StarCitizen:DLSS
More actions
Deep Learning Super Sampling, or DLSS, is an artificial intelligence rendering technology that allows for better image quality at higher fps.<ref name=":0">NVIDIA DLSS 2.0 | A Big Leap In AI Rendering, Nvidia GeForce, YouTube, 23 March 2020</ref>
Star Citizen introduced support for DLSS 2 Super Resolution in Alpha 3.23, accelerating frame rates in areas where performance isn’t limited by the CPU.<ref name=":1" /><ref>Template:Cite RSI</ref> To alleviate those limitations, Cloud Imperium Games has deployed its new StarCitizen:Vulkan Renderer, which in a future update will add CPU multi-threading, enabling player GPU to be better utilized, allowing DLSS to further increase frame rates in many more of the massive game’s locales.<ref name=":1">Star Citizen DLSS 2 Update Now Available, nvidia.com, May 21, 2024</ref>
Later in Alpha 4.5.0, support for DLSS 4's transformer model was integrated to coincide with improvements made to the Vulkan Renderer.<ref>Template:Cite RSI</ref>
Technology
Deep Learning
Nvidia RTX GPUs use Tensor cores which are specialized AI processors that allow to execute fast predictive AI models on every frame.<ref name=":0" />
Super Sampling
Super sampling is the act of rendering the image in a higher resolution than that of the monitor, and then transforming the higher resolution to the output resolution. This is a form of anti-aliasing to improve the image quality compared to directly rendering it in the native output resolution
DLSS 1.0
After an Nvidia researcher used a convolutional network to fix damaged photographs in 2016, Nvidia saw an opportunity to make games run faster by first reducing the rendering solution and then using AI the fill the missing information thus generating high quality images at higher performance.<ref name=":0" />
DLSS 2.0
DLSS 2.0 noticeably improves image quality with less artifacts and aliasing, and has an increased upscaling ratio, with 1080p able to render at 4k. The neural network works twice as fast.<ref name=":0" />
External Links
References
<references /> Template:Short description