[mythtv-users] HDHomeRun Prime failure

[f-myth-users at media.mit.edu]

Some UPS's will trip GFCI's all on their own, -especially- in a power
failure situation, which might have happened to you unnoticed if you
lost power for a moment at 2am and your UPS went to battery and back.
[Yes, I know you say it -also- tripped the GFCI when it switched to
line power again; this may be a similar issue.]

For example, I had an old APC BackUPS 650 (stepped sine wave, not true
sine wave) unit that ran with about a 100-200W typical load.  Unplugging
it from the wall supported the load.  Plugging it back in did not trip
the GFCI.  However, a -power failure- is different---in that case, the
UPS's power input is not open-circuited from being unplugged, but rather
is short-circuited through the entire rest of the grid.  And in -that-
situation, the UPS would reliably trip the GFCI (in this case, a circuit
breaker in a panel, and not one of the outlet-type GFCI's).  I presume
it backfed a few tens of milliamps through only the hot or only the
neutral, probably through some large resistor, and in a real power
failure, who cares, it's supplying a few mA at 120VAC to a dead short
(the rest of the grid).  But the GFCI saw that as an unbalance between
hot and neutral and did its job.

This mean that any power failure would tend to be survivable by other
UPS's but -not- the UPS on that GFCI, because it would trip the breaker
(unnoticed, of course) on its way down, and then when power came back,
the GFCI was still tripped, and the UPS would run down and die.

I finally had to put the UPS on a non-GFCI-protected branch.

P.S.  I agree with others' assessment that it's likely your power supply
was just marginal from caps slowly going bad, and it can't restart due
to ripple or inrush current.  So it's not a surge---it's that it just
hasn't had to start up from zero.  I've had numerous power supplies and
even motherboards with similar failure modes, all of which are recoverable
by replacing the bad caps, but in your case I'd recomend just getting a
new one from SiliconDust since it's a known failure point.