Fixed: Diagnosing a failing power supply
My son brought over his PC for some troubleshooting. He plays online games and is having issues with the machine black-screen rebooting mid-game in Apex Legends and Baldur's Gate. Additionally he's seeing screen "tearing" when watching local .mp4 video in Media Player.
We talked about reverting his video drivers and he'd done that, but it didn't fix it.
On the bench I was able to repro the problem in Baldur's Gate. Stress testing the Memory and CPU I wasn't able to trigger it, but the system crashed within thirty seconds running FurMark. The GPU and CPU temperatures looked reasonable, so something else is wrong.
A solid repro in Furmark on a clean OS image and drivers means this is almost certainly a hardware problem. I set up HWiNFO (Download HWiNFO Installer & Portable for Windows) to monitor the CPU/GPU temperatures, the 3.3/5/12V bus voltage sensors on the motherboard, and the 12V VRM input on the GPU. With that in place and Furmark running the 12V rail drooped down to ~10.9 volts. That's a problem. The spec is +/- 5% = 11.4V. I tried disconnecting the accessory loads (fancy gamer lights and shiny woo-woo fans), but that didn't make a difference. He needs a new power supply.
... but can we find some way to keep the system running until then?
Sure, with the caveat that the Power Supply could fail any minute and cook everything.
Back in the dirty OS, I setup MSI Afterburner to auto-start and limit the power consumption of the GPU. That power limit has a scale of +/- 40, but it's not clear on what the units on that are. At "-25" the 12V bus stabilized at 11.5 volts with FurMark cooking away, so I cranked it down to "-35". Viola!
That was fun and I thought I'd share.