g4 MDD warped CPU card

[65a]

So I have a dual 1.42ghz MDD. When I got it, it kernel panicked. Then stopped booting at all. I did the whole take-apart-and-rebuid dance, cleaned the motherboard and CPU card and it was fine.

I put a flashed ATI 9200 into it. Still fine, great even.

I swapped the 9200 flash chip to a full-size one, and it didn't boot. I spent a long time thinking it was the video card, until I realized it could boot headless and I could use capslock LED to determine if it crashed. It did indeed crash.

It basically exhibits random MDD symptoms, including J20 issue (didn't before) and powering off randomly. Thinking it was the PSU, I took it apart, and it looks clean (though not completely eliminated, the seller had said it was either new or recapped, and it looks that way).

The CPU card is warped. I am thinking this is the source of this MDD being haunted. I also imagine, but cannot prove, that the J20 RAM pins on the CPU card are the first ones to pull up with a warped CPU card.



Here's my plan of attack, but other ideas are welcome:
1) Clean CPU with full isopropanol immersion, some ShinEtsu microsi got messy. I don't think it's conductive, but I want to remove any chance conductive junk like corrosion bits is amongst the balls under the cache chips and CPU.
2) Try 1-3mm spacers to raise the CPU card edge clip so the bend is not pulling the connector out
3) If this fails, I might be worried the balls of the CPU have lifted off the board (nooo) or are intermittent. How reasonable is it to reheat these (e.g. a 858 style heatgun) without potentially causing the chip to come off or shorts? I know Motorola used weird non-full-melt solder balls here.
4) Buy a new CPU/MDD?

 

[65a]

I am slowly also repinning an ATX extension cable so I can completely rule out the PSU, since as we know that is not a trustworthy part regardless of appearance here. I would like a cleaner case for this guy, so if the end result is "keep obtaining MDDs and frankenstein them together until it works" I am willing to walk that path again, but it is interesting to try to debug these cursed machines.

 

[Durosity]

3) If this fails, I might be worried the balls of the CPU have lifted off the board (nooo) or are intermittent. How reasonable is it to reheat these (e.g. a 858 style heatgun) without potentially causing the chip to come off or shorts? I know Motorola used weird non-full-melt solder balls here.

That was my first thought... just at a glance it looks a bit off, particularly on the right side. @croissantking you have more experience with this than I do.. any thoughts?

 

[herd]

Can you test a different CPU to narrow the search for a problem?

It looks like something is on top of the CPU die in your picture.

 

[65a]

Can you test a different CPU to narrow the search for a problem?

No spare CPUs, getting more MDDs or CPU cards and frankensteining them is definitely a viable strategy. I do have a good collection of RAM sticks I've been going through, and a storage/datalogging multimeter I should probably hook up to the PSU rails.

It looks like something is on top of the CPU die in your picture.

I was experimenting with putting a solid cooling pad on top (since I was constantly adjust heatsink pressure and adding/removing the CPU). This didn't really change anything, also doesn't completely fit (have to loosely tighten the heatsink). Might be interesting anyway with a thinner pad just for not having to repaste in the future.

particularly on the right side.

Looking at it from different angles, it seems like if there's an issue, it will be in the middle. The problem definitely gets worse/different/better depending on how the heatsink is attached.

 

[herd]

I would not use pads. If changing the heatsink pressure or position is having an effect, then that does point to the CPU or CPU board. With OS X you can disable one CPU with a boot arg.