Gigabit broadband has gone mainstream in the UK. In the 451 home broadband deals we tracked on 16 November 2026, 133 (about 29%) advertised 900 Mbps or more, and 217 (48%) offered at least 500 Mbps.
Yet plenty of people on those lines still see a sluggish git clone, a jittery video call or a game that lags. Usually the line isn't broken. Bandwidth just isn't the only number that matters. This post walks through the other one, latency, and the bit of maths that ties the two together.
Bandwidth vs latency in one paragraph
Bandwidth is how much data the link can carry per second: the width of the pipe. Latency is how long a single packet takes to get somewhere. We usually measure it as round-trip time (RTT), the time for a packet to go out and an acknowledgement to come back. A gigabit line with 40 ms RTT moves lots of data, but every request-response exchange still waits 40 ms. For a longer, non-technical version, see our plain-English explainer on latency versus speed.
Where latency comes from
Some of it is physics. Light in optical fibre travels at roughly two-thirds of its speed in a vacuum, about 200,000 km/s. That's around 5 microseconds per kilometre, so every 100 km of fibre path adds about 1 ms to the round trip before any equipment gets involved. A server 1,000 km away by cable route costs you about 10 ms RTT from propagation alone.
The rest comes from:
- Access technology. Copper-based part-fibre (FTTC/VDSL2) usually adds more delay and jitter than full fibre (FTTP), because of line encoding and error correction on the copper segment.
- Your Wi-Fi. A congested or weak wireless link can add more latency than the entire path to a UK data centre.
- Queueing (bufferbloat). When something saturates your upload, packets queue in your router and RTT can jump by hundreds of milliseconds.
- Routing. Your traffic may not take the shortest path to the server.
The bandwidth-delay product
Here's the part that surprises developers. A single TCP connection can only have a limited amount of unacknowledged data "in flight" at once: the window. Its throughput is capped at roughly:
throughput ≈ window_size / RTT Flip it round and you get the bandwidth-delay product (BDP), the amount of data needed in flight to fill the pipe:
BDP (bytes) = bandwidth (bits/s) × RTT (s) / 8 Some worked numbers:
| Scenario | Calculation | Result |
|---|---|---|
| 64 KiB window, 20 ms RTT | 65,536 × 8 / 0.020 | ≈ 26 Mbps max |
| 64 KiB window, 80 ms RTT | 65,536 × 8 / 0.080 | ≈ 6.6 Mbps max |
| Fill 900 Mbps at 20 ms RTT | 900,000,000 × 0.020 / 8 | ≈ 2.25 MB in flight |
64 KiB was the classic TCP window limit before window scaling. Modern operating systems scale windows automatically, so you rarely hit the old ceiling on a desktop. But the relationship still holds:
- A single stream to a distant server often can't use your whole line, because slow-start ramps up gradually and any packet loss shrinks the window.
- High RTT magnifies everything. The same loss rate hurts a 100 ms path far more than a 10 ms one.
- Lots of small requests (API chatter, package installs, chatty RPC) are bounded by round trips, not bandwidth. Fifty sequential 40 ms round trips is two seconds whatever your speed.
That's also why speed test sites open several parallel connections to a nearby server. They're measuring the line's capacity, not what one connection to the other side of the world will get. To test your broadband speed properly, use a wired connection and test at different times of day.
Measure your own RTT in Python
ICMP ping needs raw sockets or a system binary. Timing a TCP handshake gives a good proxy using only the standard library:
import socket, statistics, time def tcp_rtt(host: str, port: int = 443, samples: int = 10) -> list[float]: results = [] for _ in range(samples): start = time.perf_counter() with socket.create_connection((host, port), timeout=3): pass results.append((time.perf_counter() - start) * 1000) time.sleep(0.2) return results for host in ["example.com", "github.com"]: r = tcp_rtt(host) print(f"{host:15} median {statistics.median(r):6.1f} ms " f"jitter (stdev) {statistics.stdev(r):5.1f} ms") Each call also resolves the hostname. The OS usually caches that after the first lookup, but the odd slow sample is one reason the median is a better summary than the mean. Run it on Wi-Fi and then on Ethernet. If the Wi-Fi median or jitter is much higher, your problem is in the house, not on the line.
Bufferbloat check: run the script while a large upload is in progress. If RTT jumps from, say, 15 ms to 200 ms, your router's queue is the bottleneck. Enabling smart queue management (SQM) on routers that support it often fixes this.
What this means when choosing broadband
- Don't buy bandwidth to fix latency. Going from 500 Mbps to 1.6 Gbps won't make your video calls smoother if Wi-Fi or bufferbloat is the issue.
- Full fibre generally helps with latency and jitter compared with copper-based connections, and usually gives much better upload speeds, which matters for pushing builds, backups and streaming.
- Gamers and remote workers should weigh stability and upload as much as headline download. Our guide to broadband for online gaming covers what to look for.
- Wire what you can. Ethernet to your desk or console removes the most common source of extra delay.
TL;DR
- Bandwidth = how much; latency = how soon.
- Single-connection throughput ≈ window / RTT, so distance and loss cap real-world speed.
- Round-trip-bound workloads don't get faster with more Mbps.
- Measure RTT and jitter on wired and wireless before you upgrade.
Paul Madden is Founder and Publisher of Broadband Provider. Independent UK broadband comparison: live deals scored on speed, price and contract terms, plus plain-English guides. https://broadbandprovider.co.uk
