Downloaded and had a quick look around. It was an interesting test of GPU power.
Just in the starting area:
4k Ultra, DLSS "auto", FG 4x ~160fps. Looked and felt fine.
4k Ultra, DLSS "auto", FG 2x ~90fps. Looked and felt fine.
4k Ultra, DLSS "auto", FG off ~50fps. Looked and felt ok.
Latency felt fine. Indicated latency was ~30ms render, 50-60ms PC with the higher end going to FG modes. I think it was using ultra performance setting in DLSS but it still looked good.
I did notice UI elements glitched a little when FG was on. Specifically the health box on lower corner, and also the crosshairs. This looked fine when FG off. I also found a disocclusion effect on screen edges when panning regardless of FG. For example, if I turn up, the sky that was uncovered was initially different but it caught up quickly. I think both of these were related to temporal reconstruction not handling literal edge cases well where it didn't have multi-frame data to work on, or that data was impacted by the UI.
Tried turning off RR. It looked pretty bad, like AMD's FSR4 demo with unstable lighting/shadows.
Tried lower quality presets, with and without DLSS. 16GB VRAM was a limit in several cases. At sufficiently low settings without DLSS, latency was way higher than even with FG.
There's a bit of chain link fence near the start. If you look through it without moving (either the character or viewport) then it looked fine. If you did move at all, it had a kinda distorted look like it couldn't handle the data of the two.