Dwight K. Elvey
dwight at ca2h0430.amd.com
Wed Nov 9 18:21:11 CST 2005
>From: "Sebastian Brückner" <sb at thebackend.de>
>Sridhar Ayengar schrieb:
>> Sebastian Brückner wrote:
>>> I know that those machines are supposed to run continuously but it
>>> can't be quite normal that every other time I power it up one of the
>>> supplies dies. Any hints? Is there a way to fix those things or to
>>> prevent them from blowing up altogether?
>> Are you sure that you've wired your premises electrics correctly? That
>> could cause power supplies to keep blowing.
>Well... that could be the cause. I'm running the machine off a 16A fuse
>although it is designed for 25A IIRC.
>Of course the fuse could blow on power up but I didn't expect that to do
>any damage to the power supplies... maybe that was a very bad idea(tm)?
>> Do you have enough power supplies for the boards and other options you
>> have installed in the system? That could be doing it too. Or if you
>> had a boardset with an unfused short in it or something, although I
>> would think there's probably fuses in the backplane or something.
>After losing supplies with the original configuration (24 CPUs, ~18 SCSI
>channels, 1.5GB RAM) I stripped the machine to a fairly minimal
>configuration (something like 4 cpus, 512mb ram, no extra scsi). The
>problem didn't go away.
>Once it boots (and doesn't blow as soon as I turn it on) it runs without
>problems. And since it worked in that exact configuration for years for
>the previous owner (continuously powered on though) I expect it to be
>fully working. He also said that he often had to replace the power
>supplies after power cuts, so it might just be normal...
One thought. If it uses transformers anywhere, connected
to the AC ( even switchers often do this for initial voltages ),
you need to put a MOV on the input leads. The problem is
( and I've actually seen this ) that when you cut the AC,
the core holds some energy. Since it can't pass this to
the input leads ( you know, that 90 degree phase thing ),
it causes the voltage to spike on the secondaries. Many
negative regulators could not handle the spike.
Second thought. If it is using switchers, make sure you
are providing the right AC voltage. Many of the older
switchers would actually run on 120V when setup for 220V,
for a short time. They would eventually blow in a minute
or so. The problem is that with the lower voltage, the
switcher would be running too long a duty cycle. This
would smoke the transistor because there would either
be some overlap or the cores would saturate. In either
case, it would blow the supply.
More information about the cctech