Howdy, I've had this one for a while and it's been a learning experience for sure. I might have some details wrong since it's been an off-and-on project over the last year but I've gathered this from memory as best as I can remember.

How it started
It's a 9700 Pro All-in-Wonder that I got in an eBay lot and never quite got it running at 100%. It got great framerates in-game but definitely has a memory issue so there were stray polygons with infinite length everywhere, and minor artifacting in BIOS fonts and Windows GUI. I wasn't sure what to do with it, especially after an experience reflowing a 9600XT AIW that had the same issue and I only ever managed a temporary fix.

Where it went
Seeing some of the threads on here about BGA memory work and identifying the bad chip with R3MEMID, I got a hot air station and found the suspect chip. The test results gave me MDA0 bits 52 -53, so according to the diagram that's chip U53, on the AIW Pro that's back of the board and lower left. I got the chip fluxed and hot until it wobbled a bit, but might have pressed too hard. Powered on, and SPARKS - blew a MOSFET.

Removed the dead MOSFET as well as the DDR1 chip, and powered on again after some cleanup proceeded to eBay for some new MOSFETs, got a pack of 10 of these FDS6898A. After soldering one in (as best as I could manage), I got the almost solid color textmode that I've gathered seems to indicate a voltage issue going to RAM, but not every character is completely solid. So at least the board wasn't dead and I didn't fry my ITX Pentium 4 rig, and this is probably the expected behavior with no RAM chip installed. After an attempt to reball the existing chip with probably the wrong jig, I resigned to eBay again to find an appropriate replacement. Running R3MEMID headless with a batch file just logged that every bit was bad.

How it's going
After three attempts to buy RAM ended in lost packages refunded, missing inventory refunded, and possibly one that never shipped so I just filed a claim, I got four of these FBGA-144 K4D263238G-GC2A. They aren't 1:1 replacements for the same timing but seemed close enough. It's actually been months since I started this project at this point and the chips sit in my garage for about another month until I get around to installing one, the hot Florida summer hasn't helped either. Some flux on the clean pads, tape off the surroundings, and heat the chip gently and gradually until a satisfying pop into place. Boot up, and no change.

This is where I read the forum threads again and see the voltage to RAM issue being a cause for similar output, so I re-check my MOSFET install job thinking that capacitor I blew over and reseated might not be connected. Some multimeter poking around and tracing a nudged pad to another cap, and I reflow the connections on the MOSFET. Hey, that pin has continuity to a nearby capacitor now. Boot up, and we have normal output with minor artifacting like where we started. Load up Morrowind and we still have solid polygons going every direction, maybe not the framerates it had before though, I remember it averaging above 70 on that same save location but now it's 30-45 at 800x600, not what I expect from a 9700 Pro but those infinite polygons might be intense. Let's run R3MEMID again. MDA0 bits 52-53, where we started. The modified NWVMT is giving a different result here, but I might be running it wrong - bus=128 gives me chips 3 and 7 bad, but bus=256 says every odd number chip is bad. That's a really different result from two bits in R3MEMID.

Back where we started isn't bad since it's at least functional, just dealing with the artifacting again while the card seems otherwise stable with what seem to be appropriate replacement components. So now I'd rather get a few sets of eyes on this before I go replacing the same chip again or poking around with my multimeter while the card is powered. I have 3 new chips left but I'm wondering if it could be a voltage or trace issue instead.

• This is an AIW card, so I'm curious if the chip diagram listing might be different
• I don't really know where to begin with considering voltage issues, does the card need to be AGP slotted for testing that or can I get by with the berg connector alone? Any schematics on this particular card?
• For what it's worth, I haven't removed the fan and heatsink yet. No clue what the shim and thermal compound situation is under there

If I can get this one saved, the 9600XT AIW is next. That one started with artifacting but now only blocky BIOS text like there's no voltage to the RAM, but I haven't touched any of the MOSFETs on it. Much appreciation for any guidance I can get on next steps here.