Frame Generation Latency: NVIDIA DLSS 3 vs AMD FSR 3 vs Reflex Integration: Edge Cases, Failure Modes, and Root-Cause Debugging

[Austin Howard]

I’ve been stress-testing frame gen latency across DLSS 3, FSR 3, and Reflex in competitive titles, and I’m hitting some genuinely annoying edge cases. For anyone running FSR 3 Frame Interpolation on a 30-series card via the Flux mod—how are you measuring actual pointer-to-photon latency? My ReShade + RTSS combo seems to skew readings by 8-12ms, and I can’t tell if that’s the overlay adding input delay or the interpolation pipeline itself. When you’re toggling between native 240Hz and 144Hz-with-frame-gen mid-session, does Reflex still track correctly, or does the clock switch desync the input queue?

On the failure-mode side, what happens to decoupled UI when the optical flow estimator gets confused by fast camera snaps or particle-heavy effects? I’ve seen the whole frame buffer hitch for 40+ ms on sudden mouse flicks in FPS games, which is catastrophic for tracking. Does anyone have a reliable way to isolate whether the stutter is coming from the flow field computation or from the reconstructor filling a dropped frame? And has anyone benchmarked whether disabling UI frame gen separately actually reduces perceived input lag, or is it mostly a theoretical win?

"Original Xbox & PlayStation 2 CRT gaming corner in the den."

[Sam Young]

The thing everyone gets wrong is treating frame generation as a free win rate—on a competitive title it usually makes you slower, not faster, because you’re injecting interpolated frames into a render queue that Reflex then has to drain. DLSS 3’s Optical Flow Unit is genuinely better at this than F

[Author]

Posts