Frame Generation Latency: NVIDIA DLSS 3 vs AMD FSR 3 vs Reflex Integration: Advanced Implementation and Real-World Architecture

[Isaac Smith]

I’ve been running controlled latency tests across DLSS 3, FSR 3, and native rendering in competitive titles like Valorant and CS2, and I’m hitting some confusing numbers. With FSR 3’s frame generation, the input-to-photon delay seems to creep up significantly compared to DLSS 3’s implementation—am I measuring this correctly, or is my test methodology flawed? Specifically, how much of that latency penalty is the optical flow interpolation buffer itself versus the injected frames not receiving fresh input samples? And does anyone have hard data on how much Reflex actually recovers when FSR 3 FG is active, since I’ve read that FSR 3’s approach handles the latency compensation differently under the hood?

Mai mult importantly, I’m trying to understand the UI decoupling mechanism—when you navigate menus with frame generation on, does the interpolated output lag behind the actual UI input, or does the system detect non-game content and bypass interpolation? In fast-tick-rate shooters, even menu input stutter matters for my reaction timing. Has anyone dug into the engine-level hooks, like how Unreal’s frame delivery pipeline or the DirectX 11 vs 12 paths affect this? I’m particularly interested in whether the latency cost is consistent across both APIs or if one stack handles the interpolation queue more efficiently.

🎮 Battlestation & Rig: Ryzen 7 9800X3D | RX 7900 XTX Taichi | 32GB DDR5 6000 CL28 | Corsair 5000D | ASUS ROG Swift Pro PG248QP 24.1″ 540Hz Esport-TN | Logitech G Cloud + Razer DeathAdder V3 Pro + Endgame Gear EM-C Pad | Sennheiser HD 560S Open-Back + Creative Sound BlasterX G6 DAC

📍 Munich (Germany)

"Pixel response over resolution every single day."

[Tyler Roberts]

Here’s the thing most people get wrong about this comparison: frame generation doesn’t just “add frames,” it fundamentally changes the latency pipeline. DLSS 3 leans on the dedicated Optical Flow Unit (OFU) on Ada Lovelace, which computes motion vectors between rendered frames to synthesize intermediates. The catch is those synthetic frames introduce

"────────────────────────────────────────"

"Consistent frame times matter infinitely more than peak FPS."

[Jayden Bell]

Here’s where the real architectural divergence matters: DLSS 3’s frame generation runs on a dedicated Optical Flow Unit (OFU) silicon block, which computes motion vectors and depth more accurately than FSR 3’s purely algorithmic Fluid Motion Frames approach. That hardware acceleration is why DLSS 3 interpolated frames generally get the occlusion and edge artifacts right

"Corsair 5000D Airflow | 10x Noctua NF-A12x25 PWM Fans | Sub-60°C Thermals"

[Chase Davis]

Great topic, and honestly one where a lot of the marketing gloss needs stripping away. The core thing people miss is that frame generation itself adds latency to the pipeline — you’re inserting synthetic frames between real renders, and if left unmanaged that balloons your input lag. What saves it is the Reflex/Ant

"Valorant Immortal 2 | 800 DPI @ 0.31 In-Game Sens • NA East"

[Author]

Posts