x440-g2-48p fans / temperature readings

  • 16 May 2020
  • 2 replies

Very similar to previously reported issues with fan speeds / temperature readings on 440-G2 models, however I’m running which should have all these issues addressed.

3 switch stack of 440-G2-48p.  After leaving them unplugged overnight for a test, they were reading 60C by the time they booted up - all 3 of them.  Immediately read 60 and never increased beyond that.  Rooms are 21C.  This test was with zero PoE devices connected.  In use, they are powering 3-10 devices each - not a lot of power going out.  They are cool to the touch.

Fans are running about 7500 rpm consistently on all 3.  2 of them in a rack - not a big deal.  1 node is in an office environment and is running way too loud.  I believe the issue is with the temp reading consistently high, not with the fan controller itself.  In the minute or two it takes to boot with no load, there is no way I can believe those are running 60c. 

I’ve experimented with the DEBUG HAL commands, carefully monitoring temperature on the office located one, and it holds the reported temp at significantly reduced fan speed.  So that’s a temporary fix, but I’m not happy leaving it that way.  Does anyone know if a) there is any issue with these temperatures reading high? and b) any safe workarounds for it?

Maybe this is just how fast they run - I’ve frankly never paid much attention as these are almost always installed in closets.


EDIT: another thing to mention - the noise level difference between 7500ish RPM and 5500RPM is huge.  Over 10dB.  Of course, at 100% it sounds like a jet engine...

2 replies

Userlevel 3

I would not wonder if this issue is back again in XOS 30.6.

This FAN/Temp issue was fixed even more than one time. Same with 10g flapping on X670 where debounce timers are not correct.

You can only create a case.

I hope Extreme will implement a dedicated temp sensor in X445 (and hope they have done this in X435) and will not calculate temp based on cpu values.

I considered the same - so I rolled the stack back to 30.3 (where it is explicitly addressed) with no change.  Thanks.