Thank you everyone for the suggestions.
To narrow down the problem:
No, I'm not running interlaced video anywhere in the signal chain. We banished that years ago. Also, the tearing doesn't look anything like deinterlacing artifacts. I'm very attuned to the comb like artifacts on the edges of fast moving objects. This is the entire screen being shifted below a certain point. See attached screenshot.
Yes, I have the latest BIOS and driver updates installed.
Yes, I'm running the output out of the NVIDIA card and not the integrated graphics anymore.
No, I don't have a second M.2 NVME dedicated to replay (no free slots), although the problems started after the installation of the NVIDIA card, so I'm pretty sure that's not the source of the issue (and wouldn't that result in dropped frames and not tearing?).
Getting the screenshot might turn out to be the smoking gun. I ran through our last game and couldn't see any tearing. I actually got the screenshot from a replay export from an earlier game. Unless I totally misunderstand how replay works, this means the tearing is on the input, not the output.
I'm running a Magewell Eco Capture Quad SDI M.2 capture card.
It's been very warm lately and the air conditioning breaking in the press box didn't help. This caused thermal throttling of the Intel chip, which is why I went with the NVIDIA GPU in the first place - more headroom so I could maintain frame rate even when throttling. So I've been watching temperatures closely and it just so happens that the same screen that has the temps for the Magewell Eco Capture Quad SDI M.2 also lists PCIe settings. The Magewell
sometimes drops to Gen 2 x2 speeds. It's supposed to run at Gen 2 x4. According to the
Magewell KB, tearing may result if "bandwidth of Eco Capture card [is] insufficient".
I think the NVIDIA card is stealing lanes from the Magewell since it's on the same PCIe bus (the CPU). According NVIDIA's datasheet, the RTX A1000 should be running at Gen 4 x8 speeds although GPU-Z says it wants Gen 4 x16. It's in a slot that can only run Gen 4 x8.
So it appears that the NVIDIA card is the cause of the problem, but not the source if that makes sense.
I've swapped the Magewell and the NVME so the NVME is on the CPU bus and the Magewell is on the PCH bus. That's probably better anyway since the CPU bus is faster and the NVME needs Gen 4 x 4. Initial testing with HWiNFO shows that all devices are running with the lanes they should be.
Hopefully, this will resolve the issue.
Tearing.png
(871kb) downloaded 1 time(s).