What are the different speeds of Ethernet?

Ethernet speed is the nominal bit rate of the physical link between two network interfaces. The family has grown from 10 Mb/s to 100 Mb/s, 1 Gb/s, 2.5 Gb/s, 5 Gb/s, 10 Gb/s and, at the data-centre end of the spectrum, hundreds of gigabits per second. The number printed next to a port is not the same thing as guaranteed application throughput. Frame overhead, the upper-layer protocol, switch queues, device processing time, storage and even a PLC communication task can all become the real bottleneck.
That distinction matters in automation. A 100 Mb/s link between a PLC and an HMI does not mean the HMI needs 100 Mb/s of process data. Replacing the link with Gigabit Ethernet does not automatically make the PLC scan or the HMI refresh ten times faster. To interpret the number correctly, it helps to understand how Ethernet works, which physical medium is in use and how much traffic the application actually produces.
The main Ethernet speed families
| Nominal rate | Common name / example PHY | Typical medium | Where it is commonly useful |
|---|---|---|---|
| 10 Mb/s | 10BASE-T | Twisted-pair copper | Legacy networks and some low-bandwidth embedded systems |
| 100 Mb/s | 100BASE-TX / Fast Ethernet | Cat 5 or better copper, plus fibre variants | PLCs, HMIs, remote I/O and many industrial devices |
| 1 Gb/s | 1000BASE-T / Gigabit Ethernet | Cat 5e or better, plus fibre | Machine networks, SCADA, engineering stations and general uplinks |
| 2.5 Gb/s | 2.5GBASE-T | Suitable Cat 5e/Cat 6 channels | Multi-gig access over existing structured cabling |
| 5 Gb/s | 5GBASE-T | Channel-dependent Cat 5e/Cat 6 and better | High-bandwidth edge devices and multi-gig access |
| 10 Gb/s | 10GBASE-T / 10GbE fibre | Cat 6A for 100 m, plus DAC and fibre options | Servers, industrial vision, high-rate logging and line backbones |
| 25 / 40 Gb/s | 25GbE / 40GbE, including BASE-T variants | Fibre and DAC are common; Cat 8 for short BASE-T channels | Data-centre access, aggregation and backbone links |
| 50 / 100 / 200 / 400 Gb/s | High-speed Ethernet PHY families | Primarily optics, DAC and specialised short-reach electrical media | High-capacity switching and data-centre fabrics |
| 800 Gb/s | 800GbE — covered by IEEE 802.3df-2024 | High-speed optical and short-reach electrical media | Very high-capacity data-centre and backbone connections |
The table is not a cable shopping list. The same nominal Ethernet rate may be delivered by several PHYs over copper, fibre or direct-attach media. A design that simply says “10 Gb/s” is incomplete: the Ethernet cable type, reach, connector, EMC environment and the actual PHY supported at each port must match.
800 Gb/s is standardised; 1.6 Tb/s is still a project as of September 2026
Older speed summaries often stop at 200 or 400 Gb/s and describe “1 Tb/s and beyond” as the next step. That is now out of date. IEEE 802.3df-2024 is a published amendment covering 800 Gb/s Ethernet.[1] The IEEE P802.3dj project goes further and includes 200, 400, 800 Gb/s and 1.6 Tb/s Ethernet. Its second Standards Association recirculation ballot closed successfully in August 2026, but the project was still in Task Force/draft work in September 2026.[2] Calling 1.6 Tb/s a completed IEEE Ethernet standard at this point would therefore be premature.
What does “1 Gb/s” actually tell you?
1 Gb/s means a nominal line rate of 1000 Mb/s. Bits and bytes are different units: eight bits make one byte. The raw bit-rate equivalent is therefore about 125 MB/s. A 100 MB file contains roughly 800 Mb, so the absolute lower bound at a perfect 1 Gb/s line rate is about 0.8 seconds. Real transfers take longer because Ethernet framing, inter-frame gaps, IP/TCP or other protocol overhead, software and storage all consume time or capacity.
