Qm8790 odd behavior

[bprince]

hey all,

inherited a QM8790 IB switch. seems to work ok, except for a couple odd behaviors.

1. unless I have a module plugged into port 11 none of the other ports will come up
2. seeing some link flapping. we didn’t splurge to get Nvidia AOCs. the BER on our third-party AOCs looks decent 1e-9 or better, FEC looks decent no more than bin 2:3. is there some way to set the criteria for unhealth link to keep it from resetting? I assume this is what is causing the flapping

thanks in advance

[ssimcoejr]

Hi bprince,

Welcome, and thank you for posting your inquiry to the NVIDIA developer forums!

Unfortunately, third-party cables are not supported with our switches, unless they are specifically noted in the firmware release notes (link). Different manufacturers have different specifications, our switches cannot account for all of them. Historically, we have seen issues (including high effective BER, port flapping, and failure to link up) with certain third-party manufacturers of cables that are not in our test matrix.

If the link is flapping, this is a physical issue - there’s no configurable setting that can resolve this. Our PHY firmware will automatically attempt all possible tunings within its spec (IE: increasing bias current, etc) to maintain a proper signal prior to a link flapping.

Regarding the port #11 behavior, that could either be a hardware issue or a cable compatibility issue. It’s not a behavior we’ve seen before.

If you have a valid support contract, and are experiencing these issues with supported cables (per that link above), we highly recommend opening a ticket with our NVIDIA Enterprise Experience team for further triage.

Best regards,
NVIDIA Enterprise Experience