Cabling · physical-layer fingerprint

Your ethernet is capped at exactly 90 Mbps.

Not 89, not 91. Consistently 94 down, sometimes 90, never higher. The number is so specific it tells you exactly what happened. This article is a 5-minute read and an even shorter fix.

The diagnosis, in one paragraph

Your wired ethernet link has auto-negotiated down from Gigabit (1000BASE-T) to Fast Ethernet (100BASE-TX). The Fast Ethernet standard's theoretical maximum is 100 Mbps; practical maximum after framing overhead is around 94 Mbps, which is why the cap is so consistent across devices and tests. The downshift happened because the cable, the connectors, or the NIC failed the Gigabit negotiation, and the two endpoints fell back to the slower standard that requires fewer working wire pairs.

The cause is almost always physical. Almost always a cable. Almost always fixable in ten minutes.

Why the cap is exactly this number

Gigabit ethernet (1000BASE-T) requires all four pairs of wires in a Cat5e or better cable, all four pairs working at full signal integrity, end to end. Fast Ethernet (100BASE-TX) requires only two of the four pairs and tolerates more signal degradation.

When your NIC and the router or switch on the other end try to negotiate a link, they exchange test signals on all four pairs. If any pair fails — a broken wire, a damaged connector, excess length, severe interference — they don't negotiate at a "slightly degraded" Gigabit speed. The standard doesn't allow that. They drop down to the next supported speed that works, which is 100 Mbps.

Software reports your link as "Connected at 100 Mbps." Speed tests, downloads, and file transfers all top out at ~94 Mbps because that's the practical ceiling at the negotiated link rate. The 6 Mbps gap between 100 Mbps advertised and 94 Mbps real is ethernet framing overhead, plus the small inefficiencies of TCP.

The 30-second verification

Before you do anything else, confirm the link is actually at 100 Mbps. This rules out the possibility that the router or the ISP is the bottleneck.

Windows. Open Settings → Network & Internet → Ethernet → click on your connected adapter. Look for "Link speed (Receive/Transmit)." If it reads "100/100 (Mbps)," the diagnosis is confirmed. Gigabit would read "1000/1000."

macOS. Open System Settings → Network → Ethernet → Details. Look for "Link speed." Same pattern — 100 Mbps means downshifted, 1 Gbps means full.

Linux. Run ethtool eth0 (substituting your interface name). The "Speed:" line shows the negotiated rate. 100Mb/s confirms the fingerprint.

Once you see "100 Mbps" reported by the OS, you know the problem is between the NIC and the router's port, and you know it's a physical-layer failure. The remaining question is which physical thing.

The 80% case: replace the cable

The single most common cause is a damaged patch cable between the wall jack (or router) and the device. Patch cables get bent, crushed under furniture, chewed by pets, and pinched in office chair wheels. The damage is often invisible from the outside.

The fix: swap the patch cable with a known-good one. Use a pure-copper Cat6 patch cable (affiliate) rather than CCA (copper-clad aluminum). Unplug the ethernet on both ends, plug in the new cable, re-check the link speed in the OS. If it reads "1000 Mbps," the old cable was the problem. Move on.

If the cap returns after the new cable, the damaged cable is somewhere else — most likely in the wall, between the router and the wall jack, or at a damaged keystone jack.

The next-most-common cases

01

The in-wall cable run has degraded

Why this triggers the downshift

A cable inside a wall can fail from age, rodent damage, or installation damage that wasn't visible at the time. Common in older Cat5 runs that worked fine for years until one pair finally failed below the threshold the NIC accepts.

What to do about it

If you can access the cable on both ends (the wall jack and the patch panel or other end), test continuity with a basic cable tester. Replace the cable if a pair has failed. If the cable is unreachable, run a new one along the baseboard as a temporary fix and plan to pull a replacement when you have the chance.

02

A keystone jack or wall outlet has broken pins

Why this triggers the downshift

Wall jacks are punched-down terminations. The contact pins are small spring-loaded teeth that can break or bend. Insert the patch cable a few times in and out and a previously-fine jack can fail one pair.

