My Windows 11 PC has a Ryzen 7 5800X3D, RTX 4070 Ti SUPER, 32 GB of RAM, and a 3440×1440 monitor. I never overclock the CPU or GPU. The system originally used a 7–8-year-old Thermaltake 700 W 80 Plus power supply, which I later replaced with a Thermaltake Toughpower GF A3 850 W.
The trouble started after I experimented with a mod that enabled DLSS Neural Rendering in Starfield. The game worked, but opening and closing menus frequently caused crashes, possibly because of VRAM-related problems. Later, while playing Rocket League, the display suddenly went black, the GPU fans ramped to 100%, and I had to force the PC off. The same thing happened again while I was only using Firefox.
At the time, I was using NVIDIA driver 616.56. I removed the driver with DDU from Safe Mode and installed version 610.74. Windows Reliability History showed LiveKernelEvent 141 errors, which made me concerned about a GPU or hardware fault.
I then replaced the old PSU, cleaned out a large amount of dust, and installed a GPU support bracket. Since then, Windows has booted normally and two consecutive 3DMark Steel Nomad runs completed without crashes, artifacts, driver failures, or fan issues. The scores were 5,544 and 5,574, close to the roughly 5,613 average for similar hardware.
Because I changed both the driver and PSU, I cannot tell which resolved the problem. The DLSS mod, driver 616.56, and the aging PSU are all possible causes. I also noticed before the replacement that the power button sometimes required two presses before the PC started, while it now powers on immediately.
Would two successful benchmarks be enough to consider the GPU and new PSU stable? Does this sound more like a faulty driver or DLSS modification, or could the old PSU have been failing and triggering the black screen, 100% fan behavior, and LiveKernelEvent 141?
1 Answer
The DLSS Neural Rendering mod is the most obvious initial trigger, since the crashes began after using it and it was already known to cause VRAM or rendering instability. That said, the old PSU should not be dismissed. A failing or aging power supply can cause sudden black screens and leave the GPU fans stuck at full speed, and the delayed response from the power button is another reason to be suspicious of it. Since the driver was also changed, you cannot identify the single cause with certainty, but removing the mod and replacing the questionable PSU were both sensible fixes. Two clean benchmarks with normal scores are encouraging, though continuing to use the system normally for a while is a better stability test than one benchmark alone.
That matches my thinking. The DLSS experiment may have exposed the problem, but the PSU replacement seems significant because the power button now works on the first press every time. I am going to avoid the mod and keep monitoring the system rather than deliberately reproduce the crash.