How Modern Upscalers Work
Rather than rendering every single pixel natively, modern upscalers (NVIDIA DLSS, AMD FSR, Intel XeSS) render the game at a lower internal resolution (e.g., 1080p) and reconstruct the image to your native display resolution (e.g., 1440p or 4K) using temporal accumulation vectors and AI neural networks.
DLSS (Deep Learning Super Sampling)
NVIDIA DLSS runs on dedicated Tensor Cores built into GeForce RTX graphics cards. Because it leverages hardware-accelerated autoencoders, DLSS generally provides the cleanest temporal stability, least ghosting, and best reconstruction of fine details like power lines and mesh fences.
AMD FSR (FidelityFX Super Resolution)
FSR is open-source and runs across almost all modern GPUs (Nvidia, AMD, and Intel). While FSR Quality delivers great performance, it uses spatial-temporal shaders rather than dedicated tensor hardware, which can occasionally introduce shimmering on fine foliage.
Frame Generation: How It Differs from Upscaling
Frame Generation (DLSS 3 & FSR 3) does not reconstruct pixels; instead, it generates entirely new synthetic frames between two traditionally rendered frames using Optical Flow accelerators. It doubles perceived smoothness on high-refresh monitors, but requires a base frame rate of at least 50-60 FPS for responsive input latency.