802.3 requires 1000BASE-T interface to do autonegotiation. But many network gear can disable autoneg and force 1000M-FDX operation for 1000BASE-T interface (such as "no autonegotiation / duplex full / speed 1000" ).

Despite that the spec requires autoneg and many gear venders also recommend to enable autoneg, my colleague insists on disabling autoneg. He prefer fixed configuration because autoneg-enabled 1000-T port sometimes "falls back" into 100/10M mode due to bad cable (for example one of 8 wires in Cat5 cable is cut).... Furthermore autoneg-disabled 1000BASE-T link in our lab is apparently working well.

My question: Does something wrong occur if I set "no autoneg" to each end of 1000BASE-T port? (e.g. BER increase or link flap ?)

AFAIK 1000BASE-X's autoneg includes remote fault signalling which detects uni-directional fiber cut. It makes senses to enable autonegotiation for all 1000BASE-X link.

  • also, have a look at udld May 27, 2016 at 21:31
  • Did any answer help you? If so, you should accept the answer so that the question doesn't keep popping up forever, looking for an answer. Alternatively, you could provide and accept your own answer.
    – Ron Maupin
    Aug 14, 2017 at 4:28

2 Answers 2


Autonegotiation should only be disabled where necessary. Properly compliant 802.3 hardware will send and respond to information in link pulses. Only in rare instances -- i.e. metro-ethernet -- are link pulses missing. And it is a violation of 802.3.

Setting speed/duplex does not disable negotiation; it limits what is advertised.

Note: The Cisco 2960S's at my fingers don't even support turning off negotiation. The Adtran hardware, luckily, does. (or I'd have no internet access.)

  • Right, and for access switches you may have endpoints plug in that have various capabilities so you really want auto-negotiation. May 27, 2016 at 21:25
  • Setting speed/duplex does not disable negotiation; it limits what is advertised. - at least on Fast Ethernet equipment it does disable Auto Negotiation and works by carrier detection only. On nearly all hardware I've worked with it's still that way for 10/100 speeds. It's only changed for 1000BASE-T or faster - because AN is mandatory there.
    – Zac67
    Jan 13 at 7:51
  • The question wasn't about FastEthernet. With FE, it's implementation specific how explicit speed/duplex effects autoneg. With gig, AN is required, but can still be explicitly turned off. (at least, expecting AN can be turned off. as I mentioned, some metro-e doesn't follow the rules.)
    – Ricky
    Jan 13 at 12:25

In all the PHYs that I have worked with, if Auto-Negotiation is disabled and speed is forced to 1000 Mbps, the PHY treats this as if Auto-Negotiation is enabled and the advertisement is limited to 1000 Mbps.

For example, here's the description of the Auto-Negotiation Enable bit in the Copper Control Register in the Marvell Alaska family of PHYs:

If Register 0_0.12 is set to 0 and speed is manually forced to 1000 Mbps in Registers 0.13 and 0.6, then Auto-Negotiation will still be enabled and only 1000BASE-T full-duplex is advertised if register 0_0.8 is set to 1, and 1000BASE-T half-duplex is advertised if 0.8 is set to 0.

Auto-Negotiation is mandatory per IEEE for proper operation in 1000BASE-T.

There may exist some PHYs that really do allow Auto-Negotiation to be disabled and 1000 Mbps to be forced, but these PHYs would require manual configuration of the master/slave clock relationship (which is the reason Auto-Negotiation is nominally mandatory for 1000BASE-T).

PHYs that implement a proprietary "downshift" feature often allow that feature to be disabled for cases where no link is better than a slow link. Whether the network operating system exposes this option to administrators is implementation-dependent.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.