How it works

Two GPUs, one idle

Most gaming laptops (and many desktops) ship with two GPUs: a dedicated GPU that renders the game, and an integrated GPU built into the CPU that usually sits idle while you game. GlassBreakr uses that idle integrated GPU.

The pipeline

  1. Capture the game's frames.
  2. Optical flow — an AI model (RIFE-Nano, our compact model derived from the MIT-licensed RIFE project — Huang et al., ECCV 2022) estimates the motion between two consecutive frames, running on the integrated GPU via DirectML.
  3. Warp + blend — intermediate frames are generated from that motion.
  4. Display the real and generated frames in sequence.

Because steps 2–3 run on the integrated GPU, the measured AI load on the dedicated GPU is 0% in heterogeneous mode — the dedicated GPU is left free to render the game.

What you get

The output frame rate is about 2x the game's. Examples measured live (Subway Surfers): ~30 fps → 63 fps; and the FlowUp pipeline reaches 45-48 fps at 720p native with a warp step of just 1.7-2.2ms.

What "output FPS" means (and what it doesn't)

Output FPS counts real frames plus AI-interpolated frames. It improves perceived smoothness. It does not speed up the game's simulation, input handling, or physics, and it does not render at a higher native resolution (one mode's 540p output is AI-upscaled, not native). Interpolation adds a small, measured latency — see Benchmarks.

A note on the engine beyond gaming

The interpolation engine itself (RIFE optical flow + warp) is portable and also runs offline on Windows, Linux and macOS for video files. The real-time game overlay, however, is Windows-only.