During its Architecture Day 2021 presentation, Intel revealed some details about an AI-powered super sampling solution calling it Xe Super Sampling (XeSS), which will have two modes, one as an effective solution for Intel’s hardware and the other that will run for Nvidia, AMD, and integrated graphics.
The way the presentation describes XeSS makes it more similar to Nvidia’s DLSS than AMD’s latest FSR. It is a profane solution, same as DLSS that will employ details from previous frames to upscale the game to higher intent while saving performance.
The presentation embraces the exposition of XeSS upscaling the real-time demo Intel had made from 1080p to 4k. Though this wasn’t a demo of a real game and didn’t even include any frame rate counters, the upscale image was more like an endemic 4k image. Intel confirmed that the frame time upscaling from 1080p will be slower than just running the game, but still faster than an endemic 4k image.
The hardware-accelerated supersampling mode, XMX used in the demo, will work exclusively on Intel’s recently launched Arc graphic card, using their Xe matrix extension cores. In an explanation of its other versions Intel said, it should run on any hardware with DP4a instructions, including Nvidia GPUs, AMD GPUs, and integrated graphics chips. Not getting into deep detail on working of this version of DP4a, Intel called it, “smart quality performance trade-off”.
Intel also announced that this month, the developers will be able to attain SDK for the XMX version of XeSS, with the launch of the DP4a version later this year. It also planned to make tools and SDK open-source later after Xe Super Sampling will be fully grown.
Xe Super Sampling will be available after the launch of Intel’s Alchemist Graphic Cards in early 2022.