DLSS | 深度學習超高取樣

Deep Learning Super Sampling (DLSS) is a new image rendering technology developed by NVIDIA, which uses deep learning to produce an image that looks like a higher-resolution version of an original lower resolution image.

This technology is enabled by training a neural network on thousands of images from a specific video game both with and without anti-aliasing enabled. The network then learns how to improve the image to anti-aliasing quality without requiring any performance overhead. This trained neural network is then packaged into a video game driver, and is run during game play using the CUDA cores featured on NVIDIA’s GPUs.

DLSS is currently enabled on NVIDIA’s RTX20 Series GPU cards with a select number of video games (currently around 25 titles).

DLSS can enable much higher image resolution without requiring the same amount of performance overhead as anti-aliasing. And compared to commonly-used Temporal Anti-Aliasing (TAA) techniques, it provides much better temporal stability and image clarity. DLSS can provide a performance boost that increases framerates by up to 2X at 4K compared to NVIDIA’s previous generation of GeForce GPUs, allowing users to enjoy a super smooth gaming experience at 60 FPS or more.

GIGABYTE’s range of NVIDIA RTX20 Series graphics cards are all enabled with DLSS technology. In addition, GIGABYTE is also a leading manufacturer of GPU server hardware that are used for deep learning to develop revolutionary technologies such as DLSS. Some of GIGABYTE’s most powerful servers that are used for deep learning include the G291, G292, G481 and G482 Series.