My RTX 3060 Ti randomly crashes while I'm using the computer. The monitor connected to the card suddenly loses signal, the GPU fans immediately ramp up to full speed, and Windows stops detecting the card. I have to restart the PC, and sometimes it takes two or three restarts before the GPU is detected and produces video output again.
I've reinstalled the NVIDIA drivers several times and even performed a clean Windows installation, but the issue continues. As a temporary workaround, I connected one monitor to the motherboard's integrated graphics and left the other connected to the RTX 3060 Ti. When the dedicated GPU crashes, only its display cuts out; I can still use Windows through the integrated graphics.
The same failure also occurs when I try loading a small local language model. As soon as the GPU begins handling the workload, the screen goes black, the fans reach 100%, and the card disappears from Windows until I restart.
Does this point to a failing graphics card, overheating or hotspot problems, a motherboard or PCIe fault, or a PSU and power-delivery issue? What tests would best identify the faulty component?
4 Answers
The combination of a black screen, fans suddenly locking at 100%, and the card disappearing until a reboot usually points to a graphics-card fault rather than a driver problem. Try reseating the card and its power connectors, then test it in another PCIe slot if your motherboard has one. If the same behavior follows the card, the GPU is the leading suspect.
Reseat the GPU and check every PCIe power connection carefully. If possible, use different modular PSU cables and try a different PCIe slot. A different known-good power supply or another computer would be the most conclusive test, but if the card fails in another system too, it is almost certainly defective.
Monitor both the GPU core temperature and the hotspot temperature while testing. A defective cooler, overheating memory, or failing VRM can cause this exact kind of crash. If temperatures look normal, try temporarily lowering the GPU voltage and power limit. If reducing power makes the crashes stop, that suggests the card is becoming unstable under load rather than Windows being the root cause.
Since drivers and Windows have already been reinstalled, software is less likely. Test with the integrated graphics alone, then test the RTX 3060 Ti in another compatible system or try another GPU in this computer. Comparing those results will separate a bad card from a motherboard slot or power-delivery problem.
This is especially worth testing because the crash happens as soon as a GPU workload starts. A small local model can still create a sudden, sustained load even if it does not use much system memory.