I've been troubleshooting an aging PC for the past few weeks. My GeForce GTX 1660 Super would crash under heavy load while its fans suddenly ramped up to full speed. Undervolting seemed to help temporarily, so I suspected my old power supply and replaced it with an MSI MAG A750GL, but the crashes continued.
While testing the system with one RAM stick installed, I ran a browser-based GPU stress test. The computer immediately lost power, and afterward it would not turn on. I then removed the GPU's PCIe power connection and pressed the power button, which was a serious mistake: I heard cracking noises, saw a brief flash, and noticed a burnt-plastic smell. I unplugged the system and removed the GPU. The computer still starts without it, and I don't see obvious burn marks or melted plastic on the card or motherboard.
Should I reinstall the GPU to see whether it still works, or is it safer to replace it? Also, could a GPU be damaged even if there is no visible physical damage, and what might have caused it to fail at only around 60°C?
3 Answers
The GPU can fail without visible scorching or melted plastic. Components such as VRMs, capacitors, or traces can short internally, and the protection circuits may shut the PC down before leaving an obvious mark. The PCIe power cable should never be disconnected while the system is powered or while attempting to start it; doing that can create arcing and explain the flash and smell.
The original crashes at about 60°C probably weren’t caused by normal GPU overheating, so replacing the thermal paste would not necessarily solve the issue. A failing power-delivery circuit, memory fault, unstable card, bad cable, or another hardware problem could cause crashes at safe core temperatures. Since the PC works without the card, check the replacement PSU and cables for damage, then install a known-good GPU rather than testing this one again.
Don’t reinstall it. Cracking, a flash, and a burnt smell strongly suggest an electrical short or a failed component, even if you can’t see obvious damage. Testing it again could damage the motherboard, power supply, or other parts. A GTX 1660 Super usually isn’t worth risking the rest of the system over, so replace it or have it inspected by a qualified repair technician.
That makes sense. I’m not going to risk putting it back in just to test it.
I understand now that disconnecting the PCIe power while trying to start the system was unsafe. I’ll leave the card disconnected.