Speed tests are the first thing anyone reaches for, which is reasonable — they’re free and they take thirty seconds. The trouble is that the headline number is the least useful thing on the screen, and the numbers that would actually explain the complaint are either buried or missing entirely.

This is the guide to reading them properly. Our own test is on the next page over, and it reports the buried numbers on purpose.

Copper And Fiber Are Not The Same Machine

Almost everything confusing about speed test results traces back to the physical thing carrying the signal. Three of them are common around here, and they fail in completely different ways.

Connection	How it behaves	Where it runs out
Coax — cable internet	Electrical signal over shielded copper, shared with a neighbourhood group. Far more of the available spectrum is allocated to sending data to you than to carrying it back.	Upload first, always. Then evening congestion, because the capacity is shared.
Twisted pair — DSL	Electrical signal over ordinary phone wiring. It degrades with distance: the same product is genuinely faster near the cabinet than a mile away from it.	Everything, fairly early. Distance sets the ceiling and no equipment change moves it.
Glass — fiber	Light in a strand. Distance is effectively irrelevant at city scale, and the medium has no inherent reason to favour one direction, which is why fiber plans can be symmetric.	Rarely the fiber. Nearly always something you own on your side of it.

That last row is the important one, and it is the single most common thing we end up explaining. When someone’s brand new fiber tests at 180 Mbps, the fiber is almost never what’s wrong. It is the Wi-Fi, or a router that negotiated its WAN port at 100 Mbps, or a five-year-old router that cannot actually route at line rate no matter what you feed it.

Why One Test Size Cannot Serve Both

Most speed tests transfer roughly the same amount of data for everybody. That is a compromise, and it breaks at both ends of the range.

This is why our test asks which one you have before it starts. Copper gets four connections, shorter transfers and a hard look at upload. Fiber gets eight connections and much larger ones. Both discard the first seconds rather than averaging the ramp in, because including it is how a test quietly under-reports a fast connection.

The Five Numbers, In Order Of Usefulness

Not the order they’re usually displayed in.

1. Latency under load — the one that’s usually missing

Round trip time measured while the connection is saturated, rather than while it’s idle. This is the number that explains most “my internet is fast but the call is terrible” complaints, and most speed tests do not report it at all.

The gap between idle latency and loaded latency is called bufferbloat. It happens because modems and routers were built with buffers far larger than they need. When traffic arrives faster than the line can send it, a sensible device drops a little of it, which tells the sender to slow down. A bloated one queues it instead — sometimes seconds of it. Nothing is lost, so throughput still looks excellent. But every packet now waits behind that queue, including the audio of your call.

The result is a connection that tests at 900 Mbps and adds 300 ms of delay the instant anything transfers in the background. A well-behaved 50 Mbps line will beat it on a video call every time.

Latency increase under load	Grade	What you’ll notice
Under 30 ms	A	Nothing. Calls hold up while the line is busy.
30–60 ms	B	Fine in practice.
60–100 ms	C	Occasional stumble on a call during a large transfer.
100–200 ms	D	Calls degrade whenever anything else is uploading.
Over 200 ms	F	The call drops when someone sends a file. This is the classic case.

The fix is a queue management setting — often labelled QoS, sometimes SQM, and implemented by algorithms called fq_codel or CAKE on equipment that supports them. It is a configuration change, not a purchase. Buying a faster plan does not help and frequently makes it slightly worse.

2. Upload

Your side of every video call, every cloud backup, every large attachment and every file you drop into a shared drive. On copper it is scarce: a download figure in the hundreds routinely sits alongside an upload in the teens or lower.

It also fails in a way people misread. When the upload saturates, everything feels broken — pages stop loading, the call freezes — because the acknowledgements that keep your downloads flowing are themselves trying to get out through a full pipe. The symptom looks like a total outage; the cause is one direction being full.

3. Jitter

How much the round trip varies between one measurement and the next. Audio and video care more about consistency than about speed: a steady 60 ms is comfortable, while a connection bouncing between 20 and 180 ms produces choppy audio and frozen video even though its average looks fine. High jitter on Wi-Fi is usually interference or a weak signal. On a wired connection it points at the line.

4. Latency, idle

Round trip time on a quiet connection. This is what makes a connection feel responsive rather than fast, and it is mostly a property of the technology and the distance involved. Fiber and cable to a nearby edge are typically in the low tens of milliseconds. DSL and fixed wireless sit higher. Geostationary satellite adds roughly half a second in each direction that no amount of bandwidth will remove — low-earth-orbit services are far better but still more variable than a wire.

