Bob F wrote:
My ECS Z77H2-A3 V1.2 motherboard with an I5-2500K processor worked fine
for years, then started having occasional problems when turned on, it
would not post occasionally. The problem got more frequent over months,
until it finally is next to impossible to get it to post.

When it does post, it seems to work fine, but it might take 1/2 hour of
turning the power supply on, pushing the power switch, watching it spin
the fan for 2 seconds, stopping, then powering the fan for a second,
stopping, then powering the fan for a second...... When it does
actually boot, it seems to be on the second start of the cycle.

This morning, I got it to boot, and ran it with the OCCT stress test for
40 minutes with no errors.

I then tried it with a different power supply which works fine on
another board. On the problem board, pushing the start button did
nothing. I tried jumping the green wire on the power connector to the
neighboring black wire, and that did nothing also. Unplugging the power
supply and using the jumper started it fine.

It seems to me that this is some kind of problem with the start
circuitry of the motherboard, and does not seem to affect the
processor's operation at all if it can start.

Can anyone suggest anything that might be a repair option for this kind
of problem, or point me to any references that might educate me on the
related circuitry?

1) Mostly a waste of time, but a PCI port 80 card or similar,
could show you some "progress codes". But if the Port 80
card only shows "FF" or "00", then the CPU isn't getting
a chance to execute any code. If you happen to own a PCI port 80
card, then use it. But don't waste your time and money tracking
one down. A local computer store might charge $100, for an
item worth $10 to $15 or so.

2) At a guess, Power_Good (from the PSU) is deasserting,
just after the RESET pulse finishes. And this will eventually
result in another RESET pulse being generated (looping behavior).

When the system comes up, measure the current flows on the main
ATX power cable with a clamp-on DC ammeter. What you're looking
for, is loading which is out of normal. With the "modern" power
supplies with the all-black wiring and sleeving that prevents
access to individual conductors, this is "difficult to impossible".
Thanks to "progress", we're denied a valuable debug input.

A "proxy" for overloads, is to measure the 3.3V, 5V, 12V voltages.
As those have the main DC loads. You might also check the -12V and
the +5VSB for the sake of completeness.

When using voltages as a proxy, we're looking for crossloading
effects. The short and sweet is, if one of the rails is
"abnormally low", suspect it's the one drawing too much current.

For example, the supply might have +/-5% tolerances. If one rail
is 1% below nominal, and the other rails are higher by 1%, that
hints that the below-normal one is up to something.

Each generation of computer, has a different loading pattern
for rails. The best time to establish norms, is to measure
the rails when the system is working. Or, if you have two
different manufacturers of systems in the same generation
of that hardware, compare the readings to see what typical
practice is.

My old Nforce2 board, a board which doesn't use +12V for the
processor, the +5V rail has a loading of 25 amps while gaming.
This would "tip over" almost all the supplies in the
current Newegg catalog.

Modern systems don't do that. You might expect 3.3V @ 6A or
5V @ 6A for deriving chipset voltages. These are moderate
loads compared to the PSU 3.3V @ 20A and 5V @ 20A capability.
Each hard drive, draws 1A from 5V. The optical drive can
draw a bit more than that.

The Power_Good signal from the PSU, is not the end of the
story in terms of the Power_Good concept. The purpose of
Power_Good, is to assure that the voltages are stable,
after the 35 millisecond capacitive load charge-up phase.
Hard drives, for example, won't start to spin until the
12V is a hair over 11V and rising. The hard drive, in a
sense, has its own private copy of Power_Good, which guides
its operation and loading characteristic. It doesn't
get to look at the ATX Power_Good for advice.

Some motherboard designs, use a Wired-OR of the PSU Power_Good,
plus a "good" indicator coming from various switching converters
on the motherboard.

As an example of an anomaly encountered years ago, two
brands of products would not "play nice" with one another.
An Antec supply, was "slow" raising its rail voltages. The
capacitive charging phase seemed to be slower than normal.
Then, you would combine that with a motherboard which had
OCP on VCore. The VCore converter would start converting
the 12V, before it was fully up. As the voltage drops on
input to a switcher, the current rises. The switcher
in a sense, is a "constant power" device. If the voltage
into the switcher drops by half, the current flow doubles.
Anyway, the VCore, upon starting its job based on some
timer, it sees maybe 6V on the 12V rail, thinks there is
an overload, and it "latches off". The customer motherboard
will not start. By switching off the power supply, the
latched status gets cleared, but the same thing will happen
on a second attempt. The solution is to mix different brands
of power supplies and motherboards, to avoid the "problem pair".

Your system, by looping like it is, is telling us that
no latching behavior coming from the motherboard is involved.
As that would stop the loop in a "failed state".

Yours keeps trying. That suggests, mainly, a problem
between the supply and the motherboard. Motherboard
is treating power supply "poorly". Bitch-slapping it.

This suggests some sort of issue with Power_Good, and
about all I can suggest is something is amiss with loading
which is ****ing off the power supply. Maybe it's the
"transient" of the VCore starting up with the full 12V
present.

There is a limit to how much capacitance can be on the
DC rails powered by your ATX supply. It's about 5000uF
per rail. The PSU has internal caps that account for
a fraction of that. The motherboard has bulk capacitors
for filtering too. The sum total on a particular rail,
should not exceed 5000uF to maintain sufficient phase
margin on the ATX control loop. While it would be
tempting to (as a hobbyist) suggest "adding more
capacitance", you have to be aware that these designs
only have 45 degrees of phase margin, up to a certain level.
And that spec (knowing how power supplies behave),
is hard to get. Even for an engineer, if I phone up
some fly-by-night operation in China, I would have a
hard time getting a definitive statement on that. The
above is simply a suggestion of a typical scenario.

If you had a pulse-catcher, you'd arm it right after
pushing the front panel power button, then see if it
catches a low-going transition on Power_Good. That would
be an indication that a "slug-of-current" was being drawn
by the motherboard. Such analysis is easy to do...
if we're back in the lab with our $35K storage scope :-)

My clamp-on ammeter can record the "peak" current in an
interval of time, and my voltmeter has both peak voltage
and lowest voltage detection. But these capabilities are
related to the sampling rate of the devices, and the
event we're looking for, is too short to catch
with these capabilities in measurement devices.
Even an HP meter with a 1KHz sampling rate and
peak detection, might miss what we're trying to see.

A storage scope wouldn't miss it.

*******

A second question for you might be, have you recently
changed the video card ? Try putting the old one back
as a test. Or even, pull the video card, and see if
the system enters the "missing video beep state"
properly. This would then indicate that the system
has survived the power up transient and the BIOS
code is getting to run. If the looping behavior continues,
with the (high power or new-ish) video card removed, then
you're no closer to an answer (it's not the video card
or the BIOS code that loads the video config space ROM).

Paul