Mr. Intel27 to 28 hosts a private event called “Intel Innovation”. The main event was the announcement of the 12th generation core, but in the main video, there was an explanation of the actions associated with the next disc GPU “arc” (code name: Alchemist).

The Arc is built on TSMC’s 6nm process, and the flagship model has 32 built-in Xe cores. It has all the latest features needed to run modern games, supports DirectX 12 Ultimate, and supports hardware ray tracing, mesh shader, and variable shading.

It also has a free AI acceleration function, which the company calls “XMX AI acceleration”. It seems to be similar to Nvidia’s tensor core, but as a technology to use it, introduced the high resolution technology “XeSS”.

Functions of XMX AI Acceleration

XeSS technology itself is similar to NVIDIA’s DLSS and AMD’s FidelityFX super resolution targets, and renders the game at a lower resolution to improve performance by reducing GPU load, while complementing the pixels for higher resolution viewing. However, this is to assume multiple frames or higher resolution images in the past, and the approach seems to be closer to NVIDIA’s DLSS.

This XeSS should be supported in the game as well, but the keynote address proved that “The Riftbreaker” works in Alchemist (used beta demo revision 22335). Drawing XeSS in 4K resolution can confirm that many 3D objects work smoothly, which is better than 1080p.

XeSS’s XMX AI acceleration can easily draw better 4K resolution than 1080p.

If you double or quadruple, the difference is obvious

Another feature of the Arc is the “Deep Link” technology already available in the entry level Xe MAX. Deep Link uses GPU encoder with built-in CPU and discrete GPU encoder at the same time to reduce transcoding time, but it is available in Arc.

The keynote address emphasized that using video conversion software “HandBrake” to coordinate Xe graphics and perform format conversions with a built-in CPU and arc could significantly reduce time.

Intel also supports deep linking

Demonstration of video conversion using handbrake

The conversion work was completed earlier when the GPU with the built-in CPU and the discrete GPU were collaborated using DeepLink.