This is why the negotiated link rate and application throughput should be reported separately. If a speed test shows 930 Mb/s on a 1 Gb/s link, that does not normally mean the PHY has negotiated a mysterious 930 Mb/s Ethernet mode. It is usually an application-level measurement below the nominal line rate. Our guide to the Ethernet frame explains where some of that non-payload traffic comes from.
Full duplex does not turn a 1 Gb/s port into a 2 Gb/s single stream
Modern switched Ethernet is normally full duplex. A 1 Gb/s link can therefore transmit at up to the nominal 1 Gb/s rate in one direction while receiving at up to the nominal 1 Gb/s rate in the other direction at the same time. Vendors sometimes describe that as 2 Gb/s of aggregate bidirectional capacity. It does not mean one download can run at 2 Gb/s through a 1 Gb/s interface. The directional line rate is still 1 Gb/s.
Why does a link sometimes come up slower than expected?
On twisted-pair Ethernet, link partners commonly use auto-negotiation to advertise their capabilities and select a mutually supported operating mode. A 10/100/1000 switch port connected to a 100 Mb/s PLC can therefore establish a 100 Mb/s link. Cisco's troubleshooting documentation treats speed and duplex negotiation as a core part of Ethernet link diagnosis.[3]
IEEE 802.3bz is particularly clear for 2.5GBASE-T and 5GBASE-T: auto-negotiation is mandatory, and these rates are defined for full-duplex operation.[4] The practical rule is simple: the link can only use a mode that both endpoints and the physical channel support. A Gigabit-capable switch does not make a Fast Ethernet PLC a Gigabit device.
When diagnosing a speed problem, check the NIC or Ethernet interface, the switch port, any SFP/transceiver, the patch leads and the remote device rather than blaming the cable first.
Cable category is not the speed setting
Installing a Cat 6 or Cat 6A cable does not change the electronics inside a 100 Mb/s interface. The cable must support the PHY, but the endpoints still determine which PHY is available. Ethernet Alliance BASE-T material shows the familiar progression: 10BASE-T over 100 m Cat 3, 100BASE-TX over 100 m Cat 5, 1000BASE-T over 100 m Cat 5e, 10GBASE-T over 100 m Cat 6A, and 25/40GBASE-T over 30 m Cat 8 channels.[5]
Cat 6 can support 10GBASE-T in shorter or appropriately qualified channels, but it should not be turned into a blanket “100 m at 10 Gb/s” rule. Ethernet Alliance training material distinguishes Cat 6 reach and alien-crosstalk conditions from the clearer 100 m Cat 6A case.[6] For a fuller treatment, see maximum Ethernet cable length.
Is 100 Mb/s too slow for a PLC and HMI?
Often, no. The cyclic process data exchanged by a typical PLC and HMI is tiny compared with a video stream or a large file transfer. Hundreds of registers, alarms and status values can fit comfortably within a 100 Mb/s physical link. If the screen feels slow, the limiting factor may instead be polling intervals, PLC communication load, protocol efficiency, switch congestion or the HMI project itself.
The answer changes when a cell also carries machine-vision cameras, multiple engineering clients, recipe transfers, central SCADA traffic, logging and edge computing. In that case, 1 Gb/s or 10 Gb/s uplinks may provide useful headroom. The right number comes from the traffic pattern and peak load, not from the assumption that the largest number is always best.
More bandwidth is not the same as a shorter control cycle
This is one of the most important distinctions in industrial networking. Line rate tells you how many bits the PHY can move per second. Control-cycle time depends on when data is generated, how access to the network is organised, how frames are processed in each device and when the control application runs. A generic 1 Gb/s Ethernet link does not, by itself, guarantee deterministic microsecond-level control.
EtherCAT illustrates the difference. Conventional EtherCAT uses a 100 Mb/s physical layer but processes data on the fly as the frame passes through devices, with an architecture designed for high bandwidth utilisation and predictable real-time behaviour.[7] EtherCAT G and G10 extend that architecture to 1 and 10 Gb/s for data-intensive applications.[8] That is why EtherCAT real-time performance and EtherCAT latency cannot be judged by the Mb/s figure alone.
