GameSR: Real-Time Super-Resolution for Interactive Gaming

High-resolution gaming demands significant computational resources, with challenges further amplified by bandwidth and latency constraints in cloud gaming. Existing upscalers, such as NVIDIA DLSS and AMD FSR, reduce rendering costs but require engine integration, making them unavailable for most titles and inapplicable to cloud gaming, where the client receives only compressed video and has no access to engine data. We present GameSR, a lightweight, engine-independent super-resolution model that operates directly on encoded game frames. The architecture of GameSR combines reparameterized convolutional blocks, PixelUnshuffle, and a lightweight ConvLSTM to deliver real-time upscaling with high perceptual quality. Extensive objective and subjective evaluations on popular games, such as Counter-Strike 2, Overwatch 2, FC24 and Team Fortress 2, show that GameSR reduces cloud gaming bandwidth usage by 35–56% while meeting target perceptual qualities, achieves real-time performance of up to 240 FPS for 2× scaling to 1080p (~4 ms/frame) and remains within real-time budgets for 4K upscaling (7–15 ms/frame), substantially outperforms existing super-resolution models in the literature, and achieves competitive no-reference perceptual quality compared to DLSS and FSR without accessing rendering engine data structures or modifying game source code, making GameSR a practical, engine-agnostic solution for upscaling both modern and legacy games in cloud gaming with no additional development effort.