My Windows-based video pipeline is receiving a real-time UDP stream that must be upscaled to 1080p at 60 fps with the lowest possible latency. Right now playback stutters and the GPU isn’t fully utilized. The code relies on FFmpeg and CUDA kernels; no GStreamer is in place, but I’m open to lightweight tweaks if they speed things up. Here’s the focused scope I’d like tackled within the next 2–3 days: • Profile the existing FFmpeg + CUDA chain and pinpoint the single biggest performance bottleneck. • Implement quick-win parameter or kernel adjustments that restore smooth, real-time playback. • Supply the patched code, build/run steps, and a concise optimization summary so I can reproduce the results. Now it takes about 2s to pass the RealESRGAN model. I need to reduce this time for 60fps Remote work is fine; I’ll provide a sample UDP source and current launch script as soon as we start.