The Intel Upgrade Tool is put to the test in Tomb Raider

With the latest generations of graphics cards, the phenomenon of upgrading algorithms has gained a boost. Nvidia has Deep Learning Based Super Sampling (DLSS) and AMD recently released FX Super Resolution (FSR) in version 2.1. When Intel unveiled the Arc graphics cards, it was announced that it would also enter the game with Xe Super Sampling (XESS), which was supposed to debut with the Dolmen game.

► Intel’s XESS should reduce defects when expanding

Unfortunately XESS Failed to Dolmen, and XESS, like both hardware and software from Intel on the graphics front, has been seriously delayed. Now Digital Foundry is diving deep into the technology, which was revealed in Shadow of the Tomb Raider. In a half-hour video, the technology is tested with the Arc A770 in several quality modes and compared with a similar Nvidia DLSS, the latter of which is powered by a Geforce RTX 3070.

Just like Nvidia, Intel is investing in machine learning with XESS. In the current feature, there is still only a variant of machine learning-based supersampling that is implemented in the form of XMX, which only works with Arc graphics cards. The upscaling tool has been tested in 1080p, 1440p and 4K resolution with very positive results. The most powerful performance mode at best delivers up to 88 percent higher frame rate in 4K. The difference isn’t quite as big at 1440p, where instead it represents an improvement of about 50 percent.

When the image quality is compared side-by-side with Nvidia’s DLSS, XESS scores well in terms of image quality, very close to the competition. However, it is not completely bug-free and does cause problems with artefacts. Another major drawback is the display time. When upgrading from 720p to 1440p, the display time increases by 2ms, and from 1080p to 4K, this number reaches 3.4ms.

Since the technology has not yet been launched sharply, it may improve before XESS reaches regular users. However, it is not clear exactly when Intel plans to release it. In a future release, machine learning-based super samples written with DP4a instructions should be added, for compatibility with hardware from manufacturers more than Intel.

Also recommended for those who prefer text over video Companion article to Digital Foundry.