Intel 82578DM NIC only Connects After Cold Boot

[SYNACK]

Intel's latest batch of NICs is atrocious on our networks in Windows 7. It will just sit there trying to get an IP address forever unless it has been cold booted. This gets a bunch more reliable when you upgrade the BIOS and use the very latest drivers from the Intel site but the issue still persists.

 

Looking in the drivers advanced properties I think that I have found the culprit. For some unknown reason Intel have decided that a good default value for their "Wait on AutoNeg Complete" setting would be auto which defaults to just reporting that the card is ready whether it is or not.

 

Setting this to on seems to have made these cards much more reliable. In our environment at least the setting is stored in a standard location in the registry:

 

 
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4D36E972-E325-11CE-BFC1-08002BE10318}\0007]
"WaitAutoNegComplete"="1"

 

I have attached the REG file that also resets the default option to on. Hopefully it can help someone else. (Fitting, the attachment is 666 bytes and it is an evil problem )

82578DM_NIC_WaitAutoNeg.txt

[SYNACK]

False Alarm, its still not completely fixed but it is a massive improvement and works almost every time now across 30 machines. Dirty Intel, they can make CPUs but drivers seem to be a bit above their pay grade.

[m25man]

What switch are they connecting to?

 

These are Gigabit adapters, as such the origional 15yr old Auto-negotiate specification doesn't apply and when interfaced with a 10/100 switch port you can have exactly the issues described due to duplex mismatch issues.

 

If you disable Auto Negotiate on the NIC driver you will need to match the switch port otherwise you will see a lot of discards and drops on the switch.

 

It's a mess, not entirely Intels fault but due to the fact that a revised IEEE spec for auto negotiation just hadn't been correctly interpreted by all manufacturers yet.

 

Autonegotiation - Wikipedia, the free encyclopedia

 

If mated to a modern gigabit switch do they work correctly?

[SYNACK]

They are mated to a gigabit switch, but the ones that cause the most issues are the ports called to a rubbish D-Link switch. This issue does still show up on a Cisco gigabit switch though so I am blaming Intel for this one, the Broadcom chips work fine in the same situation.

[m25man]

You have my sympathy, after years of trouble free networking here we switched to IP phones these have a two port POE aware pass thru port for the PC.

So we have 10/100 HP POE switch to Mitel 10/100/1000 pass thru to Lenovo Desktop (Broadcom) Gigabit

All of the HP's are now logging Duplex Mismatches and the PC's are running much slower but the phones are fine!

I'm having to slowly visit each user and validate the ports/outlets/settings one by one to sort this mess out.

 

The old HP 10/100 switches were thrown out and replaced with POE ones to power the phones, but now we are having to add cabling and gigabit switch gear to separate the phone/data traffic.

It's costing £1000's, good job the owners are billionares...