Intel has published a new video talking about its recent XeSS 1.1 SDK update along with a demo of iGPU performance using the AI-based upscaler.
The new Intel XeSS 1.1 SDK brings two major updates to improve visual quality & temporal stability through a new XeSS Upscaling model for cross-vendor and integrated GPUs and an advanced XeSS Upscaling model for Arc GPUs that improves temporal stability (flicker/moire).
The new model also integrates new Kernels that include Faster XMX kernels (Arc Alchemist), Faster DP4a Kernels (TGL / ADL / RPL iGPU), and Faster Cross-Vendor kernels (SM 6.4 GPUs).
XeSS is an AI-based super-sampling technique which reconstructs a low-resolution non-antialiased frame into a high-resolution fully-antialiased frame. And it does that at a fraction of the cost of high-resolution native rendering.
During this talk, we introduce XeSS and cover some of the challenges that upscalers have to deal with. Then we follow that by going through updates that XeSS 1.1 brings and watch demos on Intel integrated and discrete GPUs. We will also talk about best practices and guidelines on how to integrate XeSS into a game title. Last topic is the upcoming XeSS Dataset Toolkit that makes it easy to collect and assemble a custom XeSS model training dataset for a particular game title.
via Game.Intel
In terms of performance, what this means is that gamers will notice a small performance boost as the XeSS Kernel runtimes are reduced across all visual fidelity and upscaling modes. For example, if you are targeting your game to run at 60 FPS, you have half the millisecond's worth of extra time off each frame for your rendering budget.
A game demo of Shadow of The Tomb Raider from Crystal Dynamics is showcased which gives us a
Read more on wccftech.com