Intel Crescent Island Xe3P GPU achieves 480GB memory using cost-optimized LPDDR5X, exceeding Nvidia and AMD competitors.

Intel has unveiled its Crescent Island GPU, which leverages the brand-new Xe3P architecture to accelerate Agentic AI applications. The Xe3P architecture represents an upgrade over the existing Xe3 architecture, which was previously teased during Panther Lake discussions. While Xe3 will feature in the next-generation Arc C-Series for client devices, Xe3P demonstrates greater scalability across a broader range of applications, from client iGPUs to data center AI GPUs. The Crescent Island GPU is designed with both power and cost optimization in mind, targeting air-cooled data center solutions, workstations, and AI inference workloads, with the architecture optimized for performance per watt.

Rather than relying on premium HBM memory like competitors NVIDIA and AMD, Intel has chosen to leverage LPDDR5X memory for Crescent Island. This approach provides Intel with a significant advantage in the cost/performance segment, particularly as sourcing HBM has become increasingly difficult due to heightened demand and elevated pricing. The GPU will feature up to 480 GB of LPDDR5X capacity, with the reference Intel PCIe card offering 160 GB of LP5X memory, while partners will have the flexibility to build ODM-branded cards with options extending up to 480 GB. This memory configuration enables a densely packed channel design that delivers significant bandwidth increases while maintaining a modest 350W TDP on the air-cooled PCIe version.

The Crescent Island GPU is exclusively designed for GPGPU workloads and eliminates traditional GPU blocks such as graphics and 3D support, freeing additional die area for AI compute. The architecture supports a broad range of data types from FP4 up to FP64, making it particularly suitable for "Tokens-as-a-service" providers and inference use cases. Intel is optimizing Crescent Island for both Perf/Watt and Perf/TCO metrics. The GPU will be backed by an open and robust software stack, with Intel currently evaluating its open and unified software platform through its existing Arc Pro B-series lineup, enabling future iterations to access these optimizations early on.

Intel is targeting customer sampling for the Crescent Island GPU during the second half of 2026, with additional details expected to emerge in the coming months.