From my research it looks like 3 blinks is CPU error (hence why the board only does that after some time without P4 connector) but the 4 blinks? I have no idea. Maybe somebody here has an idea what could 4 blinks mean?

The sound from the VRM has changed to a high pitched short whine (board still throws the 3 blinks then 4 blinks). I guess I can only replace the 7 1800uF 6.3v caps in the VRM now...

Replaced the 1800uF caps with some fake nichicons (same batch works fine on an gx270) and one 560uF poly near AGP but to no avail. Board still "squeals" from either the VRM or the amplifier coil (?) near the LAN port or the LAN chip. I knew replacing the 560uF poly wouldn't make a difference, but why not try it if this board is already at a loss. No idea what the issue is.

is lan working? you can try lifting lan chip VCC

No idea. LAN chip is BGA, so only way would be to desolder the VRM module that generates the proper voltage for it. But on the other hand the sound only appears when 12VP4 is plugged in and I have narrowed the sound down to ISL6556 PWM controller. No idea why I heard it coming from the amplifier coil or LAN chip.

The MOSFET's on the back (in the middle row) are getting warm when the board whines, which would meant a short. Whats weird is that all of them have the same readings as the ones that don't heat up. All caps have 3 ohms in circuit. My idea was to replace the ISL6556 PWM chip since the noise was coming from it, but now I am quite baffled if thats the case.

Might be faulty driver, depends on how cheap can you get it. $5 on alibaba sounds ok, but paying $10-20 for a maybe chip sounds excessive. I would whip out a scope first to check how it drives mosfets and if it really opens both up and down pairs at the same time.

Since I fixed the original PSU I have gotten the idea to try it on this board.
Sadly with the p4 connector the board was doing some sort of power up/down thing. So I've done the only logical thing - plug the p4 connector while the board is running. (very logical)

Turns out this wasn't a good idea and now the whole VRM is shorted to GND 🤦
I think it'll go away after the MOSFET's and whatnot takes a break but not sure.