Ethernet slow on one machine while other devices on the same network are fast — the laptop gets 20 Mbps via cable while the router shows 500 Mbps to other devices — points clearly at the machine rather than the network. The question is whether it’s a hardware limitation (old adapter, cable), a driver issue, or a Windows configuration problem. If you want the full context, see our Complete Guide to Fixing Windows, Browser, and Software Errors.
First: verify with a speed test. Run speedtest.net or fast.com directly, not in the background. Note both download and upload. If the machine consistently gets under 100 Mbps on Gigabit ethernet while other devices get 300–900 Mbps, there’s a genuine problem. If speeds are 80–120 Mbps on a “Gigabit” connection, that might be normal — older cables or adapters often cap at Fast Ethernet (100 Mbps) rather than Gigabit.
Check the link speed first
Settings → Network and internet → Ethernet → Properties → look at the “Link speed (Send/Receive)” value at the top. This shows what speed the NIC and switch actually negotiated. If it says 100/100 Mbps rather than 1000/1000 Mbps, the connection is operating at Fast Ethernet speed regardless of what the hardware supports.
100 Mbps link speed usually means: the cable is Category 5 (not 5e or 6), the cable is damaged, one pin in an RJ45 connector is bad, or the NIC or switch port has a duplex mismatch. Try a different ethernet cable first — this is free and takes 60 seconds.
Driver — the most impactful software fix
Windows Update sometimes swaps working manufacturer Ethernet drivers for generic versions. The generic driver handles basic connectivity but may not correctly configure Offload settings, Interrupt Moderation, or Receive Buffer sizes — all of which significantly affect throughput on Gigabit connections.
Device Manager → Network Adapters → right-click the Ethernet adapter → Properties → Driver tab → check the driver date. If it changed recently: Roll Back Driver. Otherwise: download the Ethernet driver from the laptop or motherboard manufacturer’s support page. Manufacturer drivers for Realtek and Intel NICs are specifically optimised for the adapter’s hardware capabilities.
Adapter advanced settings — the hidden performance controls
Device Manager → right-click Ethernet adapter → Properties → Advanced tab. These settings control how the NIC handles traffic at a hardware level:
- Speed & Duplex: should be “Auto Negotiation” — if manually set to 100Mbps Full Duplex, that’s a hard cap. Set back to Auto.
- Receive Buffers / Transmit Buffers: increase to 512 or higher if available. Larger buffers allow the adapter to handle more data without waiting for the CPU.
- Interrupt Moderation Rate: “Adaptive” or “Moderate” balances CPU usage and throughput. “Extreme” reduces CPU interrupts but can drop throughput under heavy load.
- Large Send Offload (IPv4/IPv6): should be Enabled — offloads packet segmentation to the NIC hardware rather than the CPU.
- Checksum Offload: should be Enabled — offloads checksum calculations to the NIC.
After adjusting, restart the machine. These settings affect throughput significantly on Gigabit connections.
TCP auto-tuning
Windows TCP Receive Window Auto-Tuning optimises the receive buffer size for each connection based on network conditions. If disabled or set incorrectly — common after VPN installs, some optimisation tools, or manual changes — throughput is artificially capped even on Gigabit connections.
netsh int tcp show globalCheck “Receive Window Auto-Tuning Level” — it should say “normal.” If it shows “disabled” or “restricted”:
netsh int tcp set global autotuninglevel=normalRestart the machine. Also worth running:
netsh int tcp set global chimney=enabled
netsh int tcp set global rss=enabledRSS (Receive Side Scaling) distributes network processing across CPU cores, which improves throughput on multi-core machines significantly.
Cable quality and physical inspection
A Cat5e or Cat6 cable that works fine for 100 Mbps may not reliably carry Gigabit speeds if it has a tight bend, is too long (over 90 metres), or has connector pins that aren’t making full contact. Gigabit ethernet uses all 4 pairs of wires; a damaged pair causes negotiation to fall back to 100 Mbps.
Inspect the cable: look for kinks, sharp bends, or pinched sections. Try a different cable entirely. If the new cable immediately gives better speed, the old cable was the problem.
