How Ping Speed Test works

Ping Speed Test runs Cloudflare's open-source speed test engine in your browser against our own servers on Cloudflare's network. Ping is the median of about 20 HTTP round trips, speed is the 90th percentile of a series of transfers, and the bufferbloat grade is how much ping rises under load. Every threshold is listed below with its source. Nothing you measure is stored.

Who runs it

Ping Speed Test is published by Bopito LLC. The test is free, needs no account and runs in your browser.

It's an estimate of your connection at the moment you ran it, measured one way and explained here. It isn't certified by anyone, and other tests measure differently (why results differ).

The engine and our servers

The test uses Cloudflare's open-source speed test engine, @cloudflare/speedtest (MIT license), from a pinned copy served by our own site. We point it at our own servers: a small program on Cloudflare's network (a Worker) that answers at pingspeedtest.com. It sends test data for downloads, reads and discards uploads, and hands out login details for the packet-loss relay. Cloudflare runs it at the location nearest to you.

The engine can report results to Cloudflare when a test finishes. We switch that off: both reporting addresses are set to nothing, so your result goes nowhere but your screen.

Ping

Your browser sends about 20 small HTTP requests, one after another, each asking for zero bytes. The browser's own timing data records the time from sending each request to the first byte of the reply. Ping is the median of those round trips: the middle value, so one stray slow reply doesn't move it.

This is an HTTP round trip, not ICMP. Browsers can't send ICMP, the message the ping command uses. An HTTP round trip includes a little work at each end, so it can read a few milliseconds higher than an ICMP ping to the same place.

Jitter

Jitter is the average change between consecutive pings. For pings of 20, 22, 19, 35 and 21 ms, the changes are 2, 3, 16 and 14 ms, so jitter is about 9 ms. The ping test uses about 20 pings; the jitter test about 60, which catches more spikes.

Download and upload

The test transfers random data in a growing sequence:

Direction Transfers
Download 10 × 100 kB, 8 × 1 MB, 6 × 10 MB, 4 × 25 MB, 3 × 100 MB
Upload 8 × 100 kB, 6 × 1 MB, 4 × 10 MB, 4 × 25 MB, 3 × 50 MB

Each transfer gives one speed. The sequence stops stepping up once a transfer takes more than a second, so slower lines finish sooner and use less data. Your result is the 90th percentile of the speeds measured: faster than most transfers, but not a single lucky peak.

The limit. The largest transfers (100 MB down, 50 MB up) measure up to about 1 Gbps. Above about 950 Mbps, the result reads "at least 950 Mbps".

Loaded latency and the bufferbloat grade

While downloads and uploads run, the test sends a ping about every 400 milliseconds. Only pings taken during transfers that last at least a quarter of a second count, so the line is really busy. Ping during download is the median of up to 20 of those pings; ping during upload is worked out the same way.

The rise under load is the larger of the two loaded pings minus your idle ping, never below zero. The grade:

Grade Rise under load What it means
A Under 20 ms Ping barely rises under load.
B 20 to 49 ms A small rise; most people won't notice.
C 50 to 99 ms Noticeable lag while something else downloads or uploads.
D 100 to 199 ms Games and calls suffer whenever the line is busy.
F 200 ms or more Severe bufferbloat: any download or upload makes everything lag.

The scale is our own. Other bufferbloat tests use their own scales, so the same line can get a different letter elsewhere. Bufferbloat grades explained has examples.

Packet loss

A browser can't send raw UDP packets, but it can open a WebRTC connection. The test gets short-lived login details (valid for three minutes) from our Worker for Cloudflare Realtime TURN, a relay on Cloudflare's network. It then opens a WebRTC data channel from your browser, through the nearest relay and back to your browser. The channel is set to relay only, with nothing sent twice and no ordering, so it behaves like UDP: a lost message stays lost.

The test sends a burst of small numbered messages and waits a few seconds for stragglers. Loss is the share that never came back. If the relay connection can't open at all (a firewall, a VPN or a browser setting blocking UDP or WebRTC), the result says "UDP looks blocked" instead, and the rest of the test stands.

The verdicts and their sources

The full test and the lag test mark each use with a tick or a cross. A use passes only if every number meets its threshold.

Use Ping Jitter Packet loss Bufferbloat Download Upload
Competitive gaming 50 ms or less 10 ms or less 1% or less B or better 3 Mbps 1 Mbps
Video calls 150 ms or less 30 ms or less 2% or less C or better 4 Mbps 4 Mbps
4K streaming – – 5% or less – 15 Mbps –

Where each comes from:

  • 4K streaming, 15 Mbps down: Netflix recommends 15 Mbps or higher for 4K Ultra HD (Netflix: internet speed recommendations).
  • Video calls, 4 Mbps each way: Microsoft Teams' "best performance" figure for video, 4 Mbps up and down (Microsoft Teams: bandwidth requirements). It's the highest of the three tiers Microsoft lists.
  • Latency, jitter, loss and bufferbloat limits: our thresholds. Fast games judge hits within fractions of a second, so we hold competitive gaming to 50 ms ping, 10 ms jitter and 1% loss. Call apps buffer a little audio, which absorbs more variation, so calls get 150 ms, 30 ms and 2%. Streaming buffers ahead, so only heavy loss (over 5%) affects it. The bufferbloat limit is the grade at which a busy line starts to hurt each use.
  • Gaming speeds (3 Mbps down, 1 Mbps up): our threshold. Play itself needs little bandwidth; downloads and updates need more, but they don't affect lag.

When something fails, the result links one guide to start with, in this order: packet loss, then bufferbloat, then jitter, then ping, then speed.

What the test doesn't measure

  • ICMP ping. Browsers can't send it. Every ping here is an HTTP round trip.
  • Game servers. A browser can't reach them. Our ping goes to the nearest Cloudflare location, which is usually closer than a game's server, so in-game ping is usually higher.
  • Wi-Fi signal. The test sees the whole connection, Wi-Fi included, but can't measure signal strength or tell Wi-Fi delay from line delay. Running it once on Wi-Fi and once on a cable shows the difference (Wi-Fi vs Ethernet).
  • Speeds above about 1 Gbps. See the limit above.
  • Other routes. Your connection to other sites and services may take different paths with different results.

Privacy

  • Nothing is stored. We have no database. Your results stay in your browser; "Copy result" makes plain text there and sends nothing.
  • Your IP address reaches our Worker, as it does any website. The Worker turns it into a one-way hash used only as a key for rate limits (to stop floods of tests), and doesn't log it.
  • Location and network name. The Worker reads which Cloudflare location answered and your network's name from Cloudflare's request data, and returns them to your browser for display. They aren't stored.
  • Analytics. Google Analytics 4 receives that a test started and finished, the test type, and coarse bands only: ping and jitter in ranges (such as "20–50 ms"), loss in ranges, download in ranges and the grade letter. Never exact numbers, your network name or your IP address.

The privacy policy covers the rest of the site.

Limits on use

To keep the test free and the servers fair, each network can run a few tests a minute. Past that, the test says "Lots of tests from your network right now" and asks you to try again in a minute. It never retries on its own.

Sources