Matching speed to the job
Simple PLC–HMI machine: if both devices are Fast Ethernet, adding a Gigabit converter merely to increase the number on the label is unlikely to improve the application. Check polling, protocol settings and errors first.
Machine cell and engineering network: a Gigabit switch backbone gives useful headroom when a PLC, HMI, engineering workstation, SCADA client and auxiliary devices share the same infrastructure. Individual 100 Mb/s endpoints can still be connected where appropriate.
Machine vision and high-rate acquisition: camera count, resolution, frame rate and compression can exhaust a 1 Gb/s link quickly. 2.5, 5, 10 Gb/s or faster interfaces can make a measurable difference here.
Inter-line or inter-building backbone: a 10 Gb/s fibre uplink may solve both bandwidth and reach/EMC problems more effectively than increasing every edge-port speed.
Data-centre fabric: 25, 50, 100, 200, 400 and 800 Gb/s Ethernet belong to a different design class from the typical PLC port and should be selected for aggregate fabric capacity rather than machine-level tag traffic.
Six checks when an Ethernet link feels slow
- Read the capability of both ports. Look for 10/100, 10/100/1000, 2.5G, 5G or 10G support in the device data.
- Check the negotiated link rate. A managed switch, operating system or device diagnostics page should show the active mode.
- Inspect duplex and error counters. Duplex mismatch is especially relevant on legacy 10/100 links.
- Evaluate the channel for the actual PHY. Length, termination quality, twists, screening and patch points matter as well as the category printed on the jacket.
- Separate line rate from application response. A 1 Gb/s link can be healthy while an HMI tag refresh is slow because the protocol or PLC task is the bottleneck.
- Find the narrowest point end to end. The NIC, switch uplink, server, storage or upper-layer protocol can each cap performance.
For the packet path itself, continue with how data is transmitted in an Ethernet network. To understand what rides on top of the link, see protocols that operate over Ethernet. If the real design choice is wired versus wireless mobility, use the separate Wi-Fi versus Ethernet comparison.
Quick answers
Is 1 Gb/s exactly ten times faster than 100 Mb/s?
At the nominal line-rate level, yes. The application itself will not necessarily finish work ten times faster because other limits remain.
Will Cat 6 turn a 100 Mb/s PLC into a Gigabit PLC?
No. Cabling cannot raise the maximum PHY rate supported by the device electronics.
Can a Gigabit device connect to a 100 Mb/s device?
Usually yes when both ports support a common mode; auto-negotiation can establish the link at 100 Mb/s.
Is my Ethernet link speed the same as my internet speed?
No. A local 1 Gb/s link can feed an internet connection that is only 100 Mb/s; the two bottlenecks are at different parts of the path.
Does 10 Gb/s Ethernet automatically improve PLC control?
No. It gives more bandwidth, while deterministic control also depends on cycle timing, jitter, device processing and protocol architecture.
Related Ethernet guides
What is Ethernet? · How does Ethernet work? · Cable types · Gigabit Ethernet · PoE · Advantages and disadvantages · Network security · Ethernet evolution
Technical sources
- IEEE 802.3 Working Group, IEEE Std 802.3df-2024 and 800 Gb/s Ethernet scope. IEEE 802.3 document.
- IEEE P802.3dj Task Force, project status including 1.6 Tb/s and the August 2026 recirculation ballot. IEEE announcement.
- Cisco, Ethernet speed/duplex auto-negotiation and troubleshooting. Cisco technical document.
- IEEE P802.3bz — 2.5GBASE-T and 5GBASE-T. IEEE Task Force archive.
- Ethernet Alliance BASE-T roadmap covering representative rates, media and reaches. Ethernet Alliance document.
- Ethernet Alliance, supported 10GBASE-T link-segment distances. 10GBASE-T training document.
- EtherCAT Technology Group, EtherCAT functional principle and bandwidth utilisation. ETG technology page.
- EtherCAT Technology Group, EtherCAT G/G10 at 1 and 10 Gb/s. ETG EtherCAT G page.