This is a fun one. I am slowly leaning towards my GPU might be dying but I would like some expert insight.
Every now and then my computer will start micro-stuttering. It stutters almost every 300-500ms exact. Mouse movements, key strokes, windows dragging, all that stuff freezes for a brief moment then jumps to where it should be. Audio never lags out. It occurs most frequently when I launch NZXT Cam or when playing PSO2. When it occurs in PSO2 it almost always begins right after a load screen. I have never had this occur in any other game or software. When it occurs, rebooting does not solve the issue. I just have to wait it out then it disappears on its own. Someone told me maybe something is triggering windows memory diagnostic and its running in the background, but I have disabled those tasks and can't find the service anywhere when this happens.
https://www.youtube.com/watch?v=6PVPVGcE-Rc This is a video I took with the game still running in the background. However, this is exactly the stuttering that continues even with the game closed and after reboot.
I have reinstalled windows. Tried Windows 1909 and 2004. Tried multiple nvidia drivers. The problem
never happens if I use my onboard video. I have troubleshot
a lot over the past few days and what I have discovered is below.
My specs are:
- Windows 10 Pro x64
- Aorus Master z390
- i7 9900k @5.0GHz
- 4x8GB Gskill Trident @ 3600MHz
- EVGA 980ti
- NZXT Cam and EVGA Precision X1 for monitoring/cooling software
When the stuttering occurs there are 3 threads that get pinned higher than dxgmms2.sys when inspecting System -> Threads with process explorer. ntoskrnl.exe!KeRaiseIrqlToDpcLevel. The issue is they never resolve once the stuttering begins.
I did an xperf with no game running and attached (xperf.png) a lovely little 20s blurb of spikes. Each spike is utilizing 100% of a core. However, it happens so fast task manager never picks it up. Task manager only says ~5% of CPU is being used. If you zoom in on a spike you get to see (zoomed-in-spike.png) that the spikes consist of 2 stacks, nvlddmkm.sys and ntoskrnl.exe. One is the nvidia driver and the other is windows kernel. The driver is firing off 35x more than the kernel based on the count, so I am leaning more towards driver issue. That is why I tried different drivers and whatnot.
Using latencymon (latencymon2.png) there are a lot of DPC counts for the nvidia driver and an equal number for the kernel and wdf01000.sys. No idea if this means anything. I know enough to be dangerous but not enough to make decisions for myself.
So something is causing the kernel and the driver to fight each other I guess. And since this build is less than a month old, minus the GPU, I am leaning to the GPU is dying. Sadly, I don't have a spare to test with, nor do I want to buy a new one since the 3xxx series is coming out later this year. Buying now I would be outside of the step up window :(. Any insight would be insanely helpful! I have posted this similar issue to the Nvidia boards with 0 traction. Just someone telling me my system is fine.