While the new DLSS 4.5 Dynamic Multi-Frame Generation is an excellent step forward in scaling performance, I believe there is a gap in addressing 0.1% low micro-hiccups in otherwise high-performance scenarios.
The Problem: Dynamic Frame Gen currently functions as a reactive system—it scales the multiplier (2x, 4x, etc.) based on load. This creates a “startup” delay (latency-switch) when the GPU suddenly encounters an engine-level hitch or traversal stutter. Even with a high-end RTX 4090/5090, the transition from “Off” or “Low” to “High” multiplier can be felt as a perceptible shift in frame pacing and input response.
The Proposed Solution:
I propose a “Frame Smoothing Floor” (or “Idle Mode”) option within the Dynamic FG settings.
-
Mechanism: Rather than the system toggling to “Off” when the target frame rate is met, allow an “Always-On”1.1x or 1.2x idle state.
-
Benefit: By maintaining the optical flow vector calculation and VRAM buffers in a “warm” state at a constant, low-overhead multiplier, the system eliminates the “startup latency” when a sudden frame drop occurs.
-
Result: The system acts as an “Active Suspension” for frame times. When an unoptimized engine hitches and a native frame is delayed, the AI is already active and can seamlessly inject the synthetic frame to smooth the dip, keeping the frame-time graph a flat line rather than forcing the VRR/G-Sync module to react to a sudden stutter.
Why this matters: With the new Auto Shader Compilation (ASC) handling pre-baked stutters, this “Smoothing Floor” would address the remaining “Traversal/Engine” stutters. This would effectively turn high-end GPUs into precision stability tools, prioritizing frame pacing perfection over raw multiplier volume.