Interesting short circuit

[donileo]

So I'm diagnosing an issue with a server that I own which is failing to connect some hard drives to the internal controller card that manages them. The sata connection  board (SuperMicro BPN-SAS-826A: https://www.supermicro.com/manuals/other/BPN-SAS-826A.pdf) has three data (SAS) connectors for connecting external hard drives to an internal card but the board itself is only working for 2 out of the 3 data connectors. Power to the board is fed via three standard 4-pin (12v, GND, GND, 5V) molex from the power supply. I proceed to test the side responsible for the drives missing (4 drives, all on the right side, managed by data connector #3 and power molex #3) by removing the board since I've isolated the issue to the board - no power supply is connected to the board and molex connectors on the board are male. When testing, I notice the 5V DC male molex pin is shorted to GND with a resistance of ~46Ω which strikes me as very odd and as the most likely culprit for the issue. I look around for the most likely component causing the issue and trace it to be 2 Panasonic Polymer SEP 300 600µF 16V capacitors which measure that same resistance across their pins while in circuit. 4 other capacitors of these same type and same purpose measure as being OPEN while only those two don't / are shorted. The board is pretty symmetrical in its layout and function so its pretty easy to see which components are meant for the same thing on the middle and left side of the board.

So finally time for my question: Given that the 5V DC line is shorted to the ground pins I was wondering why in the world I got no error or audible feedback when the machine was previously turned on. I went ahead and measured the voltage across the 5V DC line when connected to the shorted 5V / GND pins and power to the system but lo and behold there is actually 5V there on the line and the 5V line is NOT shorted to ground. So how can that be? With power off, GND is shorted to the 5V pin, with power on, the short disappears.. Huh?

As a another note, the two other molex connectors #1, #2, do not have 5V pin shorted to ground, drives connected there work fine.

Is this a behavior of those two bad Polymer caps? Short circuit when no power, open circuit when power applied?
Maybe a behavior of the power supply; which the manual states has short circuit protection?

[Stray Electron]

  My guess would be that even with the shorted caps, the DC power supply is able to supply enough current to keep the 5V rail at 5 volts (or near enough) so that it doesn't trip the power failure detection circuit. Most likely, the PCB circuit runs that connect to the two shorted caps have just enough resistance in them to limit the current to a level that the 5V PSU is able to deal with.  46 Ohms at 5 volts is only about 1/10 of an amp of current and any modern PC power supply can easily provide far more current than that.

[amyk]

46R is not a short, it's a 109mA load which will dissipate over 0.5W of power. If a cap is dissipating that much it should become very obviously hot in normal operation.

That said, you should be aware that a lot of circuitry with multiple power rails have "discharge FETs" to ensure power-down sequencing, which will show the same "shorted when off, normal when on" behaviour.

[Renate]

Well aren't chips on the end being fed by 5V from one of the Molexes?
Is the one that measures 46Ω nearest the chips?

[perieanuo]

i doubt it's the caps, need more digging
from pdf i see only one led for each voltage (12 and 5V, page 2-6 D53-D54), that implies only those 2 power rails (meaning molexes are parallel) but you say it's 4 power rails, 2x12V and 2x5V
in that case, you need to diagnose the 2'nd set of 5V, maybe you got some 3v3 regulator failed or something derived from that 2'nd 5v rail who is shorted
viewing the quality of the caps in such products, i doubt you will find died caps, but maybe failed semiconductors, like some dc-dc or why not, sas controller chip
you have one working side, so identifying seems simple enough, just tracing the 5V loads on each 5v rail
but my bet is one 5V rail, and you have to look on 5V derivates/2'nd sas controller