For the past few days, my PC loses video output while gaming, although game or voice-chat audio continues briefly and the fans immediately ramp to 100%. The only recovery is holding the power button to shut the system down. The computer remains stable during Cinebench, OCCT, Blender rendering, video playback, drawing applications, and normal use.
I have already reinstalled and changed GPU drivers, reseated the graphics card, tried a Corsair RM850x and an RM1000e with their respective cables, tested different RAM, toggled XMP, run MemTest, updated the motherboard BIOS, cleaned the GPU, disconnected peripherals, changed monitors, and reinstalled Windows. None of that helped.
The problem disappears when I use an RTX 2060 Super instead of the ASUS TUF RTX 3090 OC. Limiting the 3090's power target to 70% also prevents the crashes, but that is only a workaround. Event Viewer shows no useful error besides the unexpected shutdown caused by holding the power button.
Core temperatures appear normal at roughly 70–80°C, but the GPU hotspot reached nearly 105°C during testing. Could this be a GPU hotspot, failing thermal pads, or a power-delivery issue? My system has an i7-10700K, Z490 motherboard, 32GB of DDR4-3200 memory, the RTX 3090, a Corsair RM1000e, two monitors, and a Wacom Cintiq. I am currently using the 12-pin-to-dual-8-pin cable supplied with the newer PSU; previously I used two separate 8-pin cables from the old PSU.
2 Answers
The symptoms point strongly toward the 3090 rather than the rest of the system. A black screen with audio continuing and fans going to full speed is common when the GPU stops responding. Since lowering the power limit or swapping to an RTX 2060 Super fixes it, check the 3090's hotspot and memory temperatures closely. A hotspot around 105°C is high enough to make the card suspicious, especially if the difference between the GPU core and hotspot is unusually large. Dried thermal pads, poor heatsink contact, or a failing card could all cause this.
Make sure the GPU power connectors are supplied correctly: each 8-pin socket should ideally have its own dedicated PCIe cable from the PSU, with no splitters, extensions, or adapters. Also never mix modular cables between PSU models, even if the connectors fit. Since you replaced the cables and are using the newer PSU's cable, that makes this less likely, but trying different PCIe sockets and dedicated cables is still worth doing.
I replaced all the cables when changing PSUs, so none of the RM850x cables are being used now. I tried the RM1000e's 12-pin-to-dual-8-pin cable as well as separate 8-pin cables, but the crashes still happen.
That was my suspicion too. The card is several years old, and I am considering inspecting or replacing the thermal pads, particularly because the 3090 seems stable only when its power draw is reduced.