Networking

NIC

Network Interface Card

The hardware component — integrated into the motherboard or added as an expansion card — that physically connects a computer to a network, handling the conversion between digital data and the electrical or radio signals on the wire or air.

A NIC operates at OSI Layers 1 (Physical) and 2 (Data Link). It encodes outgoing data into the signal format required by the medium (electrical pulses on Ethernet, radio waves for Wi-Fi), and decodes incoming signals back into digital frames. Every NIC has a unique burned-in MAC address used for Layer 2 frame delivery on the local network. In modern systems the NIC is almost always integrated into the motherboard chipset rather than being a separate card, but discrete PCIe NICs are common in servers and for upgrading to faster speeds like 2.5 GbE or 10 GbE.

NIC hardware components

A wired Ethernet NIC contains several functional blocks working together:

  • PHY chip (Physical Layer transceiver): handles the analogue signal processing — encoding, decoding, and line driving for the copper or fibre medium. It converts digital bits from the MAC into the modulated electrical signals placed on the cable, and vice versa.
  • MAC chip (Media Access Controller): implements the Ethernet framing standard — adding preamble, source/destination MAC addresses, EtherType, and FCS (Frame Check Sequence/CRC). The MAC enforces CSMA/CD on half-duplex links and manages the full-duplex flow control.
  • PCIe or USB interface: the host bus interface that connects the NIC to the CPU and system memory. PCIe provides low-latency DMA access to RAM; USB adds a translation layer that increases latency and CPU overhead compared to PCIe NICs.
  • RJ-45 connector: the 8P8C jack that accepts the Cat5e/Cat6 patch cable. On server NICs this may be replaced by SFP/SFP+ cages for fibre or DAC (Direct Attach Copper) connections.

NIC speeds and Ethernet standards

StandardSpeedCableTypical hardware
100BASE-TX100 MbpsCat5Legacy devices, some IoT
1000BASE-T (GbE)1 GbpsCat5eAll current consumer hardware
2.5GBASE-T2.5 GbpsCat5eGaming routers, Wi-Fi 6 AP uplinks, NAS
5GBASE-T5 GbpsCat6High-performance workstations
10GBASE-T10 GbpsCat6aServers, NAS, prosumer routers
25/40/100GbE25–100 GbpsSFP28/QSFP fibre or DACData centre servers

NIC offloading: relieving the CPU

At gigabit speeds, processing every packet in software would consume significant CPU time. Modern NICs offload several compute-intensive tasks to dedicated hardware on the card:

  • TCP Segmentation Offload (TSO): the OS hands the NIC large data buffers (up to 64 KB); the NIC splits them into MTU-sized TCP segments with correct sequence numbers and checksums. Without TSO, the CPU performs this segmentation for every segment sent.
  • Large Receive Offload (LRO): the NIC coalesces multiple incoming TCP segments belonging to the same flow into a single large buffer before passing it to the OS, reducing the number of interrupt and context switch cycles per unit of data.
  • Checksum offload: the NIC computes or verifies IPv4, TCP, and UDP checksums in hardware, removing this per-packet work from the CPU entirely.
  • Receive Side Scaling (RSS): distributes incoming packet processing across multiple CPU cores using a hash of the 5-tuple, preventing a single core from becoming a bottleneck on multi-core systems at high packet rates.

A NIC with poor driver support — or offloading features incorrectly configured — can cause high CPU usage during 1 Gbps transfers, dropped packets, or incorrect checksums that trigger retransmissions. On Linux, ethtool -k eth0 lists the offload features enabled on an interface.

NIC types

TypeInterfaceMax speedTypical use
Integrated EthernetMotherboard chipset1–2.5 GbpsDesktops, laptops, NAS
PCIe add-in cardPCIe x1/x4/x81–100 GbpsServers, speed upgrades
Wi-Fi adapter (integrated)PCIe / M.2Up to Wi-Fi 7Laptops, desktops
USB Wi-Fi dongleUSB 2.0/3.0Up to Wi-Fi 6Desktops without built-in Wi-Fi
Thunderbolt/USB-C adapterThunderbolt / USB-C1–2.5 GbpsThin laptops without RJ-45

SR-IOV for virtualisation

SR-IOV (Single Root I/O Virtualisation) is a PCIe standard that allows a single physical NIC to present multiple virtual NIC instances (Virtual Functions, VFs) directly to virtual machines, bypassing the hypervisor's software network stack. Each VM gets near-native NIC performance because packets do not pass through a virtual switch in software. SR-IOV is common in cloud and data centre servers running KVM, VMware ESXi, or Hyper-V where network throughput and latency between VMs and the physical network are critical. Consumer NICs do not support SR-IOV; it requires a NIC and platform that both advertise SR-IOV capability.

Wireless NICs: integrated vs PCIe vs USB

Laptop wireless NICs are almost always soldered directly to the motherboard or connected via an M.2 Key E slot as a small card. Desktop systems without a built-in wireless NIC can add one via a PCIe x1 card (best performance, uses an internal antenna), a USB 2.0/3.0 dongle (convenient, limited by USB overhead and small antennas), or a Thunderbolt adapter. USB Wi-Fi dongles are limited by the USB protocol's CPU overhead and the dongle's small antenna — they consistently underperform internal PCIe cards for range and throughput.

Checking NIC link speed and duplex

On Linux, ethtool eth0 reports the negotiated speed and duplex: "Speed: 1000Mb/s" and "Duplex: Full" is the expected result for a Gigabit Ethernet connection. "Speed: 100Mb/s" or "Half-duplex" indicates a cabling problem, a faulty port, or a mismatch. On Windows, Device Manager → Network Adapters → right-click the adapter → Properties → Advanced tab shows link speed. On macOS, hold Option and click the Wi-Fi icon, or use System Information → Network for wired adapters.

Wake-on-LAN and jumbo frames

Wake-on-LAN (WoL) allows a NIC to keep a small amount of circuitry powered even when the computer is off, listening for a "magic packet" (a broadcast containing the target MAC address repeated 16 times). When received, the NIC signals the motherboard to power on. WoL requires both NIC and BIOS/UEFI support, a physical Ethernet connection (Wi-Fi WoL is unreliable), and the feature enabled in firmware settings.

Jumbo frames increase the Ethernet MTU from the standard 1500 bytes to 9000 bytes (common) or larger. Larger frames reduce per-packet overhead, improving throughput for large transfers between devices that both support jumbo frames on the same switched network. All devices and switch ports in the path must have matching MTU settings — a mismatch causes fragmentation or dropped packets. Jumbo frames are most beneficial for NAS-to-workstation transfers and iSCSI storage networks.

Frequently Asked Questions

Does every computer have a NIC?

Yes. Modern devices integrate the NIC directly on the motherboard. Most laptops have both a wired Ethernet NIC and a wireless NIC. Servers typically have multiple NICs — two or four ports, often bonded together for redundancy and higher aggregate bandwidth.

What is the difference between a NIC and a network adapter?

Interchangeable terms. "NIC" originally referred to a discrete expansion card; "network adapter" is the broader term covering integrated controllers, USB dongles, and Thunderbolt adapters. Operating systems call them all "network adapters" regardless of physical form factor.

Can a NIC limit my network speed?

Yes. A 100 Mbps NIC caps throughput at 100 Mbps regardless of router or ISP speed. Most current consumer hardware is Gigabit (1 Gbps). Multi-gigabit internet plans (2.5 Gbps+) require a 2.5 GbE or 10 GbE NIC. NIC driver quality also affects CPU load and reliability during high-throughput transfers.

Related Terms

More From This Section