For the past few days, my PC has been crashing only while playing games. Both monitors and the drawing display lose their signal, but game or voice-chat audio continues briefly. The fans then ramp up to 100%, and the only way to recover is to hold the power button until the system shuts down. Event Viewer does not show a useful error afterward—only an unexpected shutdown.
The system remains stable during Cinebench, OCCT, Blender renders, YouTube playback, drawing software, and other normal use. I have already reinstalled Windows, tried several GPU driver versions, reseated the graphics card, tested a different PSU and its cables, swapped the RAM, enabled and disabled XMP, run MemTest, updated the BIOS, cleaned the GPU, disconnected extra devices, and tested another monitor.
The problem disappears when I use an RTX 2060 Super instead of the ASUS TUF RTX 3090, or when I limit the 3090's power to 70%. The current system uses an Intel i7-10700K, ASUS TUF RTX 3090 OC, Z490 motherboard, 32 GB of DDR4-3200 memory, a Corsair RM1000e PSU, and three displays. Normal reported temperatures are around 70–80°C, but the GPU hotspot reached nearly 105°C during testing. Could this be a GPU power-delivery issue, overheating hotspot, failing thermal pads, or a problem with the graphics card itself?
2 Answers
The GPU hotspot temperature is a strong clue. A core temperature around 70–80°C can look normal while a hotspot near 105°C indicates a large temperature difference across the GPU, which can point to poor cooler contact, aging thermal pads, or dried thermal paste. The fact that reducing the power limit prevents the crash also fits a card that becomes unstable when it reaches higher load or hotspot temperatures. Monitor the core-to-hotspot difference and consider having the 3090's cooler and thermal interface serviced.
Check the 3090's power connections carefully. Each 8-pin GPU socket should ideally use its own dedicated PCIe cable from the PSU, without splitters, extensions, or adapters. Also make sure every cable belongs to the RM1000e; modular PSU cables are not universally interchangeable, even between models from the same manufacturer. Since the issue follows the 3090 and disappears with a lower-power card, unstable power delivery is still worth ruling out completely.
I replaced the cables when changing PSUs and am currently using the cables supplied with the RM1000e. I also tried both a 12-pin-to-dual-8-pin cable and two separate PCIe cables, but the crashes continued.
That was my suspicion too, since the card is several years old and the problem disappears at reduced power. I’m looking into the thermal pads and cooler contact next.