DLSS / FSR

Competing spatial upscaling and frame generation technologies that allow games to render at a lower internal resolution while reconstructing a visually comparable higher-resolution output, recovering GPU performance lost to demanding settings like ray tracing or 4K rendering. DLSS (Deep Learning Super Sampling), developed by NVIDIA, uses a neural network trained on ultra-high-resolution reference frames to produce upscaled images with quality superior to simple bilinear or bicubic scaling, particularly in motion and fine detail. DLSS is exclusive to NVIDIA RTX GPUs. FSR (FidelityFX Super Resolution), developed by AMD and open-source, uses a spatial upscaling algorithm without machine learning, making it compatible with virtually any GPU, including NVIDIA cards, and requires no AI training data. FSR 4 (2025) introduced machine-learning upscaling competitive with DLSS, though limited to RDNA4 GPUs initially. Intel's XeSS offers a third option optimised for Arc GPUs. Frame Generation, available in DLSS 3 and FSR 3, synthesises additional frames between rendered frames to multiply perceived frame rates, useful for high-performance targets but not recommended for fast-paced competitive games due to latency implications. In practice, DLSS Quality and FSR Quality modes typically add minimal visual degradation while recovering 30-70% of GPU performance, making them near-mandatory settings for ray tracing at high resolutions.