How to test internet speed properly

Choose a comparison before pressing Start

To investigate broadband capacity, use a capable device connected by Ethernet to the router, where practical. Check the negotiated link rate: an adapter connected at 100 Mbps can limit a faster service. To investigate everyday experience, also test on Wi-Fi in the affected room. Keep these two sets of results separate.

For a quiet baseline, pause nonessential downloads, backups and streams on devices you control. For a workload comparison, repeat with the activity that normally causes trouble. Do not disconnect necessary services or disable required workplace security simply to raise a score.

A procedure you can repeat

1. Record the date, local time, device, connection type, room and selected server. Note whether a VPN is required or active.
2. Use the same device, server and duration for the first comparison. Keep the device in the same position for Wi-Fi trials.
3. Run the test and record download, upload, idle latency and both loaded-latency results. Avoid starting several benchmarks together.
4. Repeat three times, allowing each run to finish. Record all three rather than selecting only the fastest.
5. Change one condition, such as moving rooms or connecting Ethernet, and repeat. If the endpoint seems suspect, use another server as a separate comparison.

Auto uses Cloudflare routing; the selected network edge can change. A manually selected server makes comparisons easier to describe, but it cannot keep the whole internet route or server load constant.

Worked example: summarise without hiding variation

Suppose three download results are 180, 165 and 190 Mbps under the same recorded conditions. Sorted, they are 165, 180 and 190: the median is 180 Mbps and the range is 25 Mbps. Keep the individual readings too. This small sample describes those trials, not a guaranteed minimum or a statistical confidence interval.

If the affected room then measures 45, 52 and 48 Mbps, its median is 48 Mbps. That difference justifies investigating the wireless path; it does not establish whether placement, interference, client capability or another factor is responsible.

Read the result in the right units

Download and upload are separate directions, expressed in megabits per second. Divide Mbps by eight for decimal megabytes per second: 100 Mbps corresponds to 12.5 MB/s before considering the details of a particular transfer. Application processing, storage and remote-service limits can still extend completion time.

This tool times browser HTTP requests for latency. Its idle result is the median of 20 readings after a warm-up request; jitter is the mean absolute change between consecutive readings. It is not an ICMP ping or a direct measurement to a game server. See the latency guide before comparing numbers from different tools.

Account for test duration, data use and uncertainty

Quick, Standard and Extended select 6, 10 or 15 seconds for each transfer direction. Latency sampling and startup add time. The throughput calculation excludes an initial warm-up period and uses several concurrent requests, so it can differ from one file download.

A test transfers real data. At a constant 500 Mbps for ten seconds, one direction alone would transfer about 625 MB using decimal units. Actual usage varies; Quick is shorter, not a data cap. The panel reports tracked transfer bytes, not a carrier billing measurement. Avoid repeated benchmarks on a limited allowance.

Make evidence useful to someone else

Keep a private log with the server, conditions, all results and the real symptom, such as calls breaking up during uploads. Compare with the speed estimate and minimum, if any, in your own contract. A percentage of an entered plan speed is arithmetic, not proof of a contractual breach. Follow the provider’s required diagnostic process when reporting a fault.