I figured out that the problem was with the PWR-OK (or PWR-GOOD) line.
When I used a more standard ATX power supply, the system worked exactly like it was supposed to. I noticed that the PWR-OK line would have a nice square falling edge when AC power was removed. The other power lines (3.3V, 5V, +/-12, etc) would slowly drop out. My theory is that when the PWR-OK line goes low, the motherboard detects that power will be lost and immediatly saves some of the important settings. (such as the turn on after power failure setting). and since the power lines drop much slower than the nice square PWR-OK line, there was plenty of time to save some settings.
The power supply that needs to be used in the final application is a custom design, so it can fit in a very specific mounting location. I noticed that the PWR-OK line on the custom power supply would slowly drop at the exact same rate as the other power lines. I made some changes to the custom power supply so that the PWR-OK line has a square falling edge, and I made sure that this edge is driven low with enough time for the motherboard to save setting before the other power lines completly drop out.
This seemed to fix the problem I was having.
Hi. Thanks for your explanation. I've a similar issue with d510mo and a mini-itx case. It has a custom atx board and it uses an external 12 volts power supply. When power loss occurs when system is off (power cord on) then it restarts but if power loss occurs when system is on then no restart. I swapped the PSU with a standard one and it works fine. I noticed that when power loss occurs when system is on the light on the motherboard dim quickly as when system is off. I hope I can explain it...
Anyway, how did you fix your problem? Which modification you did?
To replay remove spam.sux
For anyone who is mystified by "made some changes to the custom power supply". There is another method of making this motherboard power-on automatically. I have developed + tested this on a D510MO with a PicoPSU. No PSU mods required. The idea is to simulate the holding of the power-on switch for a second or two after standby power hits the system. This can be very simply done by connecting a 1000uF capacitor to the PWR_ON header pins on the motherboard. Be sure to get the polarity right, or use a non-polarized cap! The only side-effect is that you must keep the motherboard powered off for at least 10 seconds after pulling power, before re-applying it if you expect it to power on again. This gives time for the 1000uF capacitor to discharge.
With this motherboard, you should use a capacitor that can take at least 3V. Open circuit looks to be 2.9V, and short looks to be 1.1mA.
This little hack should work on many other boards, probably with different capacitor values. Just calculate a decent RC time constant from open/short measurements of PWR_ON header. That being said, I've never had to fix such a problem!
Shame on PicoPSU for releasing an inferior product that doesn't slope PWR_OK signal as required.
And Intel: Can't you change board design so it's not so sensitive to power supply shutdown sequence? Powering on when you see power applied isn't that hard of a problem to solve in a more reliable way. The tiny computing world thanks you.
All you need is a pull-down resistor on the power good line. That's the modification we made to our custom power supply get it to work, and probably the modification the poster above did. The ATX spec calls for the power good signal to drop first even on a hard power failure. You can't blame Intel for issues using a power supply that doesn't meet the spec.