Vega 64 GPU burnt PCB fix next steps

[Jhongy]

Hi,

I'm wondering what my next steps should be with this board. I'm tempted to power up and test at this stage but would appreciate some advice. Here's what I've done so far:

I had a water cooling leak and a burned Vega 64. In hindsight the burning smell was around for some time prior to the system shutting down, so the leak may (or may not) be incidental. The leak spread across to the PCIe connectors, so anything could have triggered the PSU short circuit protection.

On examination, a number of MLCC caps were burned right through, burning through a couple of layers of FR4 too. The caps appear to be just filter caps underneath one of the power phase mosfets pair. I've circled the area (in red) on a picture of a good board. Two of the caps were completely vapourised, two were in bits and two remained -- one with a dead short.

I removed all of them, and removed as much of the burnt detritus as possible.

There were 6 caps, 3 pairs of two (2x 1206 10uF, 2x 0805 9uF (!), 2x 0603 2uF), all appear to be between 12v and ground. I've removed and measured the caps from an adjacent phase (circled in green) and ordered replacements with similar values (albeit replacing the 9uF caps with 10uF).

(shows the back of the PCB with areas circled)

After scrubbing away the burned parts, the board damage was quite severe. The layer below appears to be a ground plane, which was shorted (<10 milliOhms)  to the 12v traces to the caps above. I carefully dug away with a scalpel and peeled back and separated the planes and ground back burned FR4. Eventually the short cleared and I'm now showing around 8 ohms resistance.

I've filled the ground area with epoxy and plan to try to sand flat, rebuild the top tracks with some copper tape and some uv solder resist, and replace the caps (I'll try with the 0603 parts but may give up and use bigger parts).

(shows the gaping wound, covered with an initial layer of epoxy)

However, I'm concerned that 8 ohms across 12V and ground still seems rather low. Overall the resistance at the ouput of each of the power phases seems OK -- about 10 ohms. These phases seem to be connected solely to the first PCIE power connector, and the 8 ohm resistance between 12V and ground is showing there. The second connector is showing ~100 kOhms which seems a bit more reasonable. It's the differnce between the two that has me worried.

Powering the card with a limited voltage through the pcie connector shows it drawing a current ou'd expect at this resistance, with nothing obviously getting hot, but I;m wary of cranking it up to 12v and letting it draw the full 1.5-ish amps.

Taking 4-wire milliohm measurements at each power phase, I don't see any significant differences between them, so it doesn't look like the phase in question has a further short, but it's hard to be certain without removing components. I'd like to avoid removing the DirectFET packages unless I have to! On the flip side, it would be much easier to replace any directFETs while the back side is still unpopulated.

The opposite side of the board from the burn looks OK, but shows signs of blistering from the heat. If there were sill a fault, my next steps would likely be to suspect the capacitors and/or mosfets here.

(shows the front of the PCB, with blistering)

Some questions:
- I think my next step should be to power up the card in place in the pcie to see if it looks OK, but:
--- is this wise with an 8 ohm resistance between 12v and ground on one of the pcie connectors?
--- is it OK to briefly run the MOSFETs without their associated capacitors?

- Any chance that the exact values of these caps are important (i.e. do they control the power phase behaviour) -- i.e. is replacing (consistently measured for both) the 9uF caps with 10uF going to be a problem?

- Also, any tips for rebuilding the damaged PCB? Am I headed in the right direction, or likely wasting my time?

Thanks very much in advance for any advice.