Depth map and Speed Warp problems

Slow clip to 70% Optical Flow Speed Warp better, and using Depth Map node passing alpha to 2 other nodes -> Background and Subject
When rendered, the Depth Map is like 1 or 2 frames behind, causing 'ghosting'.
Doing the process as 2 renders -> Depth Map then import for SW render is a work-around.

h.265/30fps/1080p/Nvidia

Can you show us your node tree or better yet, upload a small project w/ media to a file sharing site and post a link here so someone can try? If you can't provide the media, even just the project file would help.

I wonder if a Render in Place on the speed changes before moving to Color would help here.

I wonder if a Render in Place on the speed changes before moving to Color would help here.

Interesting idea - sounds like a fix?
When doing both at 'the same time', it's like it pre-renders 1 then the other, then combines the two rasterizations for final render, and one lags behind the other for final rendering.
I finished/deleted the project using MM instead, so will have to try another later.

As a side note, Depth Map is so slow compared to MM it's hard to want to use it.
And it's use case seems very limited as subjects moving forward/backward equals poor results.
You can keyframe Depth Map attributes, but is a slow headache.

Resolve 19.0 full release.

Magic mask, no Speed Warp - same 'ghosting' issue presented after about 1 hr of editing.
Saved the project. Restarted Resolve. Issue gone.
Finished project.

It appears to be a Resolve/Nvidia Display Driver issue with the Tensor Cores getting 'out-of-sync' with Resolve, primarily after extended editing sessions.

As to new install, I went way over-board to be sure nothing in memory would affect install.
1. Restart. Wait 5 min for everything to fully load.
2. Cold boot. (shut-down, power off, wait 1 min)
3. Repeat step 1
4. install Resolve 19.0
*edit* 5. rebuild PowerGrade node tree from scratch to be sure old PowerGrade doesn't affect anything.

*Question* Is/how often is step 5 required? Need to rebuild PowerGrade between versions when? *

Since the feature is AI-based, I gather that an RTX-based GPU would be a massive boost over a GTX-based card?