I wrote this a while ago for Low End Mac, but it might be applicable to your situation:
" I was troubleshooting a DaynaPORT NuBus Ethernet card that wouldn’t link to a modern switch on RJ45. It would start the handshake, time out, retry… Déjà vu..
I ran into the exact same thing with an Asante card a while back. Switch logs showed the card trying to link, but giving up before the switch finished its auto MDI-X dance. For testing I forced the port on my switch to 10Mb/Half Duplex. It worked instantly, but that was just a band-aid.
The real issue: these old cards use a discrete link-pulse detector (26LS32 + 74LS123) to tell the DP83902 NIC when to go twisted-pair. Modern switches take a little too long doing auto MDI-X, so the card times out and bails. Neither side ever syncs.
On the Asante, the fix was swapping R28 (it's next to D3, unpopulated on that board) from 36K → 220K. That extends the timeout window just enough for the switch to finish. Added bonus: I shut off the 10Base2 side with a jumper since I wasn’t using it anyway. It has been rock solid since.
So, backing up a little bit, when the DaynaPORT showed the same exact LS123 + LS32 + DP83902 combo, I had a hunch. I checked pin 15 of the LS123 -- sheesh, same hackish RC timing circuit. This time it was R17 (100K SMD), hiding by the AUI connector. I dropped in a 220K from my pile (0805 instead of the original 1206, but who’s counting) and… it linked, no drama. "
So the trick? If your card has a LS123, check pin 15. It should connect to VCC via a resistor and be decoupled to GND via a cap. If that's so, bump up the value of the resistor by around 200k give or take to increase the timeout and it should work.
If your etherent card is different, post a high resolution picture of it and I'll see if there's anything obvious.
" I was troubleshooting a DaynaPORT NuBus Ethernet card that wouldn’t link to a modern switch on RJ45. It would start the handshake, time out, retry… Déjà vu..
I ran into the exact same thing with an Asante card a while back. Switch logs showed the card trying to link, but giving up before the switch finished its auto MDI-X dance. For testing I forced the port on my switch to 10Mb/Half Duplex. It worked instantly, but that was just a band-aid.
The real issue: these old cards use a discrete link-pulse detector (26LS32 + 74LS123) to tell the DP83902 NIC when to go twisted-pair. Modern switches take a little too long doing auto MDI-X, so the card times out and bails. Neither side ever syncs.
On the Asante, the fix was swapping R28 (it's next to D3, unpopulated on that board) from 36K → 220K. That extends the timeout window just enough for the switch to finish. Added bonus: I shut off the 10Base2 side with a jumper since I wasn’t using it anyway. It has been rock solid since.
So, backing up a little bit, when the DaynaPORT showed the same exact LS123 + LS32 + DP83902 combo, I had a hunch. I checked pin 15 of the LS123 -- sheesh, same hackish RC timing circuit. This time it was R17 (100K SMD), hiding by the AUI connector. I dropped in a 220K from my pile (0805 instead of the original 1206, but who’s counting) and… it linked, no drama. "
So the trick? If your card has a LS123, check pin 15. It should connect to VCC via a resistor and be decoupled to GND via a cap. If that's so, bump up the value of the resistor by around 200k give or take to increase the timeout and it should work.
If your etherent card is different, post a high resolution picture of it and I'll see if there's anything obvious.
