Intel confirms Xe3P architecture is not just for integrated graphics

Crescent Island GPU mockup

Only hours after Intel’s Xe3P architecture appeared in Linux kernel patches for upcoming Nova Lake-S desktop CPUs, the company has now confirmed that the same GPU design will extend into its data center portfolio.

At the OCP Global Summit, Intel announced a new inference-optimized data center GPU code-named Crescent Island . The product is based on the Xe3P architecture and is designed for AI inference workloads with a focus on efficiency and memory capacity.

Cresent Island Key features include:

• Xe3P microarchitecture with optimized performance-per-watt
• 160GB of LPDDR5X memory
• Support for a broad range of data types, ideal for “tokens-as-a-service” providers and inference

According to Intel, Crescent Island will feature 160 GB of LPDDR5X memory and support a wide range of data types suited for “tokens-as-a-service” providers and other inference applications. The GPU is described as power- and cost-optimized for air-cooled enterprise servers , offering improved performance-per-watt over previous Xe generations.

The company also confirmed that its unified AI software stack—currently being developed and tested on Arc Pro B-Series GPUs, will fully support Xe3P once the new hardware launches. Sampling for Crescent Island is planned for the second half of 2026.

With this announcement, Intel effectively confirms that Xe3P will not be limited to integrated graphics in Nova Lake systems but will also appear in dedicated GPUs for the data center. Whether this opens the door for new discrete consumer cards remains to be seen.

Source: Intel