NVIDIA Reflex 2 and Frame Warp Explained: Latency, Camera Warping, In-Painting, and Current Support

NVIDIA Reflex Low Latency is an in-game SDK technique that coordinates CPU and GPU work so the CPU does not build an unnecessarily deep render queue ahead of the GPU. Reflex 2 adds a separate technique called Frame Warp. NVIDIA describes the combination as Reflex Low Latency plus Frame Warp rather than a replacement for the original Reflex pipeline optimization.

Frame Warp is not frame generation. It does not claim to simulate and render a complete new game frame between two conventionally rendered frames. Instead, it changes the camera perspective of the frame the GPU has just rendered, using a newer camera position calculated from more recent input.

While the GPU renders a frame, NVIDIA says the CPU calculates the camera position for the next frame from the latest mouse or controller input. Frame Warp then samples that newer camera position and warps the just-rendered image to it as late as possible, immediately before the frame is sent toward the display.

The latency idea is specific: camera motion can reflect newer input without waiting for the entire scene to be rendered again from that newer viewpoint. It does not remove every other delay in the system, and it does not make simulation state, networking, display scanout or pixel response occur at the time of the late camera sample.

Warping an already-rendered view can reveal image regions that were not visible from the original camera position. NVIDIA says its predictive rendering algorithm uses camera, color and depth data from prior frames to in-paint those newly exposed holes. The goal is to present the updated camera perspective without leaving empty regions where pixels moved.

That mechanism also defines an important boundary: Frame Warp is most directly about actions that move the game camera. NVIDIA shows the in-painting process in its THE FINALS material, but current vendor documentation does not justify a universal claim about how often artifacts occur or how severe they are across games, scenes and settings.

At the January 2025 announcement, NVIDIA said Reflex 2 could reduce PC latency by up to 75%. Its published THE FINALS example used 4K maximum settings and global illumination on an RTX 5070: NVIDIA reported 56 ms without Reflex, 27 ms with Reflex Low Latency, and 14 ms with Reflex 2 Frame Warp. Those numbers describe NVIDIA’s stated test configuration, not independent Core Tech Tips measurements.

NVIDIA also reported average PC latency below 3 ms in a CPU-bottlenecked VALORANT example running above 800 FPS on an RTX 5090. Treat both examples as demonstrations of the vendor’s implementation and measurement methodology. They do not establish a fixed percentage or millisecond saving for another GPU, game, frame rate, input device or display.

Reflex discussions commonly use PC or system latency for the local input-to-display pipeline, while render latency covers a narrower rendering portion. Frame time is the duration associated with producing frames; mouse polling interval describes report opportunities from the mouse; display response describes pixel transitions and display processing; network latency describes communication with a remote server. They are related to responsiveness but are not interchangeable numbers.

A faster mouse report interval cannot be subtracted directly from a Reflex measurement, and a 240 Hz display does not by itself define total PC latency. The Core Tech Tips mouse polling, frame-time, and monitor-latency explainers cover those clocks separately so Reflex 2 can remain scoped to render-pipeline pacing and late camera warping.

Availability needs a current-date distinction from the CES 2025 announcement. NVIDIA’s current Reflex 2 technology pages, crawled in September 2026, still label Reflex Frame Warp as “coming soon.” The original announcement said THE FINALS and VALORANT would receive Reflex 2 first on GeForce RTX 50 Series GPUs, with other GeForce RTX GPU support planned later, but that launch-era roadmap is not evidence that those releases happened.

Accordingly, this guide does not present Frame Warp as generally available, does not publish a current supported-game list that NVIDIA’s present technology page does not confirm, and does not promote the old planned GPU rollout into a current requirement. Ordinary Reflex Low Latency support is much broader, but that should not be confused with shipping Reflex 2 Frame Warp support.

Reflex Low Latency can reduce queue-related delay by pacing CPU and GPU work, while Frame Warp is designed to make displayed camera orientation reflect a later input sample. These mechanisms can address meaningful portions of the local responsiveness chain when a game integrates them correctly.

They cannot eliminate network delay, server simulation cadence, every CPU stall, display scanout, pixel response, peripheral latency or all image-quality tradeoffs. Frame Warp also does not turn an old simulation result into a newly simulated world state. The useful way to evaluate Reflex 2 is therefore as an engine-integrated latency technique with a specific camera-warp stage, not as a blanket replacement for frame rate, display, input or networking performance.