Misleading power-supply test result

  • Thread starter Thread starter Percival P. Cassidy
  • Start date Start date
P

Percival P. Cassidy

I had an Antec Neo-HE 550 p/s in a machine that died a couple years ago
after a severe thunderstorm. Used my CoolMax power supply tester, which
showed that the 3.3v line was bad. This tester shows the actual voltage
only for rails that are approximately OK; in other cases it just shows a
flashing "LL" (or perhaps alternatively "HH", but I haven't seen that).

I bought another p/s and put the old one on the shelf. (I had already
planned to upgrade that machine anyway, and a couple of the cards had
burn marks, so I just replaced the motherboard as well, so didn't check
whether the old motherboard worked with the new p/s.) Two years went by,
and I decided to investigate the old p/s further. After opening it up
and deciding that doing any detailed troubleshooting was beyond my
abilities, I checked the 3.3v line with a meter and found that it was
2.998V -- less than 10% low -- so I decided to try it with a new
machine, thinking that if necessary I could boost the DRAM voltage to
make up for the low 3.3V rail. All seemed to be OK without boosting the
DRAM voltage, and I checked the 3.3V rail -- and to my surprise found
that it was now the full 3.3V.

IOW, although the other supply rails were OK with a minuscule load (the
tester), the 3.3V rail was OK only with a significant load.

Perce
 
Percival said:
I had an Antec Neo-HE 550 p/s in a machine that died a couple years ago
after a severe thunderstorm. Used my CoolMax power supply tester, which
showed that the 3.3v line was bad. This tester shows the actual voltage
only for rails that are approximately OK; in other cases it just shows a
flashing "LL" (or perhaps alternatively "HH", but I haven't seen that).

I bought another p/s and put the old one on the shelf. (I had already
planned to upgrade that machine anyway, and a couple of the cards had
burn marks, so I just replaced the motherboard as well, so didn't check
whether the old motherboard worked with the new p/s.) Two years went by,
and I decided to investigate the old p/s further. After opening it up
and deciding that doing any detailed troubleshooting was beyond my
abilities, I checked the 3.3v line with a meter and found that it was
2.998V -- less than 10% low -- so I decided to try it with a new
machine, thinking that if necessary I could boost the DRAM voltage to
make up for the low 3.3V rail. All seemed to be OK without boosting the
DRAM voltage, and I checked the 3.3V rail -- and to my surprise found
that it was now the full 3.3V.

IOW, although the other supply rails were OK with a minuscule load (the
tester), the 3.3V rail was OK only with a significant load.

Perce
Click to expand...

I can't say, whether it's the situation in your case, but some supplies
in the past, had a "minimum" loading spec. The label of such supplies had
two rows of current flow numbers. "Max" numbers established the upper limits.
But they also had a minimum, and it might be one of the rails that had
the minimum value mentioned for it. Most rails might have a minimum of "zero",
while the odd rail might have a non-zero value. Typically, the minimum
current might be in the 1 to 2 amps range.

The absolute worst ATX supply design I've seen, had a minimum current flow
on one rail, of 25% of the max current flow value. Which is a really
really bad design. The supply made up for it, in that it had an
extraordinarily high max current flow value. So high, you'd think
it would burn the wiring harness, before indicating an overcurrent
situation.

Some supplies solved this problem in the past (crudely), by placing a
load resistor across the affected rail, so that no matter what, the
regulation would fall into line. But when a supply is making claims of
being in the 82-85% efficiency range, it's pretty hard to slap in
a solution like that. They could, make the solution dynamic, and
switch in the resistor under the appropriate circumstances, but then
you'd need some kind of circuit to make the decision when to do that.
And that would affect the measured efficiency, at extremely low loads.

Paul
 
Back
Top