That's Not Right
You know when you have a model for how things work and then something happens that doesn't fit that?
One desktop ago my graphics card had a fault where the screen would go black and the fans would max. I could still hear music and control that music with hotkeys but the graphics card was well and gone; only a reboot would bring it back. As far as I could tell this would occasionally happen when a new 3D asset was streamed in, not as a function of load. A port of a mobile game triggered it on level change and that was when I realized enough was enough. Remedies available to me, like updating BIOS and reseating components hadn't prevented it. I RMA'd it and the replacement worked just fine. Still does, by all accounts.
A little bit ago my current card exhibited the same fault. I wasn't even using my desktop and the load was a paused YouTube video while I was shoveling snow. I got back in, heard the fans from across the house, and held the button. It happened again only hours later so I decided I would reseat the card before starting an RMA process.
So I flip the switch on the power supply, wait for the LEDs to die, and open the case. With how big graphics cards are I decide I'll just give it a good shimmy first and midway through the LEDs turn back on for a second.
The power is disconnected. Where did the electricity come from?
I know that there's a few voltages kicking around on the motherboard and that the LEDs are probably on the 3.3v one, given another system I had with a failed power supply that had working LEDs and IPMI interface but wouldn't turn on. I plugged that power supply into a tester and it flashed that the 12v was out of spec for a few seconds before the PSU released the magic smoke! If only all hardware tests could be so conclusive.
So I learned that the slot doesn't just take 12v but also there's a single pin pair dedicated to 3.3v for standby reasons. Also that the slot only has two pin pairs for the 12v available from the motherboard. I knew it could take up to 75 watts but I'm always shocked at how much power goes through some of these connectors. My CPU's core is at like 1.2v and its sucking down 200 watts. I work the P=IV on that and sure, it says 160 amps but I don't really wanna believe that.
But what happened with the LEDs? Was the standby power disconnected and some power trapped on the card? The D in LED stands for diode so it's not like the power ran backwards. Are the LEDs in series downstream of the GPU? Surely it's all wired in parallel, in a bus. Clearly the pins weren't making the best contact because that shimmy seems to have resolved the issues but I can't square what happened with any reasonable sequence of events.