VPN and background software eating bandwidth
A VPN running on the machine routes all traffic through its server, adding latency and potentially capping throughput at the VPN’s server bandwidth. Background processes — cloud sync actively uploading, antivirus doing a network scan, Windows Update downloading — consume available throughput and reduce measured speed test results.
Disconnect the VPN and retest. Pause OneDrive and other cloud sync. Check Task Manager → Network tab — sort by network usage. Any process consuming significant bandwidth before the speed test is artificially reducing results. The measured speed is accurate to what’s available for foreground use, not the total connection capacity.
For the wireless equivalent of this problem, our Wi-Fi slow on one device guide covers the same driver and TCP stack fixes in the wireless context — many are identical. If slow speeds are affecting all devices on the network rather than just one machine, the problem is upstream of the router and the DNS and network diagnostic guide covers ISP-side issues. Microsoft’s network adapter performance documentation covers the RSS and chimney offload settings in detail and their interaction with different NIC hardware generations.
USB ethernet adapters — a common hidden bottleneck
Laptops without a built-in RJ45 port use USB-to-Ethernet adapters. These adapters are limited by the USB bus they’re connected to. USB 3.0 has theoretical bandwidth well above Gigabit, but USB Gigabit adapters often don’t reach more than 300–500 Mbps in practice due to USB protocol overhead, CPU processing for USB packets, and adapter chip quality.
If using a USB adapter and speeds are capped at 250–400 Mbps: that’s close to what many USB Gigabit adapters actually deliver. Only higher-end adapters (Thunderbolt Ethernet, or premium USB 3.1 adapters with RTL8156 or AX88179 chips) approach true Gigabit speeds. Also ensure the adapter is in a USB 3.0 port (blue interior) — a USB 2.0 port hard-caps the adapter at 480 Mbps theoretical / ~300 Mbps actual.
Switch port and VLAN configuration
On home networks, the switch is usually the router’s built-in switch — if all devices on it are slow, it’s not a per-machine issue. But if the Ethernet port the slow machine connects to is managed (a managed switch with VLAN or QoS configuration), that port might be configured for a lower speed or have bandwidth limiting applied.
Try a different port on the switch or router. If speed immediately improves, the original port has a configuration issue or a hardware fault. On consumer routers, all ports are typically identical — if one port is slow, it might have a failing port negotiator. On managed switches: check the port configuration for any speed limits, VLAN assignment that’s routing through a slower path, or QoS rules applying bandwidth restrictions.
Wake-on-LAN and power management affecting NIC performance
Network adapters configured for Wake-on-LAN maintain a partial active state even when the computer is sleeping. Some Windows power management configurations reduce the NIC’s operating speed to support this low-power active state even when the machine is fully awake.
Device Manager → Ethernet adapter → Properties → Power Management → uncheck “Allow the computer to turn off this device to save power.” Also look in the Advanced tab for “Wake on Magic Packet” and “Wake on Pattern Match” — disabling these doesn’t affect LAN speed directly but is sometimes part of a power configuration that inadvertently reduces NIC performance mode.
Duplex mismatch — rarely talked about but still happens
Full duplex allows simultaneous send and receive; half duplex does one at a time. A duplex mismatch — where the NIC is in half duplex and the switch port is in full duplex (or vice versa) — causes significant packet collisions and retransmissions that reduce throughput to 20–40% of expected speed.
This happens when one end is set to “Auto” and the other is manually configured. In Device Manager → Advanced tab → Speed & Duplex: if it’s set to anything other than “Auto Negotiation,” change it to Auto. The switch port configuration (on managed switches) should also be Auto. Both sides in Auto negotiation mode guarantees they’ll agree on duplex.
Measuring the real cause: WinMTR and iPerf
For persistent unexplained speed caps that survive all the above: measure the actual connection performance with proper tools rather than relying on speed test websites (which test to remote servers and include internet variables).
iPerf3 measures local network throughput directly: set it up on another machine on the same network as the server (iperf3 -s) and run it as client on the slow machine (iperf3 -c [server IP]). This shows the actual throughput between two local machines, eliminating internet speed as a variable. If iPerf3 shows full Gigabit on LAN but internet speed tests show slow speeds, the local network is fine and the bottleneck is the internet connection.
