CRS312-4C+8XG and Intel E810-xxvda2 no link

I have no link with an CRS312-4C-8XG (firmware 7.23beta5) to a server with an Intel E810-XXVDA2 (Windows 2025 Server). I’m using a Mikrotik S+RJ SFP+10Gbase Copper installed in the E810 NIC connected to an copper (not Combo) port on the switch. I also habe an 10GB Marvell AQction RJ45 NIC on that server that only connects, when the negociated speed is set to 1GB full duplex (no 10GB connection at all).
I’ve tried more combinations with the E810:

  • DAC CAB-10GSFP-P3M 10Gbase CU DAC 3-m 30AWG von 10GTek
  • DAC CAB-ZSP/ZSP-P2M 25G-SPF28 DAC 2-m 30AWG von 10GTek
  • AXS85-192-M3 10G SFP+SR 850nm 300m LC von H!Fiber

All these combinations are working with an older NetGear XS708T switch, but not with the Mikrotik CRS312-4C-8XG. I’ve tried after the Intel E810 also a newer Intel E830-XXVDA2 (newest firmware) but also no link with the mikrotik CRS312

With Intel X710-T2L (2 Port 10GB/s, RJ45) I have link only when I set the NIC-Port(s) to 1GB/s. All other speeds (2.5/5/10/auto) doesn’t work (no link). It look like that this switch (CRS312-4C-8XG) hat problems with Intel NICs with more than 1GB/s. I also tried the firmware V7.23-beta5, but the result ist the same as with the stable firmware V7.22.1

I found the issue!

The CRS312-4C-8XG switch has two power supplies. When powering on the switch, you need to connect both power supplies simultaneously. If you connect only one power cable first and then the second one a few seconds later (which is common since the switch has no power button), the switch fails to negotiate 10 Gb/s and only connects at 1 Gb/s with some NICs.

My Intel 10 Gb/s NICs (E830-XXVDA2, E810-XXVDA2, X710-T2L) are unable to connect at all—not even at 1 Gb/s—under these conditions. Interestingly, once the switch has finished booting, you can disconnect either one of the power supplies and the switch continues to work normally. It seems that providing power from both sources is mandatory only during the boot process.

Is this a known hardware limitation or a case for an RMA?

"To clarify: connecting the switch with only a single PSU does not work. I tested this by connecting only PSU1, and then separately by connecting only PSU2. In both cases, I was unable to establish 10 Gb/s links.

However, if I power on the switch with both PSUs simultaneously (using a power strip), everything works perfectly across various connection types, including LC fiber, DAC SFP+, and DAC SFP28.

Interestingly, after a successful boot with both PSUs, I can disconnect one of them and then perform a reboot via the switch's web interface; it continues to function without any issues. The problem only occurs during a cold start if only one PSU is connected and the power was previously completely cut."

So, if I get this right:

  1. one of the two PSU fails (for whatever reasons)
  2. there is a blackout
  3. the boot when power resumes (with only one PSU actually working, the other failed) is not "full", i.e.:
  4. no link on SFP+ port(s)

An interesting implementation of redundancy in power supplies. :roll_eyes:

Maybe there is a need of cycling the SFP(s) after booting, but it seems to me a (rather serious for actual nework resilience) bug.