@thedrew has tried all manner of exotic methods of removing the GPU without stressing it thermally. That includes grinding away the motherboard from behind, both mechanically and with a laser cutter.
Also, both Drew and
@Aurelio have cut out the mobo around the GPU to reduce the thermal mass and the amount of heat required to desolder it, to no avail.
@Redherring32 has a proper BGA rework station, so I sent him a Vegas board. He baked a Wii chipset to remove all the moisture, and was able to reball+attach the chipset without dead shorts. My design needed some bodges but Red was more than capable of doing those. Same result: chips got warm, but no signs of life.
So no one has been successful yet, and it's unclear why. It could be that the
GPU die attach solder balls simply have a melting point too low for rework and they're shorting under the die. Or it could still be a moisture thing and no one has baked the chipset long/hot enough to prevent cracking/popcorning. Or it could be signal integrity/stackup stuff, though I think that's unlikely.
Working on my own redesign isn't a priority for me right now, but IMO here's what needs to happen for any of the redesigns to move forward:
- Full characterization of a softmodded, stripped down stock mobo or omega trim:
- current traces of all voltage rails at boot, measured with sense resistors
- logic analyzer captures of the NAND lines, and digital video lines at boot
- logic analyzer capture of boot0/boot1 debug port at boot
- Rework of redesigned mobo to include:
- current sense resistors on all voltage rails
- test pads for NAND lines, digital video lines, maybe EXI, and boot0/boot1 debug
- Proper temp-controlled reball and rework of the GPU onto new PCB (I think RedHerring's attempt is the best/most controlled yet)
- Full characterization of the redesigned mobo, just like with the original mobo:
- current traces of all voltage rails at boot, measured with sense resistors
- logic analyzer captures of the NAND lines and digital video lines at boot
- logic analyzer capture of boot0/boot1 debug pads at boot
This test plan would allow for a 1-to-1 comparison between a stock mobo and the redesigned one, and should make it pretty obvious if there are hardware issues, if boot0/boot1 are succeeding or not, etc. Hopefully someone has the bandwidth to tackle this