I have a 3 Gbps internet connection, 2.5 Gbps network cards, compatible routers, and proper cabling. Over a wired connection, I normally see around 2.2 Gbps for both downloads and uploads. However, some Ookla speed tests—depending on the selected server—report speeds of 10 or even 20 Gbps, which is higher than the physical limit of my network interfaces. Occasionally, a test shows an extremely high download result but no upload result at all. What causes these impossible-looking numbers? I'm asking about how the test reports those values, since I'm not actually downloading data at 10–20 Gbps.
4 Answers
Those results are almost certainly measurement artifacts rather than real throughput. Speed tests estimate bandwidth from how quickly data is sent and acknowledged, so brief bursts, timing differences, browser behavior, packet batching, or network-card offloading can make the calculated number much higher than the connection can sustain. The result is best treated as an estimate, not an exact measurement.
The selected test server matters too. Servers have different network paths, capacity, latency, and testing implementations, and some may produce bad or incomplete results. A result above 2.5 Gbps cannot represent actual traffic through a 2.5 Gbps interface, so the unusually high download numbers—or a missing upload result—should be considered invalid data.
Also check the units carefully. Internet speeds are normally shown in Gbps, meaning gigabits per second, while file sizes are usually measured in GB, or gigabytes. There is also a difference between Gbps and GBps. In your case, though, even 10 Gbps would still be impossible through a 2.5 Gbps network interface, so the main issue is an unreliable test result rather than extra bandwidth.
There isn’t a special networking technology that can make the interface exceed its negotiated link speed. Verify that the interface is actually linked at 2.5 Gbps and compare results using a few nearby, reputable servers or a sustained large-file transfer. Your consistent real-world result around 2.2 Gbps is much more believable than a brief 10–20 Gbps reading.
That makes sense. So the unusually high values are caused by how the test measures short bursts rather than by my connection actually exceeding the interface limit.