissions | Share of data the server sent twice | A rough hint of congestion, not packet loss; compare several runs |
For a judgement of the numbers themselves, our main speed test rates streaming, calls and gaming against our published thresholds.
Why M-Lab and Cloudflare results differ
The two tests measure different paths with different maths, so both numbers can be right at once. M-Lab’s test uses one connection. On fast fibre or when the server is far away, one connection often cannot fill the line, so the M-Lab number can be lower than a multi-connection test.
- Server location. Cloudflare answers from its nearest data centre, often inside a big Indian city. M-Lab servers usually sit in other networks, so the path is longer and the TCP round trip is higher.
- How speed is counted. M-Lab divides all data by the whole test time, including the slow first second. Our test leaves out connection warm-up, with a shorter window when a data cap ends the test early.
- Connections. M-Lab uses one. Our Data saver test uses up to two, and our Standard and Gigabit tests up to four.
- Data caps. Our Data saver compare run stops at 25 MB down and 10 MB up. On a line faster than about 60 Mbps download (or 25 Mbps upload) it shows a lower limit, marked “≥”.
Other tests add their own differences. Cloudflare’s own site reports the 90th percentile of many downloads, FAST.com downloads from Netflix servers that are often inside your provider’s network, and Ookla uses up to eight connections to a nearby server. Our guide to why speed tests differ covers them all.
Speed test in Google Search not working?
If the test in Google Search does not start or stops, you can run the same M-Lab test on this page. The Google Search Help page says Google partners with M-Lab to run that test, that it takes about 30 seconds, and that it can currently measure speeds up to 700 Mbps (Google Search Help: Test your internet speed).
When the M-Lab test fails here too, these checks help:
- Turn off a VPN, ad blocker or privacy extension for a moment, then try again. The test needs WebSocket connections to M-Lab’s servers.
- On an office, school or hostel network, ask whether a firewall blocks these tests. Try once on mobile data if your pack allows it.
- Wait a while if you see “no free test server” or “try again later”. M-Lab’s developer page says it limits each client to 40 tests a day, and mobile networks where many users share one IP address reach limits sooner (M-Lab’s developer guide).
- If only M-Lab fails, run our Cloudflare speed test. It uses different servers.
How this test works
This page does not use the Google Search widget. It asks M-Lab’s Locate service for the nearest server with room for a test, then opens one encrypted WebSocket connection for download and one for upload, each for up to 10 seconds. Speed is all data moved divided by the time, as M-Lab counts it. Latency and TCP retransmissions come from the server’s own TCP statistics.
M-Lab publishes every result, including your IP address, date and time, as open data, so the test runs only after you tick the box, every time. We do not store or receive your result; it stays on this page. A Full test uses whatever your line moves in up to 20 seconds, Data saver about 35 MB, and the Cloudflare comparison up to 37 MB. Read the full method or see what each service sees.
Questions about the M-Lab speed test
Is the speed test in Google Search better than Ookla?
Neither test is simply right or wrong; they measure in different ways. The M-Lab test in Google Search uses one connection and reports the average over the whole test, while Speedtest by Ookla uses up to eight connections to a nearby server. On fast lines a single-connection test often shows a lower number.
Is the M-Lab speed test safe to run?
The M-Lab test only sends and receives test data in your browser and installs nothing. The part to think about is privacy: M-Lab’s privacy policy says test data is made public, including your IP address, date and time. That is why this page asks for your consent before every run.
Why is my M-Lab result lower than my plan speed?
An M-Lab result is often lower than a plan speed because one connection to a server outside your provider’s network cannot always fill a fast line. Wi-Fi lowers it further. To check your plan, test on a LAN cable with our broadband speed test.
How much data does the M-Lab test use?
A full M-Lab test uses as much data as your line can move in about 10 seconds each way. At 100 Mbps that is about 125 MB each way, and more on faster lines. On mobile data, choose Data saver, which stops at about 35 MB in total.
Why does the M-Lab test say no server is free?
M-Lab answers “no capacity” when its nearby servers are busy, and it limits how many tests one client can run in a day. Many mobile users in India share one public IP address, so they can reach that limit sooner. Wait a few minutes and try again, or run the Cloudflare test meanwhile.