Allright, I'll try and get you up to speed as fast as possible in an anecdotal way.
A few years ago i could get my hands on two 2nd hand R9 290X reference models. They had been used for bitcoin mining, so they were cheap. I did not mind it, since i could buy two instead of one. The reference model is pretty loud, and since i had an interest for custom liquid cooling for a while, i decided to give it a go. I used some alphacool Nexxxos GPU blocks. They are fairly big and heavy. Did not go well, almost fried them since i made a minor mistake while mounting the blocks. After that, the 2nd GPU was basically dead. It would post to the bios, and occasionally show up on windows device manager. I was pretty sure i killed some sensitive connection on the board with heat. Meanwhile my motherboard had been in bad shape for a while as well. FX8350 at 4.9 GHz does that to a 10 phase motherboard as you might expect. I was too lazy to disassemble the whole loop, so i clocked the 8350 @ 4.3 GHz and disconnected the PCIe power connectors to the 2nd GPU. I got a 3rd GPU for even less money, from the same guy, but he wanted me to send the 2nd GPU back. Good trade, i took it. When i got the 3rd GPU, it behaved exactly like the 2nd GPU. So I kinda had a feeling i got screwed and he just sent me the same GPU back in another box. I could not prove it, and another explanation would be that i simply had destroyed my PCI lanes somehow. So i kept the thing in the loop with the PCIe power connector disconnected yet again. It ran like that for over a year. Fried motherboard and GPU both still in place. It was stable, and i had no money to replace either part anyways.
Now, with Ryzen coming out i decided it was time for an upgrade, especially since i got into streaming and youtube content creation. That gave me the opportunity to tinker with the dead GPU again, as i was still not sure if the motherboard was just fried, or it was actually the GPU that was broken. Turns out it was the GPU all this time. I am pretty sure the guy screwed me over and just send me my old GPU back. I put the GPU in the oven to give the solder some heat to melt, and lmao, the GPU revived! For 15 minutes anyways. "Maybe it was the Alphacool nexxxos heatsink and backplate applying pressure in the wrong way?" i thought. Took both GPUs apart, and studied them closely. Nothing. I put the blocks back on, but swapped them on purpose. Can't hurt, right? Wrong. 1st GPU is all acting weird somehow, while the 2nd GPU is doing the same thing.
But here is the hilarious part. I left the machine crashing and locking and doing whatever it was doing for a little while, while studying what was happening. After a rough 20 minutes it seemed stuff was getting better. It did not crash immediately, resolution was normal and i could even read GPU1's data through GPU-z. The GPU1 was reporting as "AMD Radeon R9 200 Series" in Device Manager, GPU2 now reported as well! But it was more specific: "AMD Radeon(TM) R9 290X". It did not give me much on GPU-z and whatever it gave me was either infinite or zero. But after 2 manual reboots without crashing, GPU-z managed to consistently read all the data! Temperatures, voltage, amps, everything! Yay, finally some good result! I got bold and booted up Unigine Valley, but got a swift kick in the face. Instant crash. No BSOD or crashing drivers though, so i got that going for me which is nice.
This is how far i got, but now i have no clue on how to proceed. Is there any chance of proper revival, beyond 2D clock speeds? I think that is the main issue at the moment. The GPU runs fine while it is clocked at low speeds, but as soon as it ramps up it crashes. Maybe voltage control is bad?