What to do about it

The cheapest fix is to replace the keystone jack. They cost a couple of dollars each, install in five minutes, and reuse the existing in-wall cable. Toolless keystone jacks make this easier.

03

A patch panel termination has lost a pair

Why this triggers the downshift

If you have a structured wiring setup with a patch panel, the punch-down at the panel can fail the same way a keystone jack can. Often the issue with a setup that worked fine for years and then 'started slow.'

What to do about it

Re-punch the affected port. A 110-style punch tool costs $15 and the re-termination takes a minute per pair. While you're there, verify the pair colors match the T568B standard on both ends.

04

Excess cable length

Why this triggers the downshift

The Gigabit standard guarantees performance up to 100 meters (330 feet) total — including all the cable in the walls and the patch cables at both ends. Pushing past this with cable stretched over multiple runs can occasionally trigger the downshift.

What to do about it

If your total cable run is over 100 meters, the right answer is a switch in the middle of the run. The runs from the switch to each endpoint then meet the standard individually.

05

Cable run parallel to power for long stretches

Why this triggers the downshift

Cat5e and Cat6 unshielded cables tolerate moderate EMI. Running them parallel to a 240V power line for tens of feet can induce enough noise to fail Gigabit negotiation on the affected pair.

What to do about it

Either separate the cables physically (12+ inches), use shielded cable (Cat6a S/FTP), or cross power lines at 90 degrees if you must intersect them.

06

A failing NIC

Why this triggers the downshift

Less common but real. NICs degrade with age, lightning strikes, or physical damage to the port. The failure mode often looks identical to a cable problem because the NIC's transmit or receive on one pair has gone marginal.

What to do about it

Try the suspect device on a known-good cable plugged directly into the router. If the cap persists, the NIC is the problem. USB Ethernet adapters are a $20 workaround; replacing an internal NIC is a different conversation.

07

A failing router or switch port

Why this triggers the downshift

Router LAN ports occasionally fail in ways that look like cable issues. The port works, but only at 100 Mbps.

What to do about it

Move the cable to a different LAN port on the router. If the cap disappears, the original port has failed. If you have a separate switch, swap ports there too.

What to do if the cable is in-wall and you can't easily replace it

If the run is buried in a wall, runs through a finished ceiling, or otherwise can't be replaced without significant disruption, three workarounds:

  • Add a switch and bypass the bad pair. A small unmanaged switch placed mid-run lets each segment be its own short, working cable. Works when the bad pair is in a specific section and you can put a switch on either side.
  • Run a new cable along the baseboard. Not aesthetically ideal, but solves the throughput problem immediately. White flat cables blend in reasonably well; cable raceways hide them if needed.
  • MoCA over the existing coax, if there are coax jacks in the relevant rooms. See the MoCA vs ethernet article for the trade-offs. MoCA delivers gigabit-class throughput over existing wiring.

Tools that make this easier

For anyone who deals with ethernet runs more than once a year, two tools earn their cost:

  • A basic RJ45 cable tester (affiliate). Costs $15-30. Tells you within seconds which pair has failed, so you don't have to re-terminate every pair on a suspect connection.
  • Toolless keystone jacks (affiliate). Replace a damaged wall jack in five minutes without a punch-down tool. Don't fuss with the bad jack; just swap it.

When the diagnosis is the wrong one

The "exact 90 Mbps cap" fingerprint is reliable, but related-looking caps mean different things:

  • Cap around 500-600 Mbps: probably your router's CPU is the constraint. See the ASUS gigabit-cap article or the TP-Link article.
  • Cap at exactly 100 Mbps but only on Wi-Fi: not this fingerprint. Could be a per-device Wi-Fi problem — see PC Wi-Fi slow / phone Wi-Fi fine.
  • Variable cap depending on time of day: ISP congestion or appliance interference (powerline installs only). Not a cable problem.

After the fix

Once link speed is back at 1 Gbps, run the StabilityPulse stability test to confirm jitter and loaded latency look healthy too. A cable problem causing the downshift sometimes leaves marginal-but-just-good-enough signal characteristics on the remaining pairs even after the link comes back to Gigabit. If jitter or loaded latency is unusually high even though throughput is restored, the cable might still be the right thing to replace.