Why Speed Test Results Vary: Latency, Jitter, Download and Upload Explained

A speed test reports four measurements. They describe different things and matter for different activities.

Speeds are in megabits per second. A file size in megabytes needs to be multiplied by 8 to compare: a 100 Mbps connection downloads at most about 12.5 MB per second, and in practice somewhat less once protocol overhead is counted.

For most households, latency and jitter affect day-to-day experience more than raw download speed once you're above roughly 25–50 Mbps per active user. A 4K stream needs around 15–25 Mbps; a video call needs a few Mbps but suffers badly from jitter.

Our Speed Test runs entirely in your browser against our own server, in four phases:

1. Ping: the browser makes 10 tiny requests and times each round trip. Ping is the average. Jitter is the average difference between consecutive round trips — if they come back at 20, 24, 19 and 23 ms, jitter is about 4 ms.
2. Download: the browser opens 6 parallel connections and pulls data continuously for 12 seconds, counting bytes as they arrive.
3. Upload: the same idea in reverse for 10 seconds.
4. Warm-up is discarded. TCP starts slowly and ramps up its sending rate (*slow start*), so the first 15% of each phase is thrown away and the speed is calculated from the steady-state part.

Parallel connections are used because a single TCP connection often can't fill a fast link, especially at higher latency. A single stream's throughput is limited by how much data can be "in flight" before an acknowledgement comes back. Six streams together get much closer to the line's real capacity, which is what most speed tests and most real downloads do.

Because the ping is an HTTP request rather than an ICMP ping, it includes a little application overhead and reads a few milliseconds higher than ping from a terminal would.

Server location. Every test measures the path between you and one particular server. A server in your city and one on another continent will give very different latency and often different throughput. Our test enters our hosting provider's network at its nearest edge location but is answered by our application server, which runs in a single region — so if you're far from that region, expect higher latency than a test against a server in your city. Other tests pick a server from a list, often one hosted inside your ISP's own network, which measures your connection under ideal conditions rather than the route to the sites you actually use.

Wi-Fi. This is the single largest source of variation. Distance from the router, walls, interference from neighbouring networks, and the 2.4 GHz vs 5 GHz vs 6 GHz band can cut throughput by half or more and add tens of milliseconds of jitter. A wired result is what your ISP is delivering; a Wi-Fi result is what your room is getting.

Other traffic on your network. A cloud backup, a game update or a 4K stream on another device competes for the same line.

Time of day. Shared segments of ISP networks get congested in the evening peak, typically 7–11 pm local time.

The device. Old phones, browsers with many extensions, or laptops in power-saving mode can be the bottleneck above a few hundred Mbps.

VPNs. Every byte goes through the VPN server first, adding latency and often capping throughput at whatever that server can handle.

Watch the ping figure while the download phase is running. On many connections it jumps from 15 ms idle to 200 ms or more when the line is busy. That's bufferbloat: the router or modem holds too much queued data in oversized buffers, so every packet — including your video call's — waits behind a queue of download traffic.

Bufferbloat is why a video call stutters the moment someone else starts a download, even on a fast connection. The fix is Smart Queue Management (SQM) using algorithms like fq_codel or CAKE, which keeps queues short. Many modern routers offer it under names like "QoS", "Smart Queue" or "Anti-Bufferbloat". Turn it on and set the limits to slightly below your measured speeds.

A good connection keeps latency under load within roughly 30 ms of its idle latency.

If you're checking whether you're getting what you pay for — or gathering evidence for your ISP — make the test as clean as possible:

1. Use a wired Ethernet connection directly to the router. If you can't, sit next to the router on the 5 GHz or 6 GHz band.
2. Pause other devices and downloads, including automatic cloud sync and game launchers.
3. Close other browser tabs and disconnect any VPN.
4. Run three tests a few minutes apart and take the middle result.
5. Repeat at different times of day, including the evening peak. If speeds are fine in the morning and collapse every evening, the problem is congestion in the ISP's network, not your equipment.
6. Test from a second device. If one laptop is slow and another isn't, the problem is the laptop.

Keep in mind that advertised speeds are usually "up to" figures for a wired connection. Receiving 80–95% of the plan speed on a wired test is normal; consistently below 50% at all times of day is worth raising with your ISP, with test results attached.