Intel Introduces Next-Generation Graphics — Xe3

Intel has unveiled details of its third-generation graphics architecture, Xe3. It will be part of the Panther Lake platform, replacing Xe2 and forming the foundation for both integrated laptop GPUs and discrete graphics cards.

Xe3 will come in two configurations: 4 graphics cores (32 XMX, 4 RT units, 4 MB L2 cache) and 12 graphics cores (96 XMX modules, 12 RT units, 16 MB L2 cache).

Architectural Improvements

Compared to Xe2, the new Xe3 architecture offers:

• up to 25% more threads in the Xe Vector Engine;
• support for FP8 dequantization for AI workloads;
• 33% larger L1/SLM cache;
• twice faster anisotropic filtering and stencil testing;
• a new URB Manager and an improved ray tracing unit.

Performance has increased by 50%, and power efficiency by 40% compared to Arrow Lake-H.

Software Enhancements

Intel has also improved the software layer. The new architecture now includes an updated task scheduler, DirectX Cooperative Vectors support (developed with Microsoft), and updated multimedia capabilities such as AV1, VVC, XAVC-H/H-S/S, 10-bit AVC, and the eDP 1.5 interface.

The Next Step — Xe3P

The next evolution will be Xe3P, an even more powerful architecture that will appear in Nova Lake processors and upcoming discrete GPUs. Xe3P is expected to deliver up to 60% performance improvement.