Custom TX2 Board with 100M Ethernet Instead of 1000M

Hi everyone!

I need 100M Ethernet in my custom NVIDIA Jetson TX2 carrier board due to using proprietary M12 8 pin connector with PoE-like option. It means that i have 2 diff pairs of Ethernet (green and orange) and 2 pairs for power (brown and blue).

I’ve read all datasheets on TX2 module and reference design guides and haven’t find an answer: what to do if i don’t need 4 differential pairs of ethernet? Reference design and most boards on market offer to connect 4 MDI[3…0] pairs to GBE transformer and then route it into RJ45 (or jack with magnetics). Also there was a topic but that guy didn’t use ethernet in general. - https://devtalk.nvidia.com/default/topic/916758/jetson-tx1/tx1-custom-carrier-board/

Please, give a piece of advice on such a situation. Right now i did such a thing, but have serious concerns about being right:

Should i add 4-pair GBE transformer and still use only needed pairs, or add testpoints to the currently unused carrier MDI pads, or just leave as i’ve already made?

Best wishes,
Victor Grigoryev

Hi, the reference design in OEM DG has been validated on Jetson TX2, while others not, so we can’t provide suggestion for other designs. Better to follow the reference design well, otherwise customer will have to handle that by themselves.

In other words, an official position about making 100M ethernet on the custom TX2 carrier board is unknown and NVIDIA don’t know what to do, isn’t it?

OK then, do you have any suggestions or solutions for Power-over-Ethernet connection for carriers?

As you can see in OEM DG, the reference design of ethernet part is compatible to 100M mode and so there is no dedicated ref design for it in guide.

There is no reference design for POE on TX2 as it is a kind of third-party design which should be able to convert input voltage into the range of what TX2 requests (5.5v~19.6V) and the power delivery is also in the range of what TX2 requests (15w (Max-P)).

100M mode is compatible with 1000M on OEM DG for black mini-ITX board: it is easy to check by making cable with just 2 diff pairs instead of 4. I did it and TX2 in linux console answered that 100M connection is online and working 100% well

Its schematics contains simple connection between TX2 and RJ45: TX2 connector <-> 4 diff pairs to 1000М magnetics <-> 4 diff pairs to RJ45. For an example that i’ve just mentioned, we are plugging in true 100M cable into carrier and everything is working with 2 spare diff pair traces on PCB, they are kinda floating

My question is about wiring my own board in a proper way. I don’t know how TX2 handle floating MDI signals with no 1000M but 100M magnetics if i need only 100M connection =( Quoting myself,

P.S. default PoE limitations about power consumption of 16.5W are easy to overcome by NOT using PoE in a classical way:

Topic up! Problem is still not solved! I really don’t know what to do with MDIO connection and need real technical assistance.

If not, there is a probability of changing computation platform because i can’t risk with PCB design of such a vital external interface as ethernet…

Per datasheet of BCM54610, just keeping GBE_MDI0 and GBE_MDI1 pairs connected will be OK for 100M mode. But better to keep connections of GBE_MDI2/3 on initial board (in serial with a 0 ohm resistor to MDI0/1 lines) as backup.

Thank you so much for an finding needed document! Now it totally make sense by using two full diff pairs instead of separating MDI2 and MDI3. By the way, i did some tests on a dev kit by cutting some traces just like on my table at the 1st post and it kinda worked…

…but now it is totally unnecessary. I’ll make connection just like you quoted datasheet: MDI0/1 for diff green and orange pairs, 2/3 - resistors to 0/1. Thank you a lot!