Overclock.net - An Overclocking Community - View Single Post - [Techpowerup] NVIDIA DLSS and its Surprising Resolution Limitations

View Single Post
post #15 of (permalink) Old 02-14-2019, 09:54 PM
Waiting for 7nm EUV
tpi2007's Avatar
Join Date: Nov 2010
Posts: 11,381
Rep: 894 (Unique: 503)
Quote: Originally Posted by looniam View Post
or how about the press was wrong? like i said they were back then??

yeah simple answers are best.

The press wasn't wrong, they didn't make it up on their own and Tom's Hardware specifically confirmed it with Nvidia.

While we hoped Nvidia's GeForce Experience (GFE) software wouldn't be a requisite of DLSS, we suspected it probably would be. Sure enough, the company confirmed that the features of NGX are tightly woven into GFE. If the software detects a Turing-based GPU, it downloads a package called NGX Core, which determines if games/apps are relevant to NGX. When there's a match, NGX Core retrieves any associated deep neural networks for later use.

You know where these three sites got their info from to begin with? From none other than Nvidia themselves in their Turing architecture whitepaper (page 33):



The features of NGX tightly couple to the NVIDIA driver and hardware. The NGX API provides access to several AI features for games and applications. The features are pre-trained by NVIDIA and ready for integration. The API has been designed to be thin and easy for applications to integrate multiple AI features. NGX services run on the GPU, allowing it to support multiple features and applications.

NVIDIA NGX features are managed by the NVIDIA GeForce Experience™ (GFE) application or the tech preview version of the NVIDIA Quadro Experience™ (QXP) application. After GFE or QXP is installed or updated, it looks for the presence of a Turing GPU. Once detected, the NGX Corepackage is downloaded and installed. GFE/QXP communicates with NGX Core to determine the game and application IDs present and their relevance to NGX. Different DNN models that work with various installed games and applications are then downloaded for subsequent use.

NGX DNN models can interface with CUDA 10, the DirectX and Vulkan drivers, as well as take advantage of NVIDIA TensorRT™, the high-performance deep learning inference optimizer that delivers low latency and high-throughput for deep learning inference applications. NGX models and services are accelerated by Turing’s enhanced Tensor Cores.

tpi2007 is offline