5. Download

Last, deliberately. It is the number on the advertisement and the one people quote, and for most complaints it is the least relevant thing on the page. It matters for large downloads and for several 4K streams at once — roughly 15 to 25 Mbps per stream depending on the service. A single video call needs only a few Mbps in each direction. If your download figure is in the hundreds and something still feels wrong, the answer is in one of the four numbers above it.

Why Two Speed Tests Disagree

They usually both measured something real. They just measured different things.

Cause	What’s happening
Different servers	A test to a server across the country measures a longer path with more chances for congestion. A test to a nearby edge measures your access line, which is the part you can actually do something about.
One connection vs. several	A single connection is limited by how much data can be in flight at once, which falls as distance rises. On a fast line, tests using several connections report higher figures — and are closer to the truth about the line’s capacity.
Wi-Fi in the path	The most common cause by far. A wireless client rarely sustains much more than half its advertised link rate, and far less through a wall. If you tested over Wi-Fi, you measured Wi-Fi.
Browser and device	An older laptop, a busy CPU, a battery-saver mode or a browser extension can each cap a multi-gigabit result. Above about a gigabit, the computer starts to be the bottleneck.
Time of day	Shared coax slows in the evening in a way fiber generally does not. One reading is a snapshot, not a verdict.
Megabits vs. megabytes	Plans are sold in megabits per second; downloads are usually shown in megabytes per second. Divide by eight. A 100 Mbps connection downloading at 12 MB/s is behaving exactly as sold.

What Good Actually Looks Like

Rough expectations, wired, with the connection otherwise quiet. These are what a healthy line of each type tends to look like — not a promise about any particular plan.

Connection	Typical download	Typical upload	The thing to watch
DSL	10–100 Mbps	1–20 Mbps	Distance sets the ceiling. If you’re far from the cabinet, no equipment fixes it.
Cable / coax	100–1000 Mbps	10–50 Mbps	Upload, and evening congestion.
Fiber to the home	300–2000+ Mbps	Often matches download	Whether your own equipment can keep up with it.
Fixed wireless	Varies widely	Varies widely	Latency and jitter more than speed.

One number worth committing to memory: a gigabit Ethernet port carries about 940 Mbps of usable throughput once protocol overhead is accounted for. If you pay for a gigabit plan and test at roughly 940, nothing is wrong — you have found the port. If you pay for more than a gigabit, that port is your ceiling until every piece of hardware in the path is 2.5 Gb or better, including the cable and the network adapter in the computer.

Before You Call The Provider

Worth doing yourself, in this order:

1. Test again on an Ethernet cable, plugged straight into the router. If the wired figure is much higher than the wireless one, the problem is Wi-Fi coverage — a different problem with a different fix. 2. Close what’s running. Cloud sync, backups and game updates are all measured too. 3. Check the negotiated speed of the router’s WAN port; a port that came up at 100 Mbps is a five-second explanation for a very confusing week. 4. Restart the modem and router, in that order, once. 5. Run the test three or four times across a day and keep the results — a single reading proves nothing to a provider, and a pattern proves quite a lot.

If the wired numbers match the plan and something still feels wrong, look at the loaded latency. That is usually where the answer has been the whole time.

For Businesses, The Question Is Different

On a business connection the interesting number is rarely the peak. It’s whether the connection stays predictable while everyone is using it — and what happens when it doesn’t.

A shared residential-grade line in a commercial space will test beautifully at 7am and fall apart at 2pm. If phones run over the internet, bufferbloat and jitter are not an annoyance but an outage: voice is the first thing to break and the most visible when it does. And the plan’s advertised speed says nothing about whether there is any commitment behind it, or how long a fault takes to fix — which is the difference between a consumer product and a business one.

Sizing a connection, separating voice and card traffic from everything else, and configuring the firewall to prioritise what matters is the same conversation as the speed test, and it’s covered on our networking and security page. If the connection drops often enough to matter, a second line with automatic failover is usually cheaper than the downtime it prevents.

What This Costs

Reading a speed test with you, checking router settings, turning on queue management and testing the result is usually a remote session, billed in 15-minute increments. Anything involving cabling, a survey, or replacing hardware is on-site.

Service	Rate
Remote — residential	$125/hr, 15-minute increments
Remote — commercial	$175/hr, 15-minute increments
On-site — residential	$125/hr, 1-hour minimum
On-site — commercial	$175/hr, 1-hour minimum
Cabling and access point installation	Quoted per job

Full rates are on the pricing page. If the answer turns out to be “call your provider, here is exactly what to tell them,” we will say so — that conversation is often the whole job.

More Guides

📞 727-579-4335  ·  ✉️ support@baygeeks.com  ·  Privacy Policy  ·  Terms of Service