Earlier, Intel released a preview of the Panther Lake processor architecture. The upcoming Panther Lake will use integrated graphics based on the Xe3 architecture, and their product roadmap also includes Xe3P, confirming its use in the Crescent Island data center GPUs.
However, this Xe3P GPU architecture will certainly not be used in just one product. According to videocardz, recent Linux graphics patches show that the term Xe3P has appeared under the Nova Lake processor entry, and entries about Xe3P in the kernel and Mesa mention Nova Lake series devices.
Changes in the Intel cm-compiler have mapped Nova Lake device IDs (nvl-u, nvl-h, nvl-s, nvl-hx, nvl-ul) to GMD versions 30.4.4 and 30.5.4, and added a release ID xe3-lpg to them. GMD stands for "Graphics Media Descriptor," which is essentially a device ID. Xe3P support appears in another patch in GMD 35.11.0, targeting a different hardware generation.
Intel added XMX instructions to the Xe3P DPAS (Matrix Multiply-Accumulate) path, covering micro-scaling formats such as FP8, FP4, and MXFP4 and MXFP8. These AI-oriented formats already exist in Crescent Island Xe3P GPUs, but support for them is not listed in Nova Lake information.
In fact, Nova Lake will use a mix of Xe3 and Xe3P GPUs. Because Nova Lake will be divided into five versions (U/H/S/HX/UL), the processor combinations will be diverse, and the GPU usage will follow the same logic. Most Nova Lake will use Xe3 GPUs, while Xe3P will be used in Nova Lake-H, with the top-of-the-line model having 12 Xe3P cores, serving as the successor to Panther Lake, which also features 12 Xe3 cores.
Of course, Nova Lake-H will also be paired with a discrete graphics card. This version of Nova Lake-H will use the Xe3 integrated graphics. Other versions, including Nova Lake-U/HX/S, will also use integrated graphics based on the Xe3 architecture, most likely directly using the quad-core Xe3 GPU from Panther Lake. However, compiler information indicates that some U-series models may also use Xe3P.