Antivirus deep packet inspection
Security software that inspects network traffic at a deep level — not just checking file downloads, but examining every packet — adds measurable latency and throughput reduction. The inspection layer processes each packet before it reaches the application, which on Gigabit connections becomes a significant CPU bottleneck.
Temporarily disable the network scanning component of the antivirus (not the full antivirus) and retest. If throughput improves significantly, the network scanning is the bottleneck. Configure the antivirus to exclude the local network range from deep inspection (the traffic between your machine and local devices doesn’t need deep packet inspection) while maintaining full inspection for internet traffic.
NIC segmentation offloads and Windows 11
Modern NICs can offload TCP segmentation, UDP segmentation, and receive segment coalescing to hardware, freeing CPU from these tasks and improving throughput. Windows 11 enables these by default, but some NIC drivers disable them in certain configurations or after driver updates.
Check in PowerShell (as administrator):
Get-NetAdapterAdvancedProperty -Name "Ethernet" | Where-Object {$_.DisplayName -like "*Offload*"}This shows all offload settings. Alternatively, in Device Manager → NIC → Advanced tab → look for “TCP Checksum Offload (IPv4),” “UDP Checksum Offload,” “TCP/UDP Checksum Offload (IPv6),” and “Large Send Offload v2” — all should be Enabled. Disabled offloads push processing to the CPU, which creates a throughput ceiling even on physically capable Gigabit connections.
IPv6 vs IPv4 speed differences
Some connections show different speeds depending on whether IPv4 or IPv6 is used for the speed test. If the ISP’s IPv6 routing is suboptimal or if the IPv6 configuration on the NIC has issues, IPv6 connections appear slower. Speed test sites like speedtest.net typically use whichever protocol the OS prefers.
Test with both: run the speed test normally, then temporarily disable IPv6 (Device Manager → NIC → Properties → uncheck “Internet Protocol Version 6 (TCP/IPv6)”) and test again. If there’s a significant speed difference, IPv6 configuration on the NIC or ISP side is the bottleneck. Re-enable IPv6 after testing — disabling it entirely affects various modern services that rely on it.
One practical note on expectations: “Gigabit ethernet” means 1000 Mbps of raw line speed, but actual TCP throughput on a Windows machine typically reaches 700–950 Mbps on a genuinely working Gigabit connection. Overhead from TCP headers, Windows scheduling, and NIC interrupt handling accounts for the rest. Consistently measuring 700–800 Mbps and expecting 1000 is a calibration issue, not a performance problem. The concerning scenarios are sustained throughput below 400 Mbps on a Gigabit connection with a good cable, or large variance (sometimes 900 Mbps, sometimes 50 Mbps) indicating intermittent hardware issues.
And if network slowness is intermittent — fine most of the time but slow during specific periods — track whether it correlates with other activity on the machine (a backup running, indexing running, Windows Update) or with other users on the network using bandwidth heavily. Intermittent slowness is usually one of those contextual factors rather than a persistent configuration or driver problem. The NIC monitoring in Task Manager → Performance → Ethernet shows real-time throughput, which catches those background consumption events that otherwise look mysterious when looking only at speed test point-in-time measurements.
Quick summary for what to check first: link speed in Settings → Ethernet Properties (is it showing 1000 Mbps?) → try a different ethernet cable → check the NIC driver date in Device Manager → run the TCP auto-tuning command → check Advanced tab for Speed & Duplex setting. These five steps cover the majority of Ethernet speed issues on Windows 11 and take under 10 minutes total. Everything else in this guide is for the cases where those first five steps don’t resolve it.
One thing worth adding: if the machine has both Wi-Fi and ethernet active simultaneously, Windows may route some traffic over Wi-Fi (if the metric is lower or auto-configured) even when ethernet is plugged in. This creates confusion where “ethernet is slow” when the traffic isn’t actually using ethernet at all. Settings → Network and internet → Advanced network settings → check which adapter has the lower metric. To force ethernet to be preferred: Ethernet → Properties → IPv4 → Advanced → uncheck “Automatic metric” → set Interface metric to 5 → do the same for IPv6. Wi-Fi should have a higher metric number. This forces all traffic to prefer ethernet when it’s connected. Our guide on Slow Upload Speed on Windows 11 covers an adjacent issue.
