IT House March 24 news, at the GDC 2022 Game Developer Conference, Intel detailed its resolution hypersampling technology XeSS.
Intel XeSS is a hypersampling technology that uses machine learning to reconstruct low-resolution frames into high-resolution frames, which can significantly improve the smoothness of the game at low configurations.
The core principle of XeSS is to leverage ML through XMX AI-accelerated hardware integrated on Xe-Cores for Arc Alchemist GPUs. XeSS will be fully compatible with DirectX 12 and use NN (Neural Network) based on the Intel Vector Shading Language running on Intel Arc SIMD architecture.
XeSS 4K rendering scenes with XMX requires less than half the cost compared to native resolution (4K). The technology is also DP4a compliant, which means it can run on GPUs without AI or ML acceleration engines such as XMX (compatible with some of NVIDIA and AMD's graphics), and offer similar performance and image quality.
Intel says XeSS can achieve higher scaling ratios compared to time scaling or spatial scaling without compromising quality. XeSS also comes with a built-in sharpening engine and will replace TAA anti-aliasing.
XeSS has 5 different presets including Super Performance, Performance, Balance, Quality, and Super Quality. Intel shares a new XeSS Rens demo that runs on a fixed-frequency Arc Alchemist GPU with ray tracing-enabled 1440p and 4K quality.
Tests have shown up to 153% performance improvement over native 4K quality in Super Performance mode, which provides up to 27% performance improvement over native at 4K resolution.
IT House learned that Intel also demonstrated the improvement of the size core of the 12th generation Core processor in the game, and developers can use AI acceleration, character animation, physics, collisions, audio processing and other background tasks to make the large core with stronger single-threaded performance run more demanding tasks.
In the Killer 3 demo, the 12th Generation Core hybrid architecture increased the average frame rate by 4% and load wait times by an average of 55%.
▼ 12900K Off Small Core
▼ 12900K Open small core