Intel Panther Lake with up to 12 Xe3 cores

Intel puts some claims on performance versus Arrow Lake-H.

Today, Intel is revealing the first details on Panther Lake and its associated graphics architecture, Xe3. The GPU tile for upcoming mobile CPUs will be available in configurations with either 4 or 12 Xe3 Cores. Compared to previous iterations, Intel says the new Xe Vector Engine delivers 25% more threads, introduces variable register allocation, and adds FP8 dequantization support.

For fixed functions in the Render Slice, Intel confirms a new URG manager, 2× anisotropic filter rate, and 2× stencil test rate. The 3rd Gen Xe Core also features 33% more L1 cache per SLM than the Xe2 core. Additionally, Intel notes that the Xe3 core includes an enhanced ray tracing unit supporting dynamic ray management for synchronized ray tracing.

Source: Intel

In terms of performance, the Xe3 architecture is said to be over 50% faster than Lunar Lake’s Xe2 GPU at the same power and offers over 40% better performance per watt compared to Arrow Lake-H . However, it is important to note that Panther Lake uses a different design from Lunar Lake, which is based on the Xe2P core with an XMX engine.

Source: Intel

Intel shared multiple micro-benchmarks showing 1.0× to 7.4× performance improvements, with the largest gains seen in high-register-pressure shaders (3.1×) and depth writes (7.4×).

Panther Lake will feature two GPU variants:

• 12 Xe3 Cores : 96 XMX engines, 16 MB L2 cache, and 12 ray tracing units.
• 4 Xe3 Cores : 32 XMX engines, 4 MB L2 cache, and 4 ray tracing units.

Source: Intel

The 12-core variant is designed for systems without discrete graphics, while the 4-core version targets ultra-low-power laptops or high-end notebooks paired with a discrete GPU. Intel also confirms that the 12 Xe3 GPU supports 12 PCIe lanes, while the 4 Xe3 version offers up to 20 PCIe lanes.