On October 10th, Intel previewed the Core Ultra 300 "Panther Lake" processor and also shared details about the chip's debut Xe3 graphics architecture.
Intel Introduces Xe3 Graphics Architecture: Higher Scale and Comprehensively Optimized Performance
The Xe3 graphics architecture is an expansion and upgrade of Xe2. Its scale is reflected in the fact that a single render slice can accommodate up to six Xe cores and six ray tracing units, a 50% increase compared to the four and four slices in Xe2, respectively.
"Panther Lake" includes two versions of the integrated graphics: a 4Xe GPU and a 12Xe GPU. The former features four Xe cores split across two render slices and a 4MB L2 cache; the latter boasts two full-size render slices and an increased L2 capacity of 16MB to reduce external bandwidth requirements.
In terms of performance, the Xe3 architecture introduces third-generation Xe cores, featuring eight 512-bit vector engines and eight 2048-bit XMX matrix engines. L1 cache/SLM capacity has been increased by 33%. The vector engine threads have been increased by up to 25%, with support for variable register allocation. The XMX engines can deliver 120 TOPS of computing power, and ray tracing performance has also been improved.
Overall, the Xe3 GPU in Panther Lake boasts over 50% higher graphics performance than Lunar Lake, and over 40% better energy efficiency than Arrow Lake H.
