I put one of those ATX power supply testers on a Toughpower 850 W, and the
LEDs go out after a few seconds. I happended to have a Toughpower 750 W,
and tested it too - same thing happens. Other brands of power supply
have no problem with the ATX tester (which is just a bunch of LEDs and
resistors I presume).
So why would Thermaltake object to no load (or a few mA)?
BTW the 750 W power supply was definitely okay, taken from a working
system with monster graphics card.
Monster power supplies can have a "minimum load" spec.
In this case, I had to dig a bit, to find an article
quoting a more complete spec for a Toughpower. I'm willing
to bet in many cases, you just wouldn't be able to find
a spec like this.
http://www.bjorn3d.com/2008/10/thermaltake-toughpower-850-watt-esa-power-supply/
Summarizing, for minimum load:
3.3V @ 0.5A
5.0V @ 0.5A
12V @ 4.0A (they spread the load over all outputs, but the supply is unified inside)
The +5VSB has a minimum load of 0 amps, as you'd expect. As it comes
from a separate supply and is there for soft off state. I don't know
if the industry thinks efficiency is important for soft off, and without
some spec like EUP, perhaps they don't care.
In the past, "honest" supplies used to have a line across the label,
with the "min load" specified. Then, supplies started only listing
"max load". Back in those days, at least a few supplies, used an
internal resistor to ensure load was being met, so they could have
a true "zero load" rating minimum.
But now that we're in the "87% efficiency" era, using internal
loads would spoil the efficiency.
The real question would be, why are the LEDs going out ?
One of the consequences of lacking a minimum load, was supposed
to be that the unit would not regulate to 5%. So some of the
measured voltages might be wrong. (Big deal...) Having the supply
turn off though, suggests that perhaps OVP engaged. But that would
require a pretty gross error (like maybe 7V on the 5V rail). I find
it hard to believe that would be happening. I would expect
a smaller regulation error than that.
*******
Note that, with Haswell coming this summer, some supply manufacturers
have been rating ATX supplies as "Haswell ready". And this alludes
to your very problem - some Haswell systems will have extremely low
loading, to the point that high end supplies, are not going to
have their minimum loading spec met. So some of the supplies
are being tested (and marketed) as again, truly being rated
for "zero load". Just exactly how they'll be doing this, who
knows. Perhaps "intelligent loads", that only get switched on
at low load ? It costs money, to slap MOSFETs in there to
disconnect a load when not needed.
So, the laws of physics weren't repealed. At one time, the supplies
had both a "min load" and a "max load" amperes line on the label.
Then, the marketing people chose to bury the minimum load, some
companies used internal resistors and so on. Then the "efficiency
era" arrived, and 65% or 67% efficient supplies were no longer
good enough. And at 87% efficiency, it doesn't make sense to
slap "dumb" resistive loads on the thing.
http://www.bjorn3d.com/Material/revimages/psu/TT_TP_850_ESA/Rails.jpg
There is a way to solve this. ATX supplies are "push" architecture.
That means, if the load calls for more current, they push out more
current. Now, imagine for some reason, current tries to flow back
into the supply. There is no "pull" capability, to absorb the power.
Changing the architecture to "push-pull" would give true zero load
capability, but I'm willing to bet the efficiency would no longer
be 87%. And there goes that "80%+" rating they're so proud of.
The purpose of the "pull" capability, would be so that if the
voltage wanders above Vrail + 5%, the pull portion would pull it
down again. In other words, tight closed loop feedback, in both
directions. But since this isn't going to happen, there's really
no reason to dream about it.
When I want to measure a supply here, I have a home-made load box.
That provides enough load, to meet any minimum load requirement.
It's just a bunch of resistors at a buck or two a piece, that
I can buy at my one "good" electronics store. Radio Shack would
not have the selection for a project like this. I also incorporate
an 80mm fan, to blow over the set of resistors and dissipate the
heat. I don't attempt to draw the "rated" power, just put a
representative (say 100W or 150W) load on the thing. And I haven't
had any out-of-spec operation by doing so.
(Example of the style of resistors used in the load box...)
http://www.frys.com/product/472235
Suffice to say, a guess would be your Toughpower is not "Haswell ready"
If you have a means to apply an artificial load to the ToughPower,
such as plugging a couple of hard drives to it, perhaps that
will maintain enough of a load to keep it running.
*******
You can see the basic principle of how an ATX PSU works, here.
This is one of my favorite sites, because the author of this page,
painstakingly traced out the circuit by hand (by visual examination
of a power supply and its printed circuit board).
http://www.pavouk.org/hw/en_atxps.html
On the output, on the right, you can see a common transformer,
and individual rectifiers and filter caps on the output rails.
The rectifiers "pump up the output", filling the capacitors.
But there is nothing to "drain" the capacitors, if they are
overfilled. The circuit does have feedback, and the circuit
can raise the input to the transformer, when the load on the
output is sensed to be increasing. But once the load disappears,
and the primary side reduces its pumping as much as it can, it
still leaves a zone of instability around zero load.
You will notice that example supply, does have dummy loads on it!
But the 100 ohm resistor (R50 on the extreme right) on +5V,
hardly draws much power. So it's not the kind of dummy load
I had in mind. A 5 ohm resistor would have been a dummy load.
The 100 ohms probably would not be enough to solve the "pull"
problem. But that supply is a rather low output design, so maybe
that is all that design needs for stability. A 100 ohm load
isn't going to affect the efficiency, and that supply
is an older 65% design anyway.
Paul
