My PC repeatedly black screens and restarts after a few minutes of demanding GPU use. It first happened while playing modded Cyberpunk 2077, then reproduced during Heavenly Benchmark 4.0, so it does not seem limited to one game. Lowering the GPU power limit to 80% makes the system more stable, but it can still crash under a heavy load. OCCT tests for the CPU and memory have not produced any errors. The only recurring Windows entry is Kernel-Power Event 41. The system has an RTX 3080 Ti Founders Edition and a Corsair 750 W PSU, both in a PC that is around four or five years old. Could this be a failing GPU, overheating or degraded thermal paste, or a PSU struggling with transient power spikes? What tests or next steps would help narrow it down?
3 Answers
Since lowering the power limit reduces the frequency of the crashes, the GPU power path is a strong suspect, but it does not prove the card itself is bad. Try a clean graphics-driver installation, return any GPU overclock or undervolt to stock, and test with a different PCIe power-cable arrangement. If the same card crashes another computer under a controlled benchmark, that points toward the GPU; if another GPU works normally in this PC, that narrows it down further.
GPU temperatures and hotspot temperatures are worth monitoring while running a benchmark. A card can appear to have a reasonable core temperature while the hotspot or memory becomes too hot. If temperatures climb rapidly, the fans behave oddly, or the hotspot-to-core difference is unusually large, aging thermal paste or pads could be involved. Repasting is not always obvious from visual inspection, and opening the card can be risky, so I would check logged temperatures first and treat it as a later step after testing the PSU and cables.
The 3080 Ti can produce substantial short power spikes, so the PSU is worth investigating even if its rated wattage looks sufficient. Kernel-Power Event 41 only confirms that Windows lost power unexpectedly; it does not identify the failed component. Check that the GPU power connectors are fully seated, avoid using a single daisy-chained cable if separate cables are available, and try another known-good, higher-wattage PSU if possible. Running the card in another system would also help separate a GPU problem from a system or power-delivery problem.
I was wondering whether repasting could be the cause. Is there a reliable way to confirm it before opening the graphics card?