i had some 9700/9800 cards where gpu was disattached from pcb, cards were thrown or it is an aftermatch of cooler removal ( yellow glue attached).
had to reflow a gpu to pcb. sometimes it was also accompanied with memory ic reflow/reball.

This post made me gain courage to try and tackle my dead 9550 and my 9600 which I've used as parts donor for the 9550.
9550 had an missing cap on the core which I've needed to resolder, although it looks like one end of it is disconnected (I made sure to reflow that side multiple times). Sadly it's artifacting badly ☹️ Now armed with R3MEMID and VMT I can diagnose which chips are bad. I'm pretty sure it'll say all are bad 😅...

You'd be surprised. One memory IC or missing series resistor to one can make the card look broken as hell.

tehsiggi wrote on 2025-09-01, 16:07:

You'd be surprised. One memory IC or missing series resistor to one can make the card look broken as hell.

Huh, well that makes this repair far more complicated as I had already replaced the resistor arrays (those darn things don't wanna solder on!). At least I have a pad repair kit which actually works (unlike the relife ones).

kotel wrote on 2025-09-01, 16:23:

tehsiggi wrote on 2025-09-01, 16:07:

You'd be surprised. One memory IC or missing series resistor to one can make the card look broken as hell.

Huh, well that makes this repair far more complicated as I had already replaced the resistor arrays (those darn things don't wanna solder on!). At least I have a pad repair kit which actually works (unlike the relife ones).

Go with running R3MEMID. You may have to run it "blind" if the screen is too corrupted. So make sure to use the -LOG flag, it'll create a R3MEMID.LOG with the results.

Just make sure to exit the application properly to get the log file written. If you can, try running it with a known good card, so you "recognize" the menu items 😁

Welp, I just remembered what piece of work the small TSOP-48 pads were to replace. I don't think those cards will ever work again...

Ripped pads are a tough one, I usually run away from those too. Fingers crossed!

If I may ask, what was that AGP data lines tester tool you've used in the first post? How does it work? (if you know this)

is it ok if i join you -)

found a 9250 pci on scrapyard. diagnostic show no generation on oscillator.
so i have to remove gpu to see this

after fixing the pads...

Card is alive

dm- wrote on 2025-09-04, 03:14:

is it ok if i join you -)

Please join! This is not "my thread" - it's a collection, to whoever wants to share their results! Thanks for showing it here!

kotel wrote on 2025-09-03, 16:19:

If I may ask, what was that AGP data lines tester tool you've used in the first post? How does it work? (if you know this)

It's one of those combined PCIe & AGP testers from aliexpress.

Basically it has one AGP slot and two PCIe slots (both in parallel, not sure why they included two of them).

They leverage the fact that GPUs have ESD protection diodes on their I/O pins. Those diodes are there to redirect over-voltage on the inputs to the power rails, meaning positive spikes to something like VDDC and negative ones to GND. Basically anything that is around 0.7V > than the supply voltage of that input section of 0.7 < GND will be flowing through those protection diodes. They are however not very current capable. So they don't protect the GPU from a user forcing a lot of power into an I/O pin.

However the tester uses these protection diodes in a funny way. They *short* both the cards power rails and GND together, so that there will be no current flowing into the power circuitry of the card. Then they connect that combined rail to a supply voltage around 5V.

They then connect an LED + resistor onto each I/O pin of the AGP slot. Due to the diodes in the GPU, there will be a current flow through the GND protection diode (D2 in my drawing) of the GPU.

The drawing is *very* simplified, but basically shows the idea. If an LED is much brighter than the other ones, the current does *not* flow through the GPU but there is a bridge somewhere to a power rail. If an LED is not lighting up, the connection to the GPU is missing.

For PCIe they're doing a bit more fancy stuff, they use some logic ICs + Pulsed voltage, since the data pair lines of PCIe have capacitors in series, which only allow for AC current to go through.
All in all a handy little device for a first check, if anybody is still home.