When we started testing speed tests, we were very uncertain. This was mainly because the differences between the speed tests were very large and we had no idea what speed our ISP actually delivered (that's why we started testing with a throtteld browser).
The ISP can say that you get 100 Mbps, but if one speed test measures 93 Mbps and the other speed test measures 110 Mbps, then you have no idea what is correct.
So we were relieved when we discovered that your true internet speed is 12% higher than the advertised speed.
We were convinced that this is a correct conclusion and that 100.8 Mbps is the sweet spot when the advertised speed is 100 Mbps.
However, we recently realised that Ookla speedtest also regularly measures 102 Mbps instead of 100.8 Mbps.
These speeds were recently measured in the following tests:
- Which Ookla clone is the best? - Max speed measured: 102.0 Mbps
- Are all Ookla testservers equal? - Max speed measured: 101.99 Mbps
- Differences in Ookla test servers - Max speed measured: 101.92 Mbps
- Does distance matter? - Max speed measured: 102.74 Mbps
That puzzles us.
There are only a few components that determine the measured speed:
- The speed that the ISP offers
- The capacity of the weakest link (usually the home network)
- The speed test used
It is clear that the weakest link is not the test server, the internet connection or the home network. After all, we are measuring a higher speed than expected.
That means that our ISP is delivering more than 12% of the advertised download speed or that Ookla's Speedtest is using a higher percentage than 90% of the maximum speed.
In 2023 the highest speed measured by Ookla is 101.8 Mbps and the highest speed measured by M-Lab is 101.60 Mbps.
Assumption 1, Ookla measures 90%
If we assume that the Ookla Speedtest still shows 90% of the true internet speed as measured speed, then this means that our true internet speed is 102 / 0.9 = 113.33 Mbps.
This would mean that internet providers deliver a 13.33% higher speed than the advertised speed.
What we don't understand now is why 78% of Ookla test servers offer a slightly slower speed than expected. We can't think of a good reason for this.
Assumption 2, the ISP delivers a 13.33% higher speed as advertised
Lets assume that the ISP delivers a 13.33% higher speed instead of a 12% higher speed. We can easily do this by measuring our internet speed with other accurate speed tests.
We re-measure our internet speed using the 10 most accurate speed tests:
- M-Lab 98.26 (97.05), 98.11 (97.67) and 98.12 (97.69) Mbps (NDT (MSAK))
- Ookla Speedtest 100.56, 100.55 and 100.46 Mbps (server: Odido, Amsterdam)
- RealSpeed 98.5/98.7, 99.7/99.5 and 99.0/97.5 Mbps (Router/Device)
- Think Broadband (the standalone version) 99.0, 100 and 98.8 Mbps (IPv4)
- Down at a Glance 99, 99 and 98 Mbps (5 seconds)
- Waveform (Bufferbloat and Internet) speed test 100.6, 99.3 and 98.2 Mbps
- Fireprobe 100.35, 100.29 and 100.31 Mbps
- True Speed at a Glance 110, 109 and 110 Mbps (5 seconds) Realistic
- Google Fiber 99.7, 99.3 and 99.6 Mbps
- Meter.net 100.7, 100.7 and 101.0 Mbps
The maximum speed measured by the mainstream speed tests is 101.0 Mbps (instead of 100.8 Mbs). The maximum speed measured by the realistic speed test is 110 Mbps (instead of 112 Mbps).
This would mean that speed tests would measure 89% of the maximum speed instead of 90%.
Nice try, but this still doesn't explain why 22% of Ookla test servers deliver higher than expected speeds.
Cloudflare and SpeedOf.me
Cloudflare clearly states that the precent the 90th percentiel, while SpeedOf.me measures your true internet speed.
- Cloudflare 99.5, 99.5 and 99.3 Mbps
- SpeedOf.me 111.8, 111.77 and 111.74 Mbps (max value) Realistic
Based on these results we conclude that our ISP still delivers a true internet speed of 112 Mbps.
Hence we have to conclude that when a speed of 102 Mbps is measured, the Ookla result is 102 / 112 = 91% of your true speed.
Preliminary conclusions
Our ISP still delivers 112 Mbps, that is the advertised speed of 100 Mbps plus 12%. This speed hasn't changed.
Neither the test server, nor our network, nor the internet connection are the bottleneck.
Because we always used the same Ookla speed test in the test Differences in Ookla test servers, we know that the differences are caused by the test server used by Ookla. It is almost inevitable that Ookla sometimes calculates with 90% and sometimes with 91% of the maximum speed (depending on the test server that is selected).
We find it a strange conclusion, and are therefore still puzzled by Ookla.