RTX 3080 black screen in video games, undervolted to "fix"

[thorn quail]

I am not sure why, but I've been forced to undervolt my GPU using MSI Afterburner, otherwise my PC would crash in almost any game

I had to keep trying lower and lower settings until it was stable

The "Crash" in question, is that my monitors go pitch black, and the fans go to 100%, and they just keep doing that, never ending, until I shut down my pc manually

RTX 3080 black screen in video games, undervolted to "fix"

I just updated to 576.80, as suggested in another thread

[thorn quail]

Updating the driver has not fixed the issue

[thorn quail]

Specs:

MB: Asus Z590-Plus
CPU: i9-11900k
RAM: Kingston Fury 16GB @ 3600 (DDR4)
GPU: Asus RTX 3080 10GB
PSU: Corsair RM750W 80+ Gold

[signal pendant]

Overclocking is a term that means to speed up the performance of a computer by increasing the component​’​s clock speed beyond the manufacturer's guaranteed and stable speeds, therefore increasing the speed at which that component operates at a cost of stability.

The quality of the silicon directly relates to how well it overclocks. The higher the overclock you can achieve with as low of a voltage, the more you've won the lottery. The silicon lottery usually refers to overclocking limits. A better overclocking CPU or GPU is considered "winning the silicon lottery." CPUs/GPUs have variances in how they overclock and no two chips have the same overclocking limits although they have the same die or the same model. You may even have a GPU/CPU that cannot be overclock at all without instabilities or crashes.

Keep in mind overclocking your GPU and CPU may cause issues such as degradation of the CPU/GPU, instabilities or crashes on your PC as you are running your GPU/CPU beyond the manufacturer guaranteed and stable speeds.

If you understand the risks and still wish to overclock your GPU, check the /gpuoverclocking command for basic instructions on software overclocking.

• You should not use an overclock on a GPU that is being used for work or scientific computing.
• Silicon lottery will mean that not every GPU can overclock the same. It is rare, but not impossible, for a GPU to become unstable with even just +15 MHz!
• GPU overclocking usually only has significant benefits on RTX 50 series GPUs that isn't the RTX 5090. For other GPUs, the benefit is likely in the single digits percentages.
• On GPUs utilizing GDDR6X or GDDR7, automatic ECC may mask unstable memory overclocks. This exhibits in reduced performance compared to a lower overclock. GDDR6X headroom is usually much lower than GDDR7, so it's likely you may encounter these limits with GDDR6X memory. Use the fastest speed that does not cause you to lose FPS/benchmark scores.
• On RTX GPUs, the ray-tracing and tensor multiplication cores can become unstable before the main CUDA shaders. Test with RTX features turned on to cover every case! The following lists features that utilize tensor cores:
• DLSS (including Super Resolution, Ray Reconstruction, Frame Generation)
• RTX Dynamic Vibrance/RTX HDR
• RTX Video Super Resolution/RTX Video HDR
• RTX Neural Shaders (e.g. RTX Mega Geometry, RTX Hair)
• DLDSR
• ChatRTX
• Project G-Assist
• NVIDIA Broadcast features
• NVIDIA Broadcast SDK filters in apps
• Local generative AI (e.g. Stable Diffusion/Flux, Framepack, LM Studio, local AI tools in apps)

Automatic overclocking

• Set an appropriate power limit, then use the NVIDIA App to perform automatic tuning.
• This overclock is very conservative and likely is much lower than what manual OCs can achieve, but has a very low risk of failure.
• Despite this, it is still possible for the tuning to result in an unstable OC. If this is the case, please restore settings and consider your GPU un-overclockable.

Manual overclocking

• Use MSI Afterburner to perform core and memory overclocks.
• Set an appropriate power limit. If desired, also set an appropriate temperature limit.
• Have a game or actively rendering 3D app (such as any of the non-FurMark MSI Kombustor scenes) you won't mind crashing running in a window or fullscreen borderless at the background to verify. Keep in mind that for best results, also test with games or apps using RTX features.
• For the GPU core, use +15 or +20 MHz steps while the game or app is running until it starts crashing or producing artifacts. Then subtract by two of the above steps, re-apply settings, and retest until stable.
• For GPUs with GDDR6 or GDDR6X memory, use +100 steps. Increase until it crashes, produces artifacts, or your performance drops, whichever comes first.
• For GPUs with GDDR7 memory, most testing can be skipped, and you can immediately try +2000. If this is stable, consider testing +3000. You cannot go any higher than +3000 as there is a driver level limit.

[compact beacon]

Does it crash when you run it at stock?

[thorn quail]

What does stock mean? @compact beacon

I am not overclocking

OBS: Please disregard the previous 2 large blocks of text, I can't remove them, this issue has nothing to do with overclocking

[thorn quail]

This is my HWinfo right before crashing during the Time Spy Benchmark

I have uninstalled my GPU driver using DDU

My driver version is 560.94, I am not sure if I can push it further back than that, as it is the driver I end up with after restart from DDU

[thorn quail]

I've gone off the deep end and have elected to underclock my GPU to -502 both on core and memory clock

This has caused the benchmark to no longer cause the crash

I will now try to update my driver to newest, and run